diff options
Diffstat (limited to 'src/video_core')
49 files changed, 461 insertions, 251 deletions
diff --git a/src/video_core/buffer_cache/buffer_base.h b/src/video_core/buffer_cache/buffer_base.h index 9cbd95c4b..0bb3bf8ae 100644 --- a/src/video_core/buffer_cache/buffer_base.h +++ b/src/video_core/buffer_cache/buffer_base.h @@ -18,6 +18,7 @@ namespace VideoCommon { enum class BufferFlagBits { Picked = 1 << 0, CachedWrites = 1 << 1, + PreemtiveDownload = 1 << 2, }; DECLARE_ENUM_FLAG_OPERATORS(BufferFlagBits) @@ -54,6 +55,10 @@ public: flags |= BufferFlagBits::Picked; } + void MarkPreemtiveDownload() noexcept { + flags |= BufferFlagBits::PreemtiveDownload; + } + /// Unmark buffer as picked void Unpick() noexcept { flags &= ~BufferFlagBits::Picked; @@ -84,6 +89,10 @@ public: return True(flags & BufferFlagBits::CachedWrites); } + bool IsPreemtiveDownload() const noexcept { + return True(flags & BufferFlagBits::PreemtiveDownload); + } + /// Returns the base CPU address of the buffer [[nodiscard]] VAddr CpuAddr() const noexcept { return cpu_addr; diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index e534e1e9c..98756e4da 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -23,8 +23,6 @@ BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_, common_ranges.clear(); inline_buffer_id = NULL_BUFFER_ID; - active_async_buffers = !Settings::IsGPULevelHigh(); - if (!runtime.CanReportMemoryUsage()) { minimum_memory = DEFAULT_EXPECTED_MEMORY; critical_memory = DEFAULT_CRITICAL_MEMORY; @@ -75,8 +73,6 @@ void BufferCache<P>::TickFrame() { uniform_cache_hits[0] = 0; uniform_cache_shots[0] = 0; - active_async_buffers = !Settings::IsGPULevelHigh(); - const bool skip_preferred = hits * 256 < shots * 251; uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0; @@ -100,51 +96,50 @@ void BufferCache<P>::TickFrame() { template <class P> void BufferCache<P>::WriteMemory(VAddr cpu_addr, u64 size) { - memory_tracker.MarkRegionAsCpuModified(cpu_addr, size); if (memory_tracker.IsRegionGpuModified(cpu_addr, size)) { const IntervalType subtract_interval{cpu_addr, cpu_addr + size}; ClearDownload(subtract_interval); common_ranges.subtract(subtract_interval); } + memory_tracker.MarkRegionAsCpuModified(cpu_addr, size); } template <class P> void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) { memory_tracker.CachedCpuWrite(cpu_addr, size); - const IntervalType add_interval{Common::AlignDown(cpu_addr, YUZU_PAGESIZE), - Common::AlignUp(cpu_addr + size, YUZU_PAGESIZE)}; - cached_ranges.add(add_interval); +} + +template <class P> +std::optional<VideoCore::RasterizerDownloadArea> BufferCache<P>::GetFlushArea(VAddr cpu_addr, + u64 size) { + std::optional<VideoCore::RasterizerDownloadArea> area{}; + area.emplace(); + VAddr cpu_addr_start_aligned = Common::AlignDown(cpu_addr, Core::Memory::YUZU_PAGESIZE); + VAddr cpu_addr_end_aligned = Common::AlignUp(cpu_addr + size, Core::Memory::YUZU_PAGESIZE); + area->start_address = cpu_addr_start_aligned; + area->end_address = cpu_addr_end_aligned; + if (memory_tracker.IsRegionPreflushable(cpu_addr, size)) { + area->preemtive = true; + return area; + }; + area->preemtive = + !IsRegionGpuModified(cpu_addr_start_aligned, cpu_addr_end_aligned - cpu_addr_start_aligned); + memory_tracker.MarkRegionAsPreflushable(cpu_addr_start_aligned, + cpu_addr_end_aligned - cpu_addr_start_aligned); + return area; } template <class P> void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) { - WaitOnAsyncFlushes(cpu_addr, size); ForEachBufferInRange(cpu_addr, size, [&](BufferId, Buffer& buffer) { DownloadBufferMemory(buffer, cpu_addr, size); }); } template <class P> -void BufferCache<P>::WaitOnAsyncFlushes(VAddr cpu_addr, u64 size) { - bool must_wait = false; - ForEachInOverlapCounter(async_downloads, cpu_addr, size, - [&](VAddr, VAddr, int) { must_wait = true; }); - bool must_release = false; - ForEachInRangeSet(pending_ranges, cpu_addr, size, [&](VAddr, VAddr) { must_release = true; }); - if (must_release) { - std::function<void()> tmp([]() {}); - rasterizer.SignalFence(std::move(tmp)); - } - if (must_wait || must_release) { - rasterizer.ReleaseFences(); - } -} - -template <class P> void BufferCache<P>::ClearDownload(IntervalType subtract_interval) { RemoveEachInOverlapCounter(async_downloads, subtract_interval, -1024); uncommitted_ranges.subtract(subtract_interval); - pending_ranges.subtract(subtract_interval); for (auto& interval_set : committed_ranges) { interval_set.subtract(subtract_interval); } @@ -164,7 +159,6 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am } const IntervalType subtract_interval{*cpu_dest_address, *cpu_dest_address + amount}; - WaitOnAsyncFlushes(*cpu_src_address, static_cast<u32>(amount)); ClearDownload(subtract_interval); BufferId buffer_a; @@ -192,7 +186,6 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am const IntervalType add_interval{new_base_address, new_base_address + size}; tmp_intervals.push_back(add_interval); uncommitted_ranges.add(add_interval); - pending_ranges.add(add_interval); }; ForEachInRangeSet(common_ranges, *cpu_src_address, amount, mirror); // This subtraction in this order is important for overlapping copies. @@ -205,7 +198,7 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am if (has_new_downloads) { memory_tracker.MarkRegionAsGpuModified(*cpu_dest_address, amount); } - std::vector<u8> tmp_buffer(amount); + tmp_buffer.resize(amount); cpu_memory.ReadBlockUnsafe(*cpu_src_address, tmp_buffer.data(), amount); cpu_memory.WriteBlockUnsafe(*cpu_dest_address, tmp_buffer.data(), amount); return true; @@ -441,9 +434,7 @@ void BufferCache<P>::BindComputeTextureBuffer(size_t tbo_index, GPUVAddr gpu_add template <class P> void BufferCache<P>::FlushCachedWrites() { - cached_write_buffer_ids.clear(); memory_tracker.FlushCachedWrites(); - cached_ranges.clear(); } template <class P> @@ -474,15 +465,13 @@ void BufferCache<P>::CommitAsyncFlushesHigh() { if (committed_ranges.empty()) { if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) { - if (active_async_buffers) { - async_buffers.emplace_back(std::optional<Async_Buffer>{}); - } + + async_buffers.emplace_back(std::optional<Async_Buffer>{}); } return; } MICROPROFILE_SCOPE(GPU_DownloadMemory); - pending_ranges.clear(); auto it = committed_ranges.begin(); while (it != committed_ranges.end()) { auto& current_intervals = *it; @@ -537,64 +526,65 @@ void BufferCache<P>::CommitAsyncFlushesHigh() { committed_ranges.clear(); if (downloads.empty()) { if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) { - if (active_async_buffers) { - async_buffers.emplace_back(std::optional<Async_Buffer>{}); - } + + async_buffers.emplace_back(std::optional<Async_Buffer>{}); } return; } - if (active_async_buffers) { - if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) { - auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes, true); - boost::container::small_vector<BufferCopy, 4> normalized_copies; - IntervalSet new_async_range{}; - runtime.PreCopyBarrier(); - for (auto& [copy, buffer_id] : downloads) { - copy.dst_offset += download_staging.offset; - const std::array copies{copy}; - BufferCopy second_copy{copy}; - Buffer& buffer = slot_buffers[buffer_id]; - second_copy.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset; - VAddr orig_cpu_addr = static_cast<VAddr>(second_copy.src_offset); - const IntervalType base_interval{orig_cpu_addr, orig_cpu_addr + copy.size}; - async_downloads += std::make_pair(base_interval, 1); - runtime.CopyBuffer(download_staging.buffer, buffer, copies, false); - normalized_copies.push_back(second_copy); - } - runtime.PostCopyBarrier(); - pending_downloads.emplace_back(std::move(normalized_copies)); - async_buffers.emplace_back(download_staging); - } else { - committed_ranges.clear(); - uncommitted_ranges.clear(); + if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) { + auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes, true); + boost::container::small_vector<BufferCopy, 4> normalized_copies; + IntervalSet new_async_range{}; + runtime.PreCopyBarrier(); + for (auto& [copy, buffer_id] : downloads) { + copy.dst_offset += download_staging.offset; + const std::array copies{copy}; + BufferCopy second_copy{copy}; + Buffer& buffer = slot_buffers[buffer_id]; + second_copy.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset; + VAddr orig_cpu_addr = static_cast<VAddr>(second_copy.src_offset); + const IntervalType base_interval{orig_cpu_addr, orig_cpu_addr + copy.size}; + async_downloads += std::make_pair(base_interval, 1); + runtime.CopyBuffer(download_staging.buffer, buffer, copies, false); + normalized_copies.push_back(second_copy); } + runtime.PostCopyBarrier(); + pending_downloads.emplace_back(std::move(normalized_copies)); + async_buffers.emplace_back(download_staging); } else { - if constexpr (USE_MEMORY_MAPS) { - auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes); - runtime.PreCopyBarrier(); - 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, false); - } - runtime.PostCopyBarrier(); - 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); - } + if (!Settings::IsGPULevelHigh()) { + committed_ranges.clear(); + uncommitted_ranges.clear(); } 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); + if constexpr (USE_MEMORY_MAPS) { + auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes); + runtime.PreCopyBarrier(); + 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, + false); + } + runtime.PostCopyBarrier(); + 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); + } } } } @@ -1213,16 +1203,14 @@ void BufferCache<P>::UpdateComputeTextureBuffers() { template <class P> void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size) { - memory_tracker.MarkRegionAsGpuModified(cpu_addr, size); - if (memory_tracker.IsRegionCpuModified(cpu_addr, size)) { SynchronizeBuffer(slot_buffers[buffer_id], cpu_addr, size); } + memory_tracker.MarkRegionAsGpuModified(cpu_addr, size); const IntervalType base_interval{cpu_addr, cpu_addr + size}; common_ranges.add(base_interval); uncommitted_ranges.add(base_interval); - pending_ranges.add(base_interval); } template <class P> @@ -1629,7 +1617,6 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) { 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 if (!do_not_mark) { @@ -1668,14 +1655,15 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s const bool is_nvn_cbuf = cbuf_index == 0; // The NVN driver buffer (index 0) is known to pack the SSBO address followed by its size. if (is_nvn_cbuf) { - return gpu_memory->Read<u32>(ssbo_addr + 8); + const u32 ssbo_size = gpu_memory->Read<u32>(ssbo_addr + 8); + if (ssbo_size != 0) { + return ssbo_size; + } } // Other titles (notably Doom Eternal) may use STG/LDG on buffer addresses in custom defined // cbufs, which do not store the sizes adjacent to the addresses, so use the fully // mapped buffer size for now. const u32 memory_layout_size = static_cast<u32>(gpu_memory->GetMemoryLayoutSize(gpu_addr)); - LOG_INFO(HW_GPU, "Binding storage buffer for cbuf index {}, MemoryLayoutSize 0x{:X}", - cbuf_index, memory_layout_size); return memory_layout_size; }(); const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index 656baa550..ac00d4d9d 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -188,6 +188,8 @@ public: void DownloadMemory(VAddr cpu_addr, u64 size); + std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size); + bool InlineMemory(VAddr dest_address, size_t copy_size, std::span<const u8> inlined_buffer); void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size); @@ -379,8 +381,6 @@ private: void RunGarbageCollector(); - void WaitOnAsyncFlushes(VAddr cpu_addr, u64 size); - void BindHostIndexBuffer(); void BindHostVertexBuffers(); @@ -541,13 +541,10 @@ private: std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>, Empty> uniform_buffer_binding_sizes{}; - std::vector<BufferId> cached_write_buffer_ids; - MemoryTracker memory_tracker; IntervalSet uncommitted_ranges; IntervalSet common_ranges; IntervalSet cached_ranges; - IntervalSet pending_ranges; std::deque<IntervalSet> committed_ranges; // Async Buffers @@ -572,9 +569,8 @@ private: u64 critical_memory = 0; BufferId inline_buffer_id; - bool active_async_buffers = false; - std::array<BufferId, ((1ULL << 39) >> CACHING_PAGEBITS)> page_table; + std::vector<u8> tmp_buffer; }; } // namespace VideoCommon diff --git a/src/video_core/buffer_cache/memory_tracker_base.h b/src/video_core/buffer_cache/memory_tracker_base.h index dc4ebfcaa..6036b21c9 100644 --- a/src/video_core/buffer_cache/memory_tracker_base.h +++ b/src/video_core/buffer_cache/memory_tracker_base.h @@ -66,6 +66,14 @@ public: }); } + /// Returns true if a region has been marked as Preflushable + [[nodiscard]] bool IsRegionPreflushable(VAddr query_cpu_addr, u64 query_size) noexcept { + return IteratePages<false>( + query_cpu_addr, query_size, [](Manager* manager, u64 offset, size_t size) { + return manager->template IsRegionModified<Type::Preflushable>(offset, size); + }); + } + /// Mark region as CPU modified, notifying the rasterizer about this change void MarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 query_size) { IteratePages<true>(dirty_cpu_addr, query_size, @@ -93,6 +101,15 @@ public: }); } + /// Mark region as modified from the host GPU + void MarkRegionAsPreflushable(VAddr dirty_cpu_addr, u64 query_size) noexcept { + IteratePages<true>(dirty_cpu_addr, query_size, + [](Manager* manager, u64 offset, size_t size) { + manager->template ChangeRegionState<Type::Preflushable, true>( + manager->GetCpuAddr() + offset, size); + }); + } + /// Unmark region as modified from the host GPU void UnmarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 query_size) noexcept { IteratePages<true>(dirty_cpu_addr, query_size, @@ -102,6 +119,15 @@ public: }); } + /// Unmark region as modified from the host GPU + void UnmarkRegionAsPreflushable(VAddr dirty_cpu_addr, u64 query_size) noexcept { + IteratePages<true>(dirty_cpu_addr, query_size, + [](Manager* manager, u64 offset, size_t size) { + manager->template ChangeRegionState<Type::Preflushable, false>( + manager->GetCpuAddr() + offset, 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 query_size) { diff --git a/src/video_core/buffer_cache/word_manager.h b/src/video_core/buffer_cache/word_manager.h index a42455045..a336bde41 100644 --- a/src/video_core/buffer_cache/word_manager.h +++ b/src/video_core/buffer_cache/word_manager.h @@ -26,6 +26,7 @@ enum class Type { GPU, CachedCPU, Untracked, + Preflushable, }; /// Vector tracking modified pages tightly packed with small vector optimization @@ -55,17 +56,20 @@ struct Words { gpu.stack.fill(0); cached_cpu.stack.fill(0); untracked.stack.fill(~u64{0}); + preflushable.stack.fill(0); } else { // Share allocation between CPU and GPU pages and set their default values - u64* const alloc = new u64[num_words * 4]; + u64* const alloc = new u64[num_words * 5]; cpu.heap = alloc; gpu.heap = alloc + num_words; cached_cpu.heap = alloc + num_words * 2; untracked.heap = alloc + num_words * 3; + preflushable.heap = alloc + num_words * 4; 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}); + std::fill_n(preflushable.heap, num_words, 0); } // Clean up tailing bits const u64 last_word_size = size_bytes % BYTES_PER_WORD; @@ -88,13 +92,14 @@ struct Words { gpu = rhs.gpu; cached_cpu = rhs.cached_cpu; untracked = rhs.untracked; + preflushable = rhs.preflushable; rhs.cpu.heap = nullptr; return *this; } Words(Words&& rhs) noexcept : size_bytes{rhs.size_bytes}, num_words{rhs.num_words}, cpu{rhs.cpu}, gpu{rhs.gpu}, - cached_cpu{rhs.cached_cpu}, untracked{rhs.untracked} { + cached_cpu{rhs.cached_cpu}, untracked{rhs.untracked}, preflushable{rhs.preflushable} { rhs.cpu.heap = nullptr; } @@ -129,6 +134,8 @@ struct Words { return std::span<u64>(cached_cpu.Pointer(IsShort()), num_words); } else if constexpr (type == Type::Untracked) { return std::span<u64>(untracked.Pointer(IsShort()), num_words); + } else if constexpr (type == Type::Preflushable) { + return std::span<u64>(preflushable.Pointer(IsShort()), num_words); } } @@ -142,6 +149,8 @@ struct Words { return std::span<const u64>(cached_cpu.Pointer(IsShort()), num_words); } else if constexpr (type == Type::Untracked) { return std::span<const u64>(untracked.Pointer(IsShort()), num_words); + } else if constexpr (type == Type::Preflushable) { + return std::span<const u64>(preflushable.Pointer(IsShort()), num_words); } } @@ -151,6 +160,7 @@ struct Words { WordsArray<stack_words> gpu; WordsArray<stack_words> cached_cpu; WordsArray<stack_words> untracked; + WordsArray<stack_words> preflushable; }; template <class RasterizerInterface, size_t stack_words = 1> @@ -292,6 +302,9 @@ public: (pending_pointer - pending_offset) * BYTES_PER_PAGE); }; IterateWords(offset, size, [&](size_t index, u64 mask) { + if constexpr (type == Type::GPU) { + mask &= ~untracked_words[index]; + } const u64 word = state_words[index] & mask; if constexpr (clear) { if constexpr (type == Type::CPU || type == Type::CachedCPU) { @@ -340,8 +353,13 @@ public: static_assert(type != Type::Untracked); const std::span<const u64> state_words = words.template Span<type>(); + [[maybe_unused]] const std::span<const u64> untracked_words = + words.template Span<Type::Untracked>(); bool result = false; IterateWords(offset, size, [&](size_t index, u64 mask) { + if constexpr (type == Type::GPU) { + mask &= ~untracked_words[index]; + } const u64 word = state_words[index] & mask; if (word != 0) { result = true; @@ -362,9 +380,14 @@ public: [[nodiscard]] std::pair<u64, u64> ModifiedRegion(u64 offset, u64 size) const noexcept { static_assert(type != Type::Untracked); const std::span<const u64> state_words = words.template Span<type>(); + [[maybe_unused]] const std::span<const u64> untracked_words = + words.template Span<Type::Untracked>(); u64 begin = std::numeric_limits<u64>::max(); u64 end = 0; IterateWords(offset, size, [&](size_t index, u64 mask) { + if constexpr (type == Type::GPU) { + mask &= ~untracked_words[index]; + } const u64 word = state_words[index] & mask; if (word == 0) { return; diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index e68850dc5..ebe5536de 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -223,7 +223,7 @@ void MaxwellDMA::CopyBlockLinearToPitch() { write_buffer.resize_destructive(dst_size); memory_manager.ReadBlock(src_operand.address, read_buffer.data(), src_size); - memory_manager.ReadBlockUnsafe(dst_operand.address, write_buffer.data(), dst_size); + memory_manager.ReadBlock(dst_operand.address, write_buffer.data(), dst_size); UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, src_params.origin.y, x_elements, regs.line_count, block_height, block_depth, @@ -288,11 +288,7 @@ void MaxwellDMA::CopyPitchToBlockLinear() { write_buffer.resize_destructive(dst_size); memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); - if (Settings::IsGPULevelExtreme()) { - memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); - } else { - memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); - } + memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); // If the input is linear and the output is tiled, swizzle the input and copy it over. SwizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, diff --git a/src/video_core/engines/sw_blitter/blitter.cpp b/src/video_core/engines/sw_blitter/blitter.cpp index 3c9f38559..ff88cd03d 100644 --- a/src/video_core/engines/sw_blitter/blitter.cpp +++ b/src/video_core/engines/sw_blitter/blitter.cpp @@ -5,6 +5,7 @@ #include <cmath> #include <vector> +#include "common/scratch_buffer.h" #include "video_core/engines/sw_blitter/blitter.h" #include "video_core/engines/sw_blitter/converter.h" #include "video_core/memory_manager.h" @@ -112,11 +113,11 @@ void Bilinear(std::span<const f32> input, std::span<f32> output, size_t src_widt } // namespace struct SoftwareBlitEngine::BlitEngineImpl { - std::vector<u8> tmp_buffer; - std::vector<u8> src_buffer; - std::vector<u8> dst_buffer; - std::vector<f32> intermediate_src; - std::vector<f32> intermediate_dst; + Common::ScratchBuffer<u8> tmp_buffer; + Common::ScratchBuffer<u8> src_buffer; + Common::ScratchBuffer<u8> dst_buffer; + Common::ScratchBuffer<f32> intermediate_src; + Common::ScratchBuffer<f32> intermediate_dst; ConverterFactory converter_factory; }; @@ -158,14 +159,14 @@ bool SoftwareBlitEngine::Blit(Fermi2D::Surface& src, Fermi2D::Surface& dst, const auto src_bytes_per_pixel = BytesPerBlock(PixelFormatFromRenderTargetFormat(src.format)); const auto dst_bytes_per_pixel = BytesPerBlock(PixelFormatFromRenderTargetFormat(dst.format)); const size_t src_size = get_surface_size(src, src_bytes_per_pixel); - impl->tmp_buffer.resize(src_size); + impl->tmp_buffer.resize_destructive(src_size); memory_manager.ReadBlock(src.Address(), impl->tmp_buffer.data(), src_size); const size_t src_copy_size = src_extent_x * src_extent_y * src_bytes_per_pixel; const size_t dst_copy_size = dst_extent_x * dst_extent_y * dst_bytes_per_pixel; - impl->src_buffer.resize(src_copy_size); + impl->src_buffer.resize_destructive(src_copy_size); const bool no_passthrough = src.format != dst.format || src_extent_x != dst_extent_x || src_extent_y != dst_extent_y; @@ -177,8 +178,10 @@ bool SoftwareBlitEngine::Blit(Fermi2D::Surface& src, Fermi2D::Surface& dst, const auto convertion_phase_ir = [&]() { auto* input_converter = impl->converter_factory.GetFormatConverter(src.format); - impl->intermediate_src.resize((src_copy_size / src_bytes_per_pixel) * ir_components); - impl->intermediate_dst.resize((dst_copy_size / dst_bytes_per_pixel) * ir_components); + impl->intermediate_src.resize_destructive((src_copy_size / src_bytes_per_pixel) * + ir_components); + impl->intermediate_dst.resize_destructive((dst_copy_size / dst_bytes_per_pixel) * + ir_components); input_converter->ConvertTo(impl->src_buffer, impl->intermediate_src); if (config.filter != Fermi2D::Filter::Bilinear) { @@ -195,7 +198,7 @@ bool SoftwareBlitEngine::Blit(Fermi2D::Surface& src, Fermi2D::Surface& dst, // Do actual Blit - impl->dst_buffer.resize(dst_copy_size); + impl->dst_buffer.resize_destructive(dst_copy_size); if (src.linear == Fermi2D::MemoryLayout::BlockLinear) { UnswizzleSubrect(impl->src_buffer, impl->tmp_buffer, src_bytes_per_pixel, src.width, src.height, src.depth, config.src_x0, config.src_y0, src_extent_x, @@ -218,7 +221,7 @@ bool SoftwareBlitEngine::Blit(Fermi2D::Surface& src, Fermi2D::Surface& dst, } const size_t dst_size = get_surface_size(dst, dst_bytes_per_pixel); - impl->tmp_buffer.resize(dst_size); + impl->tmp_buffer.resize_destructive(dst_size); memory_manager.ReadBlock(dst.Address(), impl->tmp_buffer.data(), dst_size); if (dst.linear == Fermi2D::MemoryLayout::BlockLinear) { diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h index 3b2f6aab6..35d699bbf 100644 --- a/src/video_core/fence_manager.h +++ b/src/video_core/fence_manager.h @@ -59,6 +59,11 @@ public: buffer_cache.AccumulateFlushes(); } + void SignalReference() { + std::function<void()> do_nothing([] {}); + SignalFence(std::move(do_nothing)); + } + void SyncOperation(std::function<void()>&& func) { uncommitted_operations.emplace_back(std::move(func)); } diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 2e7f9c5ed..295a416a8 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -283,6 +283,21 @@ struct GPU::Impl { gpu_thread.FlushRegion(addr, size); } + VideoCore::RasterizerDownloadArea OnCPURead(VAddr addr, u64 size) { + auto raster_area = rasterizer->GetFlushArea(addr, size); + if (raster_area.preemtive) { + return raster_area; + } + raster_area.preemtive = true; + const u64 fence = RequestSyncOperation([this, &raster_area]() { + rasterizer->FlushRegion(raster_area.start_address, + raster_area.end_address - raster_area.start_address); + }); + gpu_thread.TickGPU(); + WaitForSyncOperation(fence); + return raster_area; + } + /// Notify rasterizer that any caches of the specified region should be invalidated void InvalidateRegion(VAddr addr, u64 size) { gpu_thread.InvalidateRegion(addr, size); @@ -538,6 +553,10 @@ void GPU::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { impl->SwapBuffers(framebuffer); } +VideoCore::RasterizerDownloadArea GPU::OnCPURead(VAddr addr, u64 size) { + return impl->OnCPURead(addr, size); +} + void GPU::FlushRegion(VAddr addr, u64 size) { impl->FlushRegion(addr, size); } diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index 8a871593a..e49c40cf2 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -10,6 +10,7 @@ #include "core/hle/service/nvdrv/nvdata.h" #include "video_core/cdma_pusher.h" #include "video_core/framebuffer_config.h" +#include "video_core/rasterizer_download_area.h" namespace Core { class System; @@ -241,6 +242,9 @@ public: void SwapBuffers(const Tegra::FramebufferConfig* framebuffer); /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory + [[nodiscard]] VideoCore::RasterizerDownloadArea OnCPURead(VAddr addr, u64 size); + + /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory void FlushRegion(VAddr addr, u64 size); /// Notify rasterizer that any caches of the specified region should be invalidated diff --git a/src/video_core/host1x/codecs/h264.cpp b/src/video_core/host1x/codecs/h264.cpp index e87bd65fa..6ce179167 100644 --- a/src/video_core/host1x/codecs/h264.cpp +++ b/src/video_core/host1x/codecs/h264.cpp @@ -111,7 +111,7 @@ const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegist writer.WriteUe(0); writer.WriteBit(context.h264_parameter_set.entropy_coding_mode_flag != 0); - writer.WriteBit(false); + writer.WriteBit(context.h264_parameter_set.pic_order_present_flag != 0); writer.WriteUe(0); writer.WriteUe(context.h264_parameter_set.num_refidx_l0_default_active); writer.WriteUe(context.h264_parameter_set.num_refidx_l1_default_active); @@ -129,7 +129,7 @@ const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegist writer.WriteBit(context.h264_parameter_set.redundant_pic_cnt_present_flag != 0); writer.WriteBit(context.h264_parameter_set.transform_8x8_mode_flag != 0); - writer.WriteBit(true); + writer.WriteBit(true); // pic_scaling_matrix_present_flag for (s32 index = 0; index < 6; index++) { writer.WriteBit(true); diff --git a/src/video_core/query_cache.h b/src/video_core/query_cache.h index 941de95c1..1528cc1dd 100644 --- a/src/video_core/query_cache.h +++ b/src/video_core/query_cache.h @@ -255,7 +255,6 @@ private: if (!in_range(query)) { continue; } - rasterizer.UpdatePagesCachedCount(query.GetCpuAddr(), query.SizeInBytes(), -1); AsyncJobId async_job_id = query.GetAsyncJob(); auto flush_result = query.Flush(async); if (async_job_id == NULL_ASYNC_JOB_ID) { @@ -273,7 +272,6 @@ private: /// Registers the passed parameters as cached and returns a pointer to the stored cached query. CachedQuery* Register(VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr, bool timestamp) { - rasterizer.UpdatePagesCachedCount(cpu_addr, CachedQuery::SizeInBytes(timestamp), 1); const u64 page = static_cast<u64>(cpu_addr) >> YUZU_PAGEBITS; return &cached_queries[page].emplace_back(static_cast<QueryCache&>(*this), type, cpu_addr, host_ptr); diff --git a/src/video_core/rasterizer_download_area.h b/src/video_core/rasterizer_download_area.h new file mode 100644 index 000000000..2d7425c79 --- /dev/null +++ b/src/video_core/rasterizer_download_area.h @@ -0,0 +1,16 @@ +// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include "common/common_types.h" + +namespace VideoCore { + +struct RasterizerDownloadArea { + VAddr start_address; + VAddr end_address; + bool preemtive; +}; + +} // namespace VideoCore
\ No newline at end of file diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index 33e2610bc..7566a8c4e 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -12,6 +12,7 @@ #include "video_core/cache_types.h" #include "video_core/engines/fermi_2d.h" #include "video_core/gpu.h" +#include "video_core/rasterizer_download_area.h" namespace Tegra { class MemoryManager; @@ -95,6 +96,8 @@ public: virtual bool MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) = 0; + virtual RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) = 0; + /// Notify rasterizer that any caches of the specified region should be invalidated virtual void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) = 0; diff --git a/src/video_core/renderer_null/null_rasterizer.cpp b/src/video_core/renderer_null/null_rasterizer.cpp index 2b5c7defa..bf2ce4c49 100644 --- a/src/video_core/renderer_null/null_rasterizer.cpp +++ b/src/video_core/renderer_null/null_rasterizer.cpp @@ -1,6 +1,8 @@ // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include "common/alignment.h" +#include "core/memory.h" #include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" #include "video_core/renderer_null/null_rasterizer.h" @@ -46,6 +48,14 @@ bool RasterizerNull::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheTyp } void RasterizerNull::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType) {} void RasterizerNull::OnCPUWrite(VAddr addr, u64 size) {} +VideoCore::RasterizerDownloadArea RasterizerNull::GetFlushArea(VAddr addr, u64 size) { + VideoCore::RasterizerDownloadArea new_area{ + .start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE), + .end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE), + .preemtive = true, + }; + return new_area; +} void RasterizerNull::InvalidateGPUCache() {} void RasterizerNull::UnmapMemory(VAddr addr, u64 size) {} void RasterizerNull::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) {} diff --git a/src/video_core/renderer_null/null_rasterizer.h b/src/video_core/renderer_null/null_rasterizer.h index 0c59e6a1f..a8d35d2c1 100644 --- a/src/video_core/renderer_null/null_rasterizer.h +++ b/src/video_core/renderer_null/null_rasterizer.h @@ -54,6 +54,7 @@ public: void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; void OnCPUWrite(VAddr addr, u64 size) override; + VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override; void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override; diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index 22ed16ebf..400c21981 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -108,7 +108,8 @@ bool IsASTCSupported() { [[nodiscard]] bool IsDebugToolAttached(std::span<const std::string_view> extensions) { const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED"); - return nsight || HasExtension(extensions, "GL_EXT_debug_tool"); + return nsight || HasExtension(extensions, "GL_EXT_debug_tool") || + Settings::values.renderer_debug.GetValue(); } } // Anonymous namespace diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 0089b4b27..f5baa0f3c 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -433,6 +433,29 @@ bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheT return false; } +VideoCore::RasterizerDownloadArea RasterizerOpenGL::GetFlushArea(VAddr addr, u64 size) { + { + std::scoped_lock lock{texture_cache.mutex}; + auto area = texture_cache.GetFlushArea(addr, size); + if (area) { + return *area; + } + } + { + std::scoped_lock lock{buffer_cache.mutex}; + auto area = buffer_cache.GetFlushArea(addr, size); + if (area) { + return *area; + } + } + VideoCore::RasterizerDownloadArea new_area{ + .start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE), + .end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE), + .preemtive = true, + }; + return new_area; +} + void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which) { MICROPROFILE_SCOPE(OpenGL_CacheManagement); if (addr == 0 || size == 0) { @@ -1281,7 +1304,7 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, const Tegra::DMA::ImageOperand& image_operand) { std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - const auto image_id = texture_cache.DmaImageId(image_operand); + const auto image_id = texture_cache.DmaImageId(image_operand, IS_IMAGE_UPLOAD); if (image_id == VideoCommon::NULL_IMAGE_ID) { return false; } diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index ad6978bd0..410d8ffc5 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -95,6 +95,7 @@ public: VideoCommon::CacheType which = VideoCommon::CacheType::All) override; bool MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; + VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; void OnCPUWrite(VAddr addr, u64 size) override; diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 47cccd0e5..31118886f 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -231,7 +231,7 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4 [[nodiscard]] bool CanBeAccelerated(const TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info) { - if (IsPixelFormatASTC(info.format) && !runtime.HasNativeASTC()) { + if (IsPixelFormatASTC(info.format) && info.size.depth == 1 && !runtime.HasNativeASTC()) { return Settings::values.accelerate_astc.GetValue() && !Settings::values.async_astc.GetValue(); } @@ -1126,7 +1126,8 @@ bool Image::ScaleDown(bool ignore) { ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info, ImageId image_id_, Image& image, const SlotVector<Image>&) - : VideoCommon::ImageViewBase{info, image.info, image_id_}, views{runtime.null_image_views} { + : VideoCommon::ImageViewBase{info, image.info, image_id_, image.gpu_addr}, + views{runtime.null_image_views} { const Device& device = runtime.device; if (True(image.flags & ImageFlagBits::Converted)) { internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8; @@ -1217,12 +1218,12 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, const VideoCommon::ImageViewInfo& view_info, GPUVAddr gpu_addr_) - : VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_}, + : VideoCommon::ImageViewBase{info, view_info, gpu_addr_}, buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {} ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, const VideoCommon::ImageViewInfo& view_info) - : VideoCommon::ImageViewBase{info, view_info} {} + : VideoCommon::ImageViewBase{info, view_info, 0} {} ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::NullImageViewParams& params) : VideoCommon::ImageViewBase{params}, views{runtime.null_image_views} {} @@ -1282,7 +1283,7 @@ GLuint ImageView::MakeView(Shader::TextureType view_type, GLenum view_format) { ApplySwizzle(view.handle, format, casted_swizzle); } if (set_object_label) { - const std::string name = VideoCommon::Name(*this); + const std::string name = VideoCommon::Name(*this, gpu_addr); glObjectLabel(GL_TEXTURE, view.handle, static_cast<GLsizei>(name.size()), name.data()); } return view.handle; diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 0dd039ed2..1190999a8 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -314,7 +314,6 @@ private: std::unique_ptr<StorageViews> storage_views; GLenum internal_format = GL_NONE; GLuint default_handle = 0; - GPUVAddr gpu_addr = 0; u32 buffer_size = 0; GLuint original_texture = 0; int num_samples = 0; diff --git a/src/video_core/renderer_vulkan/pipeline_helper.h b/src/video_core/renderer_vulkan/pipeline_helper.h index 28b893e25..983e1c2e1 100644 --- a/src/video_core/renderer_vulkan/pipeline_helper.h +++ b/src/video_core/renderer_vulkan/pipeline_helper.h @@ -176,7 +176,7 @@ public: }; inline void PushImageDescriptors(TextureCache& texture_cache, - UpdateDescriptorQueue& update_descriptor_queue, + GuestDescriptorQueue& guest_descriptor_queue, const Shader::Info& info, RescalingPushConstant& rescaling, const VkSampler*& samplers, const VideoCommon::ImageViewInOut*& views) { @@ -190,7 +190,7 @@ inline void PushImageDescriptors(TextureCache& texture_cache, const VkSampler sampler{*(samplers++)}; ImageView& image_view{texture_cache.GetImageView(image_view_id)}; const VkImageView vk_image_view{image_view.Handle(desc.type)}; - update_descriptor_queue.AddSampledImage(vk_image_view, sampler); + guest_descriptor_queue.AddSampledImage(vk_image_view, sampler); rescaling.PushTexture(texture_cache.IsRescaling(image_view)); } } @@ -201,7 +201,7 @@ inline void PushImageDescriptors(TextureCache& texture_cache, texture_cache.MarkModification(image_view.image_id); } const VkImageView vk_image_view{image_view.StorageView(desc.type, desc.format)}; - update_descriptor_queue.AddImage(vk_image_view); + guest_descriptor_queue.AddImage(vk_image_view); rescaling.PushImage(texture_cache.IsRescaling(image_view)); } } diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 510602e8e..9627eb129 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -298,12 +298,14 @@ private: BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_, Scheduler& scheduler_, StagingBufferPool& staging_pool_, - UpdateDescriptorQueue& update_descriptor_queue_, + GuestDescriptorQueue& guest_descriptor_queue_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue, DescriptorPool& 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) { + staging_pool{staging_pool_}, guest_descriptor_queue{guest_descriptor_queue_}, + uint8_pass(device, scheduler, descriptor_pool, staging_pool, compute_pass_descriptor_queue), + quad_index_pass(device, scheduler, descriptor_pool, staging_pool, + compute_pass_descriptor_queue) { quad_array_index_buffer = std::make_shared<QuadArrayIndexBuffer>(device_, memory_allocator_, scheduler_, staging_pool_); quad_strip_index_buffer = std::make_shared<QuadStripIndexBuffer>(device_, memory_allocator_, diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 879f1ed94..5e9602905 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -63,7 +63,8 @@ class BufferCacheRuntime { public: explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_, Scheduler& scheduler_, StagingBufferPool& staging_pool_, - UpdateDescriptorQueue& update_descriptor_queue_, + GuestDescriptorQueue& guest_descriptor_queue, + ComputePassDescriptorQueue& compute_pass_descriptor_queue, DescriptorPool& descriptor_pool); void Finish(); @@ -116,12 +117,12 @@ public: void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size, VideoCore::Surface::PixelFormat format) { - update_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format)); + guest_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format)); } private: void BindBuffer(VkBuffer buffer, u32 offset, u32 size) { - update_descriptor_queue.AddBuffer(buffer, offset, size); + guest_descriptor_queue.AddBuffer(buffer, offset, size); } void ReserveNullBuffer(); @@ -130,7 +131,7 @@ private: MemoryAllocator& memory_allocator; Scheduler& scheduler; StagingBufferPool& staging_pool; - UpdateDescriptorQueue& update_descriptor_queue; + GuestDescriptorQueue& guest_descriptor_queue; std::shared_ptr<QuadArrayIndexBuffer> quad_array_index_buffer; std::shared_ptr<QuadStripIndexBuffer> quad_strip_index_buffer; diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 1a316b6eb..3bc8553e1 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -200,12 +200,12 @@ ComputePass::~ComputePass() = default; Uint8Pass::Uint8Pass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_, - UpdateDescriptorQueue& update_descriptor_queue_) + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) : ComputePass(device_, descriptor_pool, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, {}, VULKAN_UINT8_COMP_SPV), scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, - update_descriptor_queue{update_descriptor_queue_} {} + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} Uint8Pass::~Uint8Pass() = default; @@ -214,10 +214,10 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16)); 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.buffer, staging.offset, staging_size); - const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + compute_pass_descriptor_queue.Acquire(); + compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices); + compute_pass_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) { @@ -242,12 +242,12 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer QuadIndexedPass::QuadIndexedPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - UpdateDescriptorQueue& update_descriptor_queue_) + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) : ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 3>, VULKAN_QUAD_INDEXED_COMP_SPV), scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, - update_descriptor_queue{update_descriptor_queue_} {} + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} QuadIndexedPass::~QuadIndexedPass() = default; @@ -272,10 +272,10 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( const std::size_t staging_size = num_tri_vertices * sizeof(u32); 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.buffer, staging.offset, staging_size); - const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + compute_pass_descriptor_queue.Acquire(); + compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size); + compute_pass_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, num_tri_vertices, base_vertex, index_shift, @@ -304,13 +304,14 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - UpdateDescriptorQueue& update_descriptor_queue_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_, MemoryAllocator& memory_allocator_) : ComputePass(device_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV), scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, - update_descriptor_queue{update_descriptor_queue_}, memory_allocator{memory_allocator_} {} + compute_pass_descriptor_queue{compute_pass_descriptor_queue_}, memory_allocator{ + memory_allocator_} {} ASTCDecoderPass::~ASTCDecoderPass() = default; @@ -358,11 +359,11 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); const u32 num_dispatches_z = image.info.resources.layers; - update_descriptor_queue.Acquire(); - update_descriptor_queue.AddBuffer(map.buffer, input_offset, - image.guest_size_bytes - swizzle.buffer_offset); - update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); - const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + compute_pass_descriptor_queue.Acquire(); + compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; // To unswizzle the ASTC data const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info); diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index c4c8fa081..dd3927376 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -9,6 +9,7 @@ #include "common/common_types.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" +#include "video_core/renderer_vulkan/vk_update_descriptor.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" @@ -21,7 +22,6 @@ namespace Vulkan { class Device; class StagingBufferPool; class Scheduler; -class UpdateDescriptorQueue; class Image; struct StagingBufferRef; @@ -50,7 +50,7 @@ class Uint8Pass final : public ComputePass { public: explicit Uint8Pass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - UpdateDescriptorQueue& update_descriptor_queue_); + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); ~Uint8Pass(); /// Assemble uint8 indices into an uint16 index buffer @@ -61,7 +61,7 @@ public: private: Scheduler& scheduler; StagingBufferPool& staging_buffer_pool; - UpdateDescriptorQueue& update_descriptor_queue; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; }; class QuadIndexedPass final : public ComputePass { @@ -69,7 +69,7 @@ public: explicit QuadIndexedPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - UpdateDescriptorQueue& update_descriptor_queue_); + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); ~QuadIndexedPass(); std::pair<VkBuffer, VkDeviceSize> Assemble( @@ -79,7 +79,7 @@ public: private: Scheduler& scheduler; StagingBufferPool& staging_buffer_pool; - UpdateDescriptorQueue& update_descriptor_queue; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; }; class ASTCDecoderPass final : public ComputePass { @@ -87,7 +87,7 @@ public: explicit ASTCDecoderPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - UpdateDescriptorQueue& update_descriptor_queue_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_, MemoryAllocator& memory_allocator_); ~ASTCDecoderPass(); @@ -97,7 +97,7 @@ public: private: Scheduler& scheduler; StagingBufferPool& staging_buffer_pool; - UpdateDescriptorQueue& update_descriptor_queue; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; MemoryAllocator& memory_allocator; }; diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index 2a0f0dbf0..733e70d9d 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -26,13 +26,13 @@ using Tegra::Texture::TexturePair; ComputePipeline::ComputePipeline(const Device& device_, vk::PipelineCache& pipeline_cache_, DescriptorPool& descriptor_pool, - UpdateDescriptorQueue& update_descriptor_queue_, + GuestDescriptorQueue& guest_descriptor_queue_, Common::ThreadWorker* thread_worker, PipelineStatistics* pipeline_statistics, VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_, vk::ShaderModule spv_module_) - : device{device_}, pipeline_cache(pipeline_cache_), - update_descriptor_queue{update_descriptor_queue_}, info{info_}, + : device{device_}, + pipeline_cache(pipeline_cache_), guest_descriptor_queue{guest_descriptor_queue_}, info{info_}, spv_module(std::move(spv_module_)) { if (shader_notify) { shader_notify->MarkShaderBuilding(); @@ -99,7 +99,7 @@ ComputePipeline::ComputePipeline(const Device& device_, vk::PipelineCache& pipel void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, Tegra::MemoryManager& gpu_memory, Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache) { - update_descriptor_queue.Acquire(); + guest_descriptor_queue.Acquire(); buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes); buffer_cache.UnbindComputeStorageBuffers(); @@ -194,7 +194,7 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, RescalingPushConstant rescaling; const VkSampler* samplers_it{samplers.data()}; const VideoCommon::ImageViewInOut* views_it{views.data()}; - PushImageDescriptors(texture_cache, update_descriptor_queue, info, rescaling, samplers_it, + PushImageDescriptors(texture_cache, guest_descriptor_queue, info, rescaling, samplers_it, views_it); if (!is_built.load(std::memory_order::relaxed)) { @@ -204,7 +204,7 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); }); } - const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + const void* const descriptor_data{guest_descriptor_queue.UpdateData()}; const bool is_rescaling = !info.texture_descriptors.empty() || !info.image_descriptors.empty(); scheduler.Record([this, descriptor_data, is_rescaling, rescaling_data = rescaling.Data()](vk::CommandBuffer cmdbuf) { diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.h b/src/video_core/renderer_vulkan/vk_compute_pipeline.h index 78d77027f..d1a1e2c46 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.h @@ -30,7 +30,7 @@ class ComputePipeline { public: explicit ComputePipeline(const Device& device, vk::PipelineCache& pipeline_cache, DescriptorPool& descriptor_pool, - UpdateDescriptorQueue& update_descriptor_queue, + GuestDescriptorQueue& guest_descriptor_queue, Common::ThreadWorker* thread_worker, PipelineStatistics* pipeline_statistics, VideoCore::ShaderNotify* shader_notify, const Shader::Info& info, @@ -48,7 +48,7 @@ public: private: const Device& device; vk::PipelineCache& pipeline_cache; - UpdateDescriptorQueue& update_descriptor_queue; + GuestDescriptorQueue& guest_descriptor_queue; Shader::Info info; VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{}; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index baedc4424..f1bcd5cd6 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -236,13 +236,13 @@ GraphicsPipeline::GraphicsPipeline( Scheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_, vk::PipelineCache& pipeline_cache_, VideoCore::ShaderNotify* shader_notify, const Device& device_, DescriptorPool& descriptor_pool, - UpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread, + GuestDescriptorQueue& guest_descriptor_queue_, Common::ThreadWorker* worker_thread, PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages, const std::array<const Shader::Info*, NUM_STAGES>& infos) : key{key_}, device{device_}, texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, pipeline_cache(pipeline_cache_), scheduler{scheduler_}, - update_descriptor_queue{update_descriptor_queue_}, spv_modules{std::move(stages)} { + guest_descriptor_queue{guest_descriptor_queue_}, spv_modules{std::move(stages)} { if (shader_notify) { shader_notify->MarkShaderBuilding(); } @@ -449,7 +449,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { buffer_cache.UpdateGraphicsBuffers(is_indexed); buffer_cache.BindHostGeometryBuffers(is_indexed); - update_descriptor_queue.Acquire(); + guest_descriptor_queue.Acquire(); RescalingPushConstant rescaling; RenderAreaPushConstant render_area; @@ -457,7 +457,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { const VideoCommon::ImageViewInOut* views_it{views.data()}; const auto prepare_stage{[&](size_t stage) LAMBDA_FORCEINLINE { buffer_cache.BindHostStageBuffers(stage); - PushImageDescriptors(texture_cache, update_descriptor_queue, stage_infos[stage], rescaling, + PushImageDescriptors(texture_cache, guest_descriptor_queue, stage_infos[stage], rescaling, samplers_it, views_it); const auto& info{stage_infos[0]}; if (info.uses_render_area) { @@ -499,7 +499,7 @@ void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling, const bool is_rescaling{texture_cache.IsRescaling()}; const bool update_rescaling{scheduler.UpdateRescaling(is_rescaling)}; const bool bind_pipeline{scheduler.UpdateGraphicsPipeline(this)}; - const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + const void* const descriptor_data{guest_descriptor_queue.UpdateData()}; scheduler.Record([this, descriptor_data, bind_pipeline, rescaling_data = rescaling.Data(), is_rescaling, update_rescaling, uses_render_area = render_area.uses_render_area, diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h index 67c657d0e..99e56e9ad 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h @@ -64,7 +64,6 @@ class RenderPassCache; class RescalingPushConstant; class RenderAreaPushConstant; class Scheduler; -class UpdateDescriptorQueue; class GraphicsPipeline { static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage; @@ -74,7 +73,7 @@ public: Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache, vk::PipelineCache& pipeline_cache, VideoCore::ShaderNotify* shader_notify, const Device& device, DescriptorPool& descriptor_pool, - UpdateDescriptorQueue& update_descriptor_queue, Common::ThreadWorker* worker_thread, + GuestDescriptorQueue& guest_descriptor_queue, Common::ThreadWorker* worker_thread, PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages, const std::array<const Shader::Info*, NUM_STAGES>& infos); @@ -133,7 +132,7 @@ private: BufferCache& buffer_cache; vk::PipelineCache& pipeline_cache; Scheduler& scheduler; - UpdateDescriptorQueue& update_descriptor_queue; + GuestDescriptorQueue& guest_descriptor_queue; void (*configure_func)(GraphicsPipeline*, bool){}; diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index e5219e7e0..66dfe5733 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -277,11 +277,11 @@ bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) c PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, - UpdateDescriptorQueue& update_descriptor_queue_, + GuestDescriptorQueue& guest_descriptor_queue_, RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_, TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_) : VideoCommon::ShaderCache{rasterizer_}, device{device_}, scheduler{scheduler_}, - descriptor_pool{descriptor_pool_}, update_descriptor_queue{update_descriptor_queue_}, + descriptor_pool{descriptor_pool_}, guest_descriptor_queue{guest_descriptor_queue_}, render_pass_cache{render_pass_cache_}, buffer_cache{buffer_cache_}, texture_cache{texture_cache_}, shader_notify{shader_notify_}, use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()}, @@ -643,7 +643,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline( Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; return std::make_unique<GraphicsPipeline>( scheduler, buffer_cache, texture_cache, vulkan_pipeline_cache, &shader_notify, device, - descriptor_pool, update_descriptor_queue, thread_worker, statistics, render_pass_cache, key, + descriptor_pool, guest_descriptor_queue, thread_worker, statistics, render_pass_cache, key, std::move(modules), infos); } catch (const Shader::Exception& exception) { @@ -723,7 +723,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline( } Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; return std::make_unique<ComputePipeline>(device, vulkan_pipeline_cache, descriptor_pool, - update_descriptor_queue, thread_worker, statistics, + guest_descriptor_queue, thread_worker, statistics, &shader_notify, program.info, std::move(spv_module)); } catch (const Shader::Exception& exception) { diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h index 5171912d7..15aa7e224 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h @@ -82,7 +82,6 @@ class PipelineStatistics; class RasterizerVulkan; class RenderPassCache; class Scheduler; -class UpdateDescriptorQueue; using VideoCommon::ShaderInfo; @@ -102,7 +101,7 @@ class PipelineCache : public VideoCommon::ShaderCache { public: explicit PipelineCache(RasterizerVulkan& rasterizer, const Device& device, Scheduler& scheduler, DescriptorPool& descriptor_pool, - UpdateDescriptorQueue& update_descriptor_queue, + GuestDescriptorQueue& guest_descriptor_queue, RenderPassCache& render_pass_cache, BufferCache& buffer_cache, TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_); ~PipelineCache(); @@ -144,7 +143,7 @@ private: const Device& device; Scheduler& scheduler; DescriptorPool& descriptor_pool; - UpdateDescriptorQueue& update_descriptor_queue; + GuestDescriptorQueue& guest_descriptor_queue; RenderPassCache& render_pass_cache; BufferCache& buffer_cache; TextureCache& texture_cache; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index d1489fc95..64bd2f6a5 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -160,17 +160,16 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra : RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, screen_info{screen_info_}, device{device_}, memory_allocator{memory_allocator_}, state_tracker{state_tracker_}, scheduler{scheduler_}, staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler), - update_descriptor_queue(device, scheduler), - blit_image(device, scheduler, state_tracker, descriptor_pool), - render_pass_cache(device), texture_cache_runtime{device, scheduler, - memory_allocator, staging_pool, - blit_image, render_pass_cache, - descriptor_pool, update_descriptor_queue}, + guest_descriptor_queue(device, scheduler), compute_pass_descriptor_queue(device, scheduler), + blit_image(device, scheduler, state_tracker, descriptor_pool), render_pass_cache(device), + texture_cache_runtime{ + device, scheduler, memory_allocator, staging_pool, + blit_image, render_pass_cache, descriptor_pool, compute_pass_descriptor_queue}, texture_cache(texture_cache_runtime, *this), buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool, - update_descriptor_queue, descriptor_pool), + guest_descriptor_queue, compute_pass_descriptor_queue, descriptor_pool), buffer_cache(*this, cpu_memory_, buffer_cache_runtime), - pipeline_cache(*this, device, scheduler, descriptor_pool, update_descriptor_queue, + pipeline_cache(*this, device, scheduler, descriptor_pool, guest_descriptor_queue, render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()), query_cache{*this, cpu_memory_, device, scheduler}, accelerate_dma(buffer_cache, texture_cache, scheduler), @@ -502,6 +501,22 @@ bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheT return false; } +VideoCore::RasterizerDownloadArea RasterizerVulkan::GetFlushArea(VAddr addr, u64 size) { + { + std::scoped_lock lock{texture_cache.mutex}; + auto area = texture_cache.GetFlushArea(addr, size); + if (area) { + return *area; + } + } + VideoCore::RasterizerDownloadArea new_area{ + .start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE), + .end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE), + .preemtive = true, + }; + return new_area; +} + void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which) { if (addr == 0 || size == 0) { return; @@ -598,7 +613,7 @@ void RasterizerVulkan::SignalSyncPoint(u32 value) { } void RasterizerVulkan::SignalReference() { - fence_manager.SignalOrdering(); + fence_manager.SignalReference(); } void RasterizerVulkan::ReleaseFences() { @@ -631,7 +646,7 @@ void RasterizerVulkan::WaitForIdle() { cmdbuf.SetEvent(event, flags); cmdbuf.WaitEvents(event, flags, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, {}, {}, {}); }); - SignalReference(); + fence_manager.SignalOrdering(); } void RasterizerVulkan::FragmentBarrier() { @@ -653,7 +668,8 @@ void RasterizerVulkan::FlushCommands() { void RasterizerVulkan::TickFrame() { draw_counter = 0; - update_descriptor_queue.TickFrame(); + guest_descriptor_queue.TickFrame(); + compute_pass_descriptor_queue.TickFrame(); fence_manager.TickFrame(); staging_pool.TickFrame(); { @@ -777,7 +793,7 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, const Tegra::DMA::ImageOperand& image_operand) { std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - const auto image_id = texture_cache.DmaImageId(image_operand); + const auto image_id = texture_cache.DmaImageId(image_operand, IS_IMAGE_UPLOAD); if (image_id == VideoCommon::NULL_IMAGE_ID) { return false; } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 1659fbc13..b39710b3c 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -92,6 +92,7 @@ public: VideoCommon::CacheType which = VideoCommon::CacheType::All) override; bool MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; + VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; void InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) override; @@ -183,7 +184,8 @@ private: StagingBufferPool staging_pool; DescriptorPool descriptor_pool; - UpdateDescriptorQueue update_descriptor_queue; + GuestDescriptorQueue guest_descriptor_queue; + ComputePassDescriptorQueue compute_pass_descriptor_queue; BlitImageHelper blit_image; RenderPassCache render_pass_cache; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index d0a7d8f35..4d0481f2a 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -798,13 +798,13 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched BlitImageHelper& blit_image_helper_, RenderPassCache& render_pass_cache_, DescriptorPool& descriptor_pool, - UpdateDescriptorQueue& update_descriptor_queue) + ComputePassDescriptorQueue& compute_pass_descriptor_queue) : device{device_}, scheduler{scheduler_}, memory_allocator{memory_allocator_}, staging_buffer_pool{staging_buffer_pool_}, blit_image_helper{blit_image_helper_}, render_pass_cache{render_pass_cache_}, resolution{Settings::values.resolution_info} { if (Settings::values.accelerate_astc) { astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, - update_descriptor_queue, memory_allocator); + compute_pass_descriptor_queue, memory_allocator); } } @@ -1268,7 +1268,7 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) { if (Settings::values.async_astc.GetValue()) { flags |= VideoCommon::ImageFlagBits::AsynchronousDecode; - } else if (Settings::values.accelerate_astc.GetValue()) { + } else if (Settings::values.accelerate_astc.GetValue() && info.size.depth == 1) { flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; } flags |= VideoCommon::ImageFlagBits::Converted; @@ -1584,8 +1584,9 @@ bool Image::NeedsScaleHelper() const { ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info, ImageId image_id_, Image& image) - : VideoCommon::ImageViewBase{info, image.info, image_id_}, device{&runtime.device}, - image_handle{image.Handle()}, samples(ConvertSampleCount(image.info.num_samples)) { + : VideoCommon::ImageViewBase{info, image.info, image_id_, image.gpu_addr}, + device{&runtime.device}, image_handle{image.Handle()}, + samples(ConvertSampleCount(image.info.num_samples)) { using Shader::TextureType; const VkImageAspectFlags aspect_mask = ImageViewAspectMask(info); @@ -1631,7 +1632,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI } vk::ImageView handle = device->GetLogical().CreateImageView(ci); if (device->HasDebuggingToolAttached()) { - handle.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); + handle.SetObjectNameEXT(VideoCommon::Name(*this, gpu_addr).c_str()); } image_views[static_cast<size_t>(tex_type)] = std::move(handle); }; @@ -1672,7 +1673,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, const VideoCommon::ImageViewInfo& view_info, GPUVAddr gpu_addr_) - : VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_}, + : VideoCommon::ImageViewBase{info, view_info, gpu_addr_}, buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {} ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParams& params) @@ -1863,6 +1864,7 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, num_layers = std::max(num_layers, color_buffer->range.extent.layers); images[num_images] = color_buffer->ImageHandle(); image_ranges[num_images] = MakeSubresourceRange(color_buffer); + rt_map[index] = num_images; samples = color_buffer->Samples(); ++num_images; } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index c656c5386..4166b3d20 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -34,7 +34,6 @@ class ImageView; class Framebuffer; class RenderPassCache; class StagingBufferPool; -class UpdateDescriptorQueue; class Scheduler; class TextureCacheRuntime { @@ -45,7 +44,7 @@ public: BlitImageHelper& blit_image_helper_, RenderPassCache& render_pass_cache_, DescriptorPool& descriptor_pool, - UpdateDescriptorQueue& update_descriptor_queue); + ComputePassDescriptorQueue& compute_pass_descriptor_queue); void Finish(); @@ -265,7 +264,6 @@ private: VkImage image_handle = VK_NULL_HANDLE; VkImageView render_target = VK_NULL_HANDLE; VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT; - GPUVAddr gpu_addr = 0; u32 buffer_size = 0; }; @@ -336,7 +334,7 @@ public: } [[nodiscard]] bool HasAspectColorBit(size_t index) const noexcept { - return (image_ranges.at(index).aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) != 0; + return (image_ranges.at(rt_map[index]).aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) != 0; } [[nodiscard]] bool HasAspectDepthBit() const noexcept { @@ -356,6 +354,7 @@ private: u32 num_images = 0; std::array<VkImage, 9> images{}; std::array<VkImageSubresourceRange, 9> image_ranges{}; + std::array<size_t, NUM_RT> rt_map{}; bool has_depth{}; bool has_stencil{}; }; diff --git a/src/video_core/renderer_vulkan/vk_update_descriptor.h b/src/video_core/renderer_vulkan/vk_update_descriptor.h index 1c1a7020b..310fb551a 100644 --- a/src/video_core/renderer_vulkan/vk_update_descriptor.h +++ b/src/video_core/renderer_vulkan/vk_update_descriptor.h @@ -32,7 +32,7 @@ class UpdateDescriptorQueue final { // This should be plenty for the vast majority of cases. Most desktop platforms only // provide up to 3 swapchain images. static constexpr size_t FRAMES_IN_FLIGHT = 5; - static constexpr size_t FRAME_PAYLOAD_SIZE = 0x10000; + static constexpr size_t FRAME_PAYLOAD_SIZE = 0x20000; static constexpr size_t PAYLOAD_SIZE = FRAME_PAYLOAD_SIZE * FRAMES_IN_FLIGHT; public: @@ -86,4 +86,8 @@ private: std::array<DescriptorUpdateEntry, PAYLOAD_SIZE> payload; }; +// TODO: should these be separate classes instead? +using GuestDescriptorQueue = UpdateDescriptorQueue; +using ComputePassDescriptorQueue = UpdateDescriptorQueue; + } // namespace Vulkan diff --git a/src/video_core/texture_cache/formatter.cpp b/src/video_core/texture_cache/formatter.cpp index 30f72361d..6279d8e9e 100644 --- a/src/video_core/texture_cache/formatter.cpp +++ b/src/video_core/texture_cache/formatter.cpp @@ -46,7 +46,7 @@ std::string Name(const ImageBase& image) { return "Invalid"; } -std::string Name(const ImageViewBase& image_view) { +std::string Name(const ImageViewBase& image_view, GPUVAddr addr) { const u32 width = image_view.size.width; const u32 height = image_view.size.height; const u32 depth = image_view.size.depth; @@ -56,23 +56,25 @@ std::string Name(const ImageViewBase& image_view) { const std::string level = num_levels > 1 ? fmt::format(":{}", num_levels) : ""; switch (image_view.type) { case ImageViewType::e1D: - return fmt::format("ImageView 1D {}{}", width, level); + return fmt::format("ImageView 1D 0x{:X} {}{}", addr, width, level); case ImageViewType::e2D: - return fmt::format("ImageView 2D {}x{}{}", width, height, level); + return fmt::format("ImageView 2D 0x{:X} {}x{}{}", addr, width, height, level); case ImageViewType::Cube: - return fmt::format("ImageView Cube {}x{}{}", width, height, level); + return fmt::format("ImageView Cube 0x{:X} {}x{}{}", addr, width, height, level); case ImageViewType::e3D: - return fmt::format("ImageView 3D {}x{}x{}{}", width, height, depth, level); + return fmt::format("ImageView 3D 0x{:X} {}x{}x{}{}", addr, width, height, depth, level); case ImageViewType::e1DArray: - return fmt::format("ImageView 1DArray {}{}|{}", width, level, num_layers); + return fmt::format("ImageView 1DArray 0x{:X} {}{}|{}", addr, width, level, num_layers); case ImageViewType::e2DArray: - return fmt::format("ImageView 2DArray {}x{}{}|{}", width, height, level, num_layers); + return fmt::format("ImageView 2DArray 0x{:X} {}x{}{}|{}", addr, width, height, level, + num_layers); case ImageViewType::CubeArray: - return fmt::format("ImageView CubeArray {}x{}{}|{}", width, height, level, num_layers); + return fmt::format("ImageView CubeArray 0x{:X} {}x{}{}|{}", addr, width, height, level, + num_layers); case ImageViewType::Rect: - return fmt::format("ImageView Rect {}x{}{}", width, height, level); + return fmt::format("ImageView Rect 0x{:X} {}x{}{}", addr, width, height, level); case ImageViewType::Buffer: - return fmt::format("BufferView {}", width); + return fmt::format("BufferView 0x{:X} {}", addr, width); } return "Invalid"; } diff --git a/src/video_core/texture_cache/formatter.h b/src/video_core/texture_cache/formatter.h index b97147797..9ee57a076 100644 --- a/src/video_core/texture_cache/formatter.h +++ b/src/video_core/texture_cache/formatter.h @@ -274,7 +274,7 @@ struct RenderTargets; [[nodiscard]] std::string Name(const ImageBase& image); -[[nodiscard]] std::string Name(const ImageViewBase& image_view); +[[nodiscard]] std::string Name(const ImageViewBase& image_view, GPUVAddr addr); [[nodiscard]] std::string Name(const RenderTargets& render_targets); diff --git a/src/video_core/texture_cache/image_info.cpp b/src/video_core/texture_cache/image_info.cpp index 11f3f78a1..e8ddde691 100644 --- a/src/video_core/texture_cache/image_info.cpp +++ b/src/video_core/texture_cache/image_info.cpp @@ -4,6 +4,7 @@ #include <fmt/format.h> #include "common/assert.h" +#include "common/settings.h" #include "video_core/surface.h" #include "video_core/texture_cache/format_lookup_table.h" #include "video_core/texture_cache/image_info.h" @@ -22,6 +23,8 @@ using VideoCore::Surface::PixelFormat; using VideoCore::Surface::SurfaceType; ImageInfo::ImageInfo(const TICEntry& config) noexcept { + forced_flushed = config.IsPitchLinear() && !Settings::values.use_reactive_flushing.GetValue(); + dma_downloaded = forced_flushed; format = PixelFormatFromTextureInfo(config.format, config.r_type, config.g_type, config.b_type, config.a_type, config.srgb_conversion); num_samples = NumSamples(config.msaa_mode); @@ -117,6 +120,9 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept { ImageInfo::ImageInfo(const Maxwell3D::Regs::RenderTargetConfig& ct, Tegra::Texture::MsaaMode msaa_mode) noexcept { + forced_flushed = + ct.tile_mode.is_pitch_linear && !Settings::values.use_reactive_flushing.GetValue(); + dma_downloaded = forced_flushed; format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(ct.format); rescaleable = false; if (ct.tile_mode.is_pitch_linear) { @@ -155,6 +161,9 @@ ImageInfo::ImageInfo(const Maxwell3D::Regs::RenderTargetConfig& ct, ImageInfo::ImageInfo(const Maxwell3D::Regs::Zeta& zt, const Maxwell3D::Regs::ZetaSize& zt_size, Tegra::Texture::MsaaMode msaa_mode) noexcept { + forced_flushed = + zt.tile_mode.is_pitch_linear && !Settings::values.use_reactive_flushing.GetValue(); + dma_downloaded = forced_flushed; format = VideoCore::Surface::PixelFormatFromDepthFormat(zt.format); size.width = zt_size.width; size.height = zt_size.height; @@ -195,6 +204,9 @@ ImageInfo::ImageInfo(const Maxwell3D::Regs::Zeta& zt, const Maxwell3D::Regs::Zet ImageInfo::ImageInfo(const Fermi2D::Surface& config) noexcept { UNIMPLEMENTED_IF_MSG(config.layer != 0, "Surface layer is not zero"); + forced_flushed = config.linear == Fermi2D::MemoryLayout::Pitch && + !Settings::values.use_reactive_flushing.GetValue(); + dma_downloaded = forced_flushed; format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(config.format); rescaleable = false; if (config.linear == Fermi2D::MemoryLayout::Pitch) { diff --git a/src/video_core/texture_cache/image_info.h b/src/video_core/texture_cache/image_info.h index 4b7dfa315..8a4cb0cbd 100644 --- a/src/video_core/texture_cache/image_info.h +++ b/src/video_core/texture_cache/image_info.h @@ -39,6 +39,8 @@ struct ImageInfo { u32 tile_width_spacing = 0; bool rescaleable = false; bool downscaleable = false; + bool forced_flushed = false; + bool dma_downloaded = false; }; } // namespace VideoCommon diff --git a/src/video_core/texture_cache/image_view_base.cpp b/src/video_core/texture_cache/image_view_base.cpp index 04fb84bfa..d134b6738 100644 --- a/src/video_core/texture_cache/image_view_base.cpp +++ b/src/video_core/texture_cache/image_view_base.cpp @@ -4,7 +4,6 @@ #include <algorithm> #include "common/assert.h" -#include "common/settings.h" #include "video_core/compatible_formats.h" #include "video_core/surface.h" #include "video_core/texture_cache/formatter.h" @@ -16,8 +15,8 @@ namespace VideoCommon { ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_info, - ImageId image_id_) - : image_id{image_id_}, format{info.format}, type{info.type}, range{info.range}, + ImageId image_id_, GPUVAddr addr) + : image_id{image_id_}, gpu_addr{addr}, format{info.format}, type{info.type}, range{info.range}, size{ .width = std::max(image_info.size.width >> range.base.level, 1u), .height = std::max(image_info.size.height >> range.base.level, 1u), @@ -26,8 +25,7 @@ ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_i ASSERT_MSG(VideoCore::Surface::IsViewCompatible(image_info.format, info.format, false, true), "Image view format {} is incompatible with image format {}", info.format, image_info.format); - const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue(); - if (image_info.type == ImageType::Linear && is_async) { + if (image_info.forced_flushed) { flags |= ImageViewFlagBits::PreemtiveDownload; } if (image_info.type == ImageType::e3D && info.type != ImageViewType::e3D) { @@ -35,8 +33,8 @@ ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_i } } -ImageViewBase::ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info) - : image_id{NULL_IMAGE_ID}, format{info.format}, type{ImageViewType::Buffer}, +ImageViewBase::ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info, GPUVAddr addr) + : image_id{NULL_IMAGE_ID}, gpu_addr{addr}, format{info.format}, type{ImageViewType::Buffer}, size{ .width = info.size.width, .height = 1, diff --git a/src/video_core/texture_cache/image_view_base.h b/src/video_core/texture_cache/image_view_base.h index 69c9776e7..a25ae1d4a 100644 --- a/src/video_core/texture_cache/image_view_base.h +++ b/src/video_core/texture_cache/image_view_base.h @@ -24,9 +24,9 @@ enum class ImageViewFlagBits : u16 { DECLARE_ENUM_FLAG_OPERATORS(ImageViewFlagBits) struct ImageViewBase { - explicit ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_info, - ImageId image_id); - explicit ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info); + explicit ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_info, ImageId image_id, + GPUVAddr addr); + explicit ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info, GPUVAddr addr); explicit ImageViewBase(const NullImageViewParams&); [[nodiscard]] bool IsBuffer() const noexcept { @@ -34,6 +34,7 @@ struct ImageViewBase { } ImageId image_id{}; + GPUVAddr gpu_addr = 0; PixelFormat format{}; ImageViewType type{}; SubresourceRange range; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index f335009d0..b24086fce 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -491,6 +491,32 @@ void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) { } template <class P> +std::optional<VideoCore::RasterizerDownloadArea> TextureCache<P>::GetFlushArea(VAddr cpu_addr, + u64 size) { + std::optional<VideoCore::RasterizerDownloadArea> area{}; + ForEachImageInRegion(cpu_addr, size, [&](ImageId, ImageBase& image) { + if (False(image.flags & ImageFlagBits::GpuModified)) { + return; + } + if (!area) { + area.emplace(); + area->start_address = cpu_addr; + area->end_address = cpu_addr + size; + area->preemtive = true; + } + area->start_address = std::min(area->start_address, image.cpu_addr); + area->end_address = std::max(area->end_address, image.cpu_addr_end); + for (auto image_view_id : image.image_view_ids) { + auto& image_view = slot_image_views[image_view_id]; + image_view.flags |= ImageViewFlagBits::PreemtiveDownload; + } + area->preemtive &= image.info.forced_flushed; + image.info.forced_flushed = true; + }); + return area; +} + +template <class P> void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) { std::vector<ImageId> deleted_images; ForEachImageInRegion(cpu_addr, size, [&](ImageId id, Image&) { deleted_images.push_back(id); }); @@ -683,6 +709,7 @@ void TextureCache<P>::CommitAsyncFlushes() { download_info.async_buffer_id = last_async_buffer_id; } } + if (any_none_dma) { auto download_map = runtime.DownloadStagingBuffer(total_size_bytes, true); for (const PendingDownload& download_info : download_ids) { @@ -695,6 +722,7 @@ void TextureCache<P>::CommitAsyncFlushes() { } uncommitted_async_buffers.emplace_back(download_map); } + async_buffers.emplace_back(std::move(uncommitted_async_buffers)); uncommitted_async_buffers.clear(); } @@ -783,17 +811,22 @@ void TextureCache<P>::PopAsyncFlushes() { } template <class P> -ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand) { +ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload) { const ImageInfo dst_info(operand); const ImageId dst_id = FindDMAImage(dst_info, operand.address); if (!dst_id) { return NULL_IMAGE_ID; } - const auto& image = slot_images[dst_id]; + auto& image = slot_images[dst_id]; if (False(image.flags & ImageFlagBits::GpuModified)) { // No need to waste time on an image that's synced with guest return NULL_IMAGE_ID; } + if (!is_upload && !image.info.dma_downloaded) { + // Force a full sync. + image.info.dma_downloaded = true; + return NULL_IMAGE_ID; + } const auto base = image.TryFindBase(operand.address); if (!base) { return NULL_IMAGE_ID; @@ -1290,7 +1323,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA all_siblings.push_back(overlap_id); } else { bad_overlap_ids.push_back(overlap_id); - overlap.flags |= ImageFlagBits::BadOverlap; } }; ForEachImageInRegion(cpu_addr, size_bytes, region_check); @@ -1359,6 +1391,12 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA ScaleDown(new_image); } + std::ranges::sort(overlap_ids, [this](const ImageId lhs, const ImageId rhs) { + const ImageBase& lhs_image = slot_images[lhs]; + const ImageBase& rhs_image = slot_images[rhs]; + return lhs_image.modification_tick < rhs_image.modification_tick; + }); + for (const ImageId overlap_id : overlap_ids) { Image& overlap = slot_images[overlap_id]; if (True(overlap.flags & ImageFlagBits::GpuModified)) { @@ -1395,7 +1433,12 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA ImageBase& aliased = slot_images[aliased_id]; aliased.overlapping_images.push_back(new_image_id); new_image.overlapping_images.push_back(aliased_id); - new_image.flags |= ImageFlagBits::BadOverlap; + if (aliased.info.resources.levels == 1 && aliased.overlapping_images.size() > 1) { + aliased.flags |= ImageFlagBits::BadOverlap; + } + if (new_image.info.resources.levels == 1 && new_image.overlapping_images.size() > 1) { + new_image.flags |= ImageFlagBits::BadOverlap; + } } RegisterImage(new_image_id); return new_image_id; @@ -1426,7 +1469,7 @@ std::optional<typename TextureCache<P>::BlitImages> TextureCache<P>::GetBlitImag if (!copy.must_accelerate) { do { if (!src_id && !dst_id) { - return std::nullopt; + break; } if (src_id && True(slot_images[src_id].flags & ImageFlagBits::GpuModified)) { break; diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 758b7e212..0720494e5 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -179,6 +179,8 @@ public: /// Download contents of host images to guest memory in a region void DownloadMemory(VAddr cpu_addr, size_t size); + std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size); + /// Remove images in a region void UnmapMemory(VAddr cpu_addr, size_t size); @@ -205,7 +207,7 @@ public: /// Pop asynchronous downloads void PopAsyncFlushes(); - [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand); + [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload); [[nodiscard]] std::pair<Image*, BufferImageCopy> DmaBufferImageCopy( const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index de37db684..f1071aa23 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -896,11 +896,11 @@ void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8 ASSERT(copy.buffer_row_length == Common::AlignUp(mip_size.width, tile_size.width)); ASSERT(copy.buffer_image_height == Common::AlignUp(mip_size.height, tile_size.height)); if (IsPixelFormatASTC(info.format)) { - ASSERT(copy.image_extent.depth == 1); - Tegra::Texture::ASTC::Decompress(input.subspan(copy.buffer_offset), - copy.image_extent.width, copy.image_extent.height, - copy.image_subresource.num_layers, tile_size.width, - tile_size.height, output.subspan(output_offset)); + Tegra::Texture::ASTC::Decompress( + input.subspan(copy.buffer_offset), copy.image_extent.width, + copy.image_extent.height, + copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width, + tile_size.height, output.subspan(output_offset)); } else { DecompressBC4(input.subspan(copy.buffer_offset), copy.image_extent, output.subspan(output_offset)); diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 6ffca2af2..161f050b8 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -1009,6 +1009,8 @@ void Device::CollectPhysicalMemoryInfo() { device_access_memory += mem_properties.memoryHeaps[element].size; } if (!is_integrated) { + const u64 reserve_memory = std::min<u64>(device_access_memory / 8, 1_GiB); + device_access_memory -= reserve_memory; return; } const s64 available_memory = static_cast<s64>(device_access_memory - device_initial_usage); diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 7d5018151..5f1c63ff9 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -10,6 +10,7 @@ #include <vector> #include "common/common_types.h" +#include "common/settings.h" #include "video_core/vulkan_common/vulkan_wrapper.h" // Define all features which may be used by the implementation here. @@ -510,7 +511,7 @@ public: /// Returns true when a known debugging tool is attached. bool HasDebuggingToolAttached() const { - return has_renderdoc || has_nsight_graphics; + return has_renderdoc || has_nsight_graphics || Settings::values.renderer_debug.GetValue(); } /// Returns true when the device does not properly support cube compatibility. diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp index 1732866e0..e28a556f8 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp @@ -147,7 +147,7 @@ public: /// Returns whether this allocation is compatible with the arguments. [[nodiscard]] bool IsCompatible(VkMemoryPropertyFlags flags, u32 type_mask) const { - return (flags & property_flags) == property_flags && (type_mask & shifted_memory_type) != 0; + return (flags & property_flags) == flags && (type_mask & shifted_memory_type) != 0; } private: |