early-access version 1845

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1844.
+This is the source code for early-access 1845.
 
 ## Legal Notice
 

src/tests/video_core/buffer_base.cpp

@@ -536,7 +536,7 @@ TEST_CASE("BufferBase: Cached write downloads") {
     REQUIRE(rasterizer.Count() == 63);
     buffer.MarkRegionAsGpuModified(c + PAGE, PAGE);
     int num = 0;
-    buffer.ForEachDownloadRange(c, WORD, true, [&](u64 offset, u64 size) { ++num; });
+    buffer.ForEachDownloadRange(c, WORD, [&](u64 offset, u64 size) { ++num; });
     buffer.ForEachUploadRange(c, WORD, [&](u64 offset, u64 size) { ++num; });
     REQUIRE(num == 0);
     REQUIRE(!buffer.IsRegionCpuModified(c + PAGE, PAGE));

src/video_core/buffer_cache/buffer_base.h

@@ -226,19 +226,19 @@ public:
     /// Call 'func' for each CPU modified range and unmark those pages as CPU modified
     template <typename Func>
     void ForEachUploadRange(VAddr query_cpu_range, u64 size, Func&& func) {
-        ForEachModifiedRange<Type::CPU>(query_cpu_range, size, true, func);
+        ForEachModifiedRange<Type::CPU>(query_cpu_range, size, func);
     }
 
     /// Call 'func' for each GPU modified range and unmark those pages as GPU modified
     template <typename Func>
-    void ForEachDownloadRange(VAddr query_cpu_range, u64 size, bool clear, Func&& func) {
+    void ForEachDownloadRange(VAddr query_cpu_range, u64 size, Func&& func) {
-        ForEachModifiedRange<Type::GPU>(query_cpu_range, size, clear, func);
+        ForEachModifiedRange<Type::GPU>(query_cpu_range, size, func);
     }
 
     /// Call 'func' for each GPU modified range and unmark those pages as GPU modified
     template <typename Func>
     void ForEachDownloadRange(Func&& func) {
-        ForEachModifiedRange<Type::GPU>(cpu_addr, SizeBytes(), true, func);
+        ForEachModifiedRange<Type::GPU>(cpu_addr, SizeBytes(), func);
     }
 
     /// Mark buffer as picked
@@ -415,7 +415,7 @@ private:
      * @param func            Function to call for each turned off region
      */
     template <Type type, typename Func>
-    void ForEachModifiedRange(VAddr query_cpu_range, s64 size, bool clear, Func&& func) {
+    void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) {
         static_assert(type != Type::Untracked);
 
         const s64 difference = query_cpu_range - cpu_addr;
@@ -467,9 +467,7 @@ private:
             bits = (bits << left_offset) >> left_offset;
 
             const u64 current_word = state_words[word_index] & bits;
-            if (clear) {
             state_words[word_index] &= ~bits;
-            }
 
             if constexpr (type == Type::CPU) {
                 const u64 current_bits = untracked_words[word_index] & bits;

src/video_core/buffer_cache/buffer_cache.h

@@ -15,7 +15,6 @@
 #include <vector>
 
 #include <boost/container/small_vector.hpp>
-#include <boost/icl/interval_set.hpp>
 
 #include "common/common_types.h"
 #include "common/div_ceil.h"
@@ -78,9 +77,6 @@ class BufferCache {
     using Runtime = typename P::Runtime;
     using Buffer = typename P::Buffer;
 
-    using IntervalSet = boost::icl::interval_set<VAddr>;
-    using IntervalType = typename IntervalSet::interval_type;
-
     struct Empty {};
 
     struct OverlapResult {
@@ -157,7 +153,6 @@ public:
 
     /// Commit asynchronous downloads
     void CommitAsyncFlushes();
-    void CommitAsyncFlushesHigh();
 
     /// Pop asynchronous downloads
     void PopAsyncFlushes();
@@ -165,9 +160,6 @@ public:
     /// Return true when a CPU region is modified from the GPU
     [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
 
-    /// Return true when a CPU region is modified from the GPU
-    [[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size);
-
     std::mutex mutex;
 
 private:
@@ -280,6 +272,8 @@ private:
 
     void DeleteBuffer(BufferId buffer_id);
 
+    void ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id);
+
     void NotifyBufferDeletion();
 
     [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr) const;
@@ -333,7 +327,9 @@ private:
 
     std::vector<BufferId> cached_write_buffer_ids;
 
-    IntervalSet uncommitted_ranges;
+    // TODO: This data structure is not optimal and it should be reworked
+    std::vector<BufferId> uncommitted_downloads;
+    std::deque<std::vector<BufferId>> committed_downloads;
 
     size_t immediate_buffer_capacity = 0;
     std::unique_ptr<u8[]> immediate_buffer_alloc;
@@ -551,18 +547,29 @@ void BufferCache<P>::FlushCachedWrites() {
 
 template <class P>
 bool BufferCache<P>::HasUncommittedFlushes() const noexcept {
-    return !uncommitted_ranges.empty();
+    return !uncommitted_downloads.empty();
 }
 
 template <class P>
 bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept {
-    return false;
+    return !committed_downloads.empty() && !committed_downloads.front().empty();
 }
 
 template <class P>
-void BufferCache<P>::CommitAsyncFlushesHigh() {
+void BufferCache<P>::CommitAsyncFlushes() {
-    const IntervalSet& intervals = uncommitted_ranges;
+    // This is intentionally passing the value by copy
-    if (intervals.empty()) {
+    committed_downloads.push_front(uncommitted_downloads);
+    uncommitted_downloads.clear();
+}
+
+template <class P>
+void BufferCache<P>::PopAsyncFlushes() {
+    if (committed_downloads.empty()) {
+        return;
+    }
+    auto scope_exit_pop_download = detail::ScopeExit([this] { committed_downloads.pop_back(); });
+    const std::span<const BufferId> download_ids = committed_downloads.back();
+    if (download_ids.empty()) {
         return;
     }
     MICROPROFILE_SCOPE(GPU_DownloadMemory);
@@ -570,35 +577,18 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
     boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads;
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
-    for (auto& interval : intervals) {
+    for (const BufferId buffer_id : download_ids) {
-        const std::size_t size = interval.upper() - interval.lower();
+        slot_buffers[buffer_id].ForEachDownloadRange([&](u64 range_offset, u64 range_size) {
-        const VAddr cpu_addr = interval.lower();
-        const VAddr cpu_addr_end = interval.upper();
-        ForEachBufferInRange(cpu_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
-            boost::container::small_vector<BufferCopy, 1> copies;
-            buffer.ForEachDownloadRange(
-                cpu_addr, size, false, [&](u64 range_offset, u64 range_size) {
-                    VAddr cpu_addr_base = buffer.CpuAddr() + range_offset;
-                    VAddr cpu_addr_end2 = cpu_addr_base + range_size;
-                    const s64 difference = s64(cpu_addr_end2 - cpu_addr_end);
-                    cpu_addr_end2 -= u64(std::max<s64>(difference, 0));
-                    const s64 difference2 = s64(cpu_addr - cpu_addr_base);
-                    cpu_addr_base += u64(std::max<s64>(difference2, 0));
-                    const u64 new_size = cpu_addr_end2 - cpu_addr_base;
-                    const u64 new_offset = cpu_addr_base - buffer.CpuAddr();
-                    ASSERT(!IsRegionCpuModified(cpu_addr_base, new_size));
             downloads.push_back({
                 BufferCopy{
-                            .src_offset = new_offset,
+                    .src_offset = range_offset,
                     .dst_offset = total_size_bytes,
-                            .size = new_size,
+                    .size = range_size,
                 },
                 buffer_id,
             });
-                    total_size_bytes += new_size;
+            total_size_bytes += range_size;
-                    buffer.UnmarkRegionAsGpuModified(cpu_addr_base, new_size);
+            largest_copy = std::max(largest_copy, range_size);
-                    largest_copy = std::max(largest_copy, new_size);
-                });
         });
     }
     if (downloads.empty()) {
@@ -632,18 +622,6 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
     }
 }
 
-template <class P>
-void BufferCache<P>::CommitAsyncFlushes() {
-    if (Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::High) {
-        CommitAsyncFlushesHigh();
-    } else {
-        uncommitted_ranges.clear();
-    }
-}
-
-template <class P>
-void BufferCache<P>::PopAsyncFlushes() {}
-
 template <class P>
 bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
     const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE);
@@ -663,25 +641,6 @@ bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
     return false;
 }
 
-template <class P>
-bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) {
-    const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE);
-    for (u64 page = addr >> PAGE_BITS; page < page_end;) {
-        const BufferId image_id = page_table[page];
-        if (!image_id) {
-            ++page;
-            continue;
-        }
-        Buffer& buffer = slot_buffers[image_id];
-        if (buffer.IsRegionCpuModified(addr, size)) {
-            return true;
-        }
-        const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes();
-        page = Common::DivCeil(end_addr, PAGE_SIZE);
-    }
-    return false;
-}
-
 template <class P>
 void BufferCache<P>::BindHostIndexBuffer() {
     Buffer& buffer = slot_buffers[index_buffer.buffer_id];
@@ -1051,14 +1010,16 @@ void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 s
     Buffer& buffer = slot_buffers[buffer_id];
     buffer.MarkRegionAsGpuModified(cpu_addr, size);
 
-    const bool is_accuracy_high =
+    const bool is_accuracy_high = Settings::IsGPULevelHigh();
-        Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::High;
     const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue();
-    if (!is_async && !is_accuracy_high) {
+    if (!is_accuracy_high || !is_async) {
         return;
     }
-    const IntervalType base_interval{cpu_addr, cpu_addr + size};
+    if (std::ranges::find(uncommitted_downloads, buffer_id) != uncommitted_downloads.end()) {
-    uncommitted_ranges.add(base_interval);
+        // Already inserted
+        return;
+    }
+    uncommitted_downloads.push_back(buffer_id);
 }
 
 template <class P>
@@ -1142,6 +1103,7 @@ void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id,
     if (!copies.empty()) {
         runtime.CopyBuffer(slot_buffers[new_buffer_id], overlap, copies);
     }
+    ReplaceBufferDownloads(overlap_id, new_buffer_id);
     DeleteBuffer(overlap_id);
 }
 
@@ -1282,7 +1244,7 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, VAddr cpu_addr, u64 si
     boost::container::small_vector<BufferCopy, 1> copies;
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
-    buffer.ForEachDownloadRange(cpu_addr, size, true, [&](u64 range_offset, u64 range_size) {
+    buffer.ForEachDownloadRange(cpu_addr, size, [&](u64 range_offset, u64 range_size) {
         copies.push_back(BufferCopy{
             .src_offset = range_offset,
             .dst_offset = total_size_bytes,
@@ -1353,6 +1315,18 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id) {
     NotifyBufferDeletion();
 }
 
+template <class P>
+void BufferCache<P>::ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id) {
+    const auto replace = [old_buffer_id, new_buffer_id](std::vector<BufferId>& buffers) {
+        std::ranges::replace(buffers, old_buffer_id, new_buffer_id);
+        if (auto it = std::ranges::find(buffers, new_buffer_id); it != buffers.end()) {
+            buffers.erase(std::remove(it + 1, buffers.end(), new_buffer_id), buffers.end());
+        }
+    };
+    replace(uncommitted_downloads);
+    std::ranges::for_each(committed_downloads, replace);
+}
+
 template <class P>
 void BufferCache<P>::NotifyBufferDeletion() {
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {

src/video_core/engines/fermi_2d.cpp

@@ -67,10 +67,10 @@ void Fermi2D::Blit() {
     };
     Surface src = regs.src;
     const auto bytes_per_pixel = BytesPerBlock(PixelFormatFromRenderTargetFormat(src.format));
-    const auto is_copy_out_of_bound =
+    const auto need_align_to_pitch =
         src.linear == Tegra::Engines::Fermi2D::MemoryLayout::Pitch && src.width == config.src_x1 &&
         config.src_x1 > static_cast<s32>(src.pitch / bytes_per_pixel) && config.src_x0 > 0;
-    if (is_copy_out_of_bound) {
+    if (need_align_to_pitch) {
         auto address = src.Address() + config.src_x0 * bytes_per_pixel;
         src.addr_upper = static_cast<u32>(address >> 32);
         src.addr_lower = static_cast<u32>(address);

src/video_core/fence_manager.h

@@ -96,23 +96,6 @@ public:
         }
     }
 
-    void TryReleasePendingFences() {
-        while (!fences.empty()) {
-            TFence& current_fence = fences.front();
-            if (ShouldWait() && !IsFenceSignaled(current_fence)) {
-                return;
-            }
-            PopAsyncFlushes();
-            if (current_fence->IsSemaphore()) {
-                gpu_memory.template Write<u32>(current_fence->GetAddress(),
-                                               current_fence->GetPayload());
-            } else {
-                gpu.IncrementSyncPoint(current_fence->GetPayload());
-            }
-            PopFence();
-        }
-    }
-
 protected:
     explicit FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
                           TTextureCache& texture_cache_, TTBufferCache& buffer_cache_,
@@ -142,6 +125,23 @@ protected:
     TQueryCache& query_cache;
 
 private:
+    void TryReleasePendingFences() {
+        while (!fences.empty()) {
+            TFence& current_fence = fences.front();
+            if (ShouldWait() && !IsFenceSignaled(current_fence)) {
+                return;
+            }
+            PopAsyncFlushes();
+            if (current_fence->IsSemaphore()) {
+                gpu_memory.template Write<u32>(current_fence->GetAddress(),
+                                               current_fence->GetPayload());
+            } else {
+                gpu.IncrementSyncPoint(current_fence->GetPayload());
+            }
+            PopFence();
+        }
+    }
+
     bool ShouldWait() const {
         std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         return texture_cache.ShouldWaitAsyncFlushes() || buffer_cache.ShouldWaitAsyncFlushes() ||

src/video_core/gpu_thread.cpp

@@ -8,7 +8,6 @@
 #include "common/settings.h"
 #include "common/thread.h"
 #include "core/core.h"
-#include "core/core_timing.h"
 #include "core/frontend/emu_window.h"
 #include "video_core/dma_pusher.h"
 #include "video_core/gpu.h"
@@ -84,17 +83,6 @@ void ThreadManager::StartThread(VideoCore::RendererBase& renderer,
     rasterizer = renderer.ReadRasterizer();
     thread = std::thread(RunThread, std::ref(system), std::ref(renderer), std::ref(context),
                          std::ref(dma_pusher), std::ref(state));
-    gpu_sync_event = Core::Timing::CreateEvent(
-        "GPUHostSyncCallback", [this](std::uintptr_t, std::chrono::nanoseconds) {
-            if (!state.is_running) {
-                return;
-            }
-
-            OnCommandListEnd();
-            const auto time_interval = std::chrono::nanoseconds{500 * 1000};
-            system.CoreTiming().ScheduleEvent(time_interval, gpu_sync_event);
-        });
-    system.CoreTiming().ScheduleEvent(std::chrono::nanoseconds{500 * 1000}, gpu_sync_event);
 }
 
 void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
@@ -140,9 +128,6 @@ void ThreadManager::ShutDown() {
         state.cv.notify_all();
     }
 
-    system.CoreTiming().UnscheduleEvent(gpu_sync_event, 0);
-    system.CoreTiming().RemoveEvent(gpu_sync_event);
-
     if (!thread.joinable()) {
         return;
     }

src/video_core/gpu_thread.h

@@ -20,10 +20,6 @@ class DmaPusher;
 } // namespace Tegra
 
 namespace Core {
-namespace Timing {
-class CoreTiming;
-struct EventType;
-} // namespace Timing
 namespace Frontend {
 class GraphicsContext;
 }
@@ -154,7 +150,6 @@ private:
 
     SynchState state;
     std::thread thread;
-    std::shared_ptr<Core::Timing::EventType> gpu_sync_event;
 };
 
 } // namespace VideoCommon::GPUThread

src/video_core/renderer_vulkan/vk_fence_manager.cpp

@@ -34,10 +34,6 @@ bool InnerFence::IsSignaled() const {
     if (is_stubbed) {
         return true;
     }
-    if (scheduler.IsFree(wait_tick)) {
-        return true;
-    }
-    scheduler.Refresh();
     return scheduler.IsFree(wait_tick);
 }
 

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -604,7 +604,7 @@ void RasterizerVulkan::ReleaseFences() {
     if (!gpu.IsAsync()) {
         return;
     }
-    fence_manager.TryReleasePendingFences();
+    fence_manager.WaitPendingFences();
 }
 
 void RasterizerVulkan::FlushAndInvalidateRegion(VAddr addr, u64 size) {

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -83,10 +83,6 @@ public:
         return master_semaphore->IsFree(tick);
     }
 
-    void Refresh() const noexcept {
-        return master_semaphore->Refresh();
-    }
-
     /// Waits for the given tick to trigger on the GPU.
     void Wait(u64 tick) {
         master_semaphore->Wait(tick);

src/video_core/texture_cache/texture_cache.h

@@ -769,20 +769,20 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
     const ImageBase& src_image = slot_images[src_id];
 
     // TODO: Deduplicate
-    const std::optional dst_base = dst_image.TryFindBase(dst.Address());
+    const std::optional src_base = src_image.TryFindBase(src.Address());
-    const SubresourceRange dst_range{.base = dst_base.value(), .extent = {1, 1}};
+    const SubresourceRange src_range{.base = src_base.value(), .extent = {1, 1}};
-    const ImageViewInfo dst_view_info(ImageViewType::e2D, images.dst_format, dst_range);
+    const ImageViewInfo src_view_info(ImageViewType::e2D, images.src_format, src_range);
-    const auto [dst_framebuffer_id, dst_view_id] = RenderTargetFromImage(dst_id, dst_view_info);
+    const auto [src_framebuffer_id, src_view_id] = RenderTargetFromImage(src_id, src_view_info);
     const auto [src_samples_x, src_samples_y] = SamplesLog2(src_image.info.num_samples);
     const Region2D src_region{
         Offset2D{.x = copy.src_x0 >> src_samples_x, .y = copy.src_y0 >> src_samples_y},
         Offset2D{.x = copy.src_x1 >> src_samples_x, .y = copy.src_y1 >> src_samples_y},
     };
 
-    const std::optional src_base = src_image.TryFindBase(src.Address());
+    const std::optional dst_base = dst_image.TryFindBase(dst.Address());
-    const SubresourceRange src_range{.base = src_base.value(), .extent = {1, 1}};
+    const SubresourceRange dst_range{.base = dst_base.value(), .extent = {1, 1}};
-    const ImageViewInfo src_view_info(ImageViewType::e2D, images.src_format, src_range);
+    const ImageViewInfo dst_view_info(ImageViewType::e2D, images.dst_format, dst_range);
-    const auto [src_framebuffer_id, src_view_id] = RenderTargetFromImage(src_id, src_view_info);
+    const auto [dst_framebuffer_id, dst_view_id] = RenderTargetFromImage(dst_id, dst_view_info);
     const auto [dst_samples_x, dst_samples_y] = SamplesLog2(dst_image.info.num_samples);
     const Region2D dst_region{
         Offset2D{.x = copy.dst_x0 >> dst_samples_x, .y = copy.dst_y0 >> dst_samples_y},

src/video_core/texture_cache/types.h

@@ -133,8 +133,8 @@ struct BufferImageCopy {
 };
 
 struct BufferCopy {
-    u64 src_offset;
+    size_t src_offset;
-    u64 dst_offset;
+    size_t dst_offset;
     size_t size;
 };
 

src/yuzu/debugger/profiler.cpp

@@ -143,24 +143,25 @@ void MicroProfileWidget::hideEvent(QHideEvent* ev) {
 }
 
 void MicroProfileWidget::mouseMoveEvent(QMouseEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, 0);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale, 0);
     ev->accept();
 }
 
 void MicroProfileWidget::mousePressEvent(QMouseEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, 0);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale, 0);
     MicroProfileMouseButton(ev->buttons() & Qt::LeftButton, ev->buttons() & Qt::RightButton);
     ev->accept();
 }
 
 void MicroProfileWidget::mouseReleaseEvent(QMouseEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, 0);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale, 0);
     MicroProfileMouseButton(ev->buttons() & Qt::LeftButton, ev->buttons() & Qt::RightButton);
     ev->accept();
 }
 
 void MicroProfileWidget::wheelEvent(QWheelEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, ev->delta() / 120);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale,
+                              ev->angleDelta().y() / 120);
     ev->accept();
 }