early-access version 3584

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 3583.
+This is the source code for early-access 3584.
 
 ## Legal Notice
 

src/video_core/buffer_cache/buffer_cache.h

@@ -131,33 +131,15 @@ std::optional<VideoCore::RasterizerDownloadArea> BufferCache<P>::GetFlushArea(VA
 
 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);
     }
@@ -177,7 +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;
@@ -205,7 +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.
@@ -492,7 +472,6 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
     }
     MICROPROFILE_SCOPE(GPU_DownloadMemory);
 
-    pending_ranges.clear();
     auto it = committed_ranges.begin();
     while (it != committed_ranges.end()) {
         auto& current_intervals = *it;
@@ -1232,7 +1211,6 @@ void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 s
     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>

src/video_core/buffer_cache/buffer_cache_base.h

@@ -381,8 +381,6 @@ private:
 
     void RunGarbageCollector();
 
-    void WaitOnAsyncFlushes(VAddr cpu_addr, u64 size);
-
     void BindHostIndexBuffer();
 
     void BindHostVertexBuffers();
@@ -547,7 +545,6 @@ private:
     IntervalSet uncommitted_ranges;
     IntervalSet common_ranges;
     IntervalSet cached_ranges;
-    IntervalSet pending_ranges;
     std::deque<IntervalSet> committed_ranges;
 
     // Async Buffers

src/video_core/texture_cache/texture_cache.h

@@ -1397,27 +1397,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
         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)) {
-            new_image.flags |= ImageFlagBits::GpuModified;
-            const auto& resolution = Settings::values.resolution_info;
-            const SubresourceBase base = new_image.TryFindBase(overlap.gpu_addr).value();
-            const u32 up_scale = can_rescale ? resolution.up_scale : 1;
-            const u32 down_shift = can_rescale ? resolution.down_shift : 0;
-            auto copies = MakeShrinkImageCopies(new_info, overlap.info, base, up_scale, down_shift);
-            if (overlap.info.num_samples != new_image.info.num_samples) {
-                runtime.CopyImageMSAA(new_image, overlap, std::move(copies));
-            } else {
-                runtime.CopyImage(new_image, overlap, std::move(copies));
-            }
-        }
-        if (True(overlap.flags & ImageFlagBits::Tracked)) {
-            UntrackImage(overlap, overlap_id);
-        }
-        UnregisterImage(overlap_id);
-        DeleteImage(overlap_id);
-    }
     ImageBase& new_image_base = new_image;
     for (const ImageId aliased_id : right_aliased_ids) {
         ImageBase& aliased = slot_images[aliased_id];
@@ -1440,6 +1419,33 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             new_image.flags |= ImageFlagBits::BadOverlap;
         }
     }
+
+    SynchronizeAliases(new_image_id);
+
+    for (const ImageId overlap_id : overlap_ids) {
+        Image& overlap = slot_images[overlap_id];
+        if (True(overlap.flags & ImageFlagBits::GpuModified) &&
+            overlap.modification_tick > new_image.modification_tick) {
+            new_image.flags |= ImageFlagBits::GpuModified;
+            const auto& resolution = Settings::values.resolution_info;
+            const SubresourceBase base = new_image.TryFindBase(overlap.gpu_addr).value();
+            const u32 up_scale = can_rescale ? resolution.up_scale : 1;
+            const u32 down_shift = can_rescale ? resolution.down_shift : 0;
+            auto copies = MakeShrinkImageCopies(new_info, overlap.info, base, up_scale, down_shift);
+            if (overlap.info.num_samples != new_image.info.num_samples) {
+                runtime.CopyImageMSAA(new_image, overlap, std::move(copies));
+            } else {
+                runtime.CopyImage(new_image, overlap, std::move(copies));
+            }
+            new_image.modification_tick = overlap.modification_tick;
+        }
+        if (True(overlap.flags & ImageFlagBits::Tracked)) {
+            UntrackImage(overlap, overlap_id);
+        }
+        UnregisterImage(overlap_id);
+        DeleteImage(overlap_id);
+    }
+
     RegisterImage(new_image_id);
     return new_image_id;
 }