2 files changed, 48 insertions, 28 deletions

diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 8be7bd594..9b2698fad 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -272,13 +272,19 @@ std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainBuffer(GPUVAddr gpu_ad
     if (!cpu_addr) {
         return {&slot_buffers[NULL_BUFFER_ID], 0};
     }
-    const BufferId buffer_id = FindBuffer(*cpu_addr, size);
+    return ObtainCPUBuffer(*cpu_addr, size, sync_info, post_op);
+}
+
+template <class P>
+std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainCPUBuffer(
+    VAddr cpu_addr, u32 size, ObtainBufferSynchronize sync_info, ObtainBufferOperation post_op) {
+    const BufferId buffer_id = FindBuffer(cpu_addr, size);
     Buffer& buffer = slot_buffers[buffer_id];
 
     // synchronize op
     switch (sync_info) {
     case ObtainBufferSynchronize::FullSynchronize:
-        SynchronizeBuffer(buffer, *cpu_addr, size);
+        SynchronizeBuffer(buffer, cpu_addr, size);
         break;
     default:
         break;
@@ -286,11 +292,11 @@ std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainBuffer(GPUVAddr gpu_ad
 
     switch (post_op) {
     case ObtainBufferOperation::MarkAsWritten:
-        MarkWrittenBuffer(buffer_id, *cpu_addr, size);
+        MarkWrittenBuffer(buffer_id, cpu_addr, size);
         break;
     case ObtainBufferOperation::DiscardWrite: {
-        VAddr cpu_addr_start = Common::AlignDown(*cpu_addr, 64);
-        VAddr cpu_addr_end = Common::AlignUp(*cpu_addr + size, 64);
+        VAddr cpu_addr_start = Common::AlignDown(cpu_addr, 64);
+        VAddr cpu_addr_end = Common::AlignUp(cpu_addr + size, 64);
         IntervalType interval{cpu_addr_start, cpu_addr_end};
         ClearDownload(interval);
         common_ranges.subtract(interval);
@@ -300,7 +306,7 @@ std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainBuffer(GPUVAddr gpu_ad
         break;
     }
 
-    return {&buffer, buffer.Offset(*cpu_addr)};
+    return {&buffer, buffer.Offset(cpu_addr)};
 }
 
 template <class P>
@@ -538,7 +544,7 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
         it++;
     }
 
-    boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads;
+    boost::container::small_vector<std::pair<BufferCopy, BufferId>, 16> downloads;
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
     for (const IntervalSet& intervals : committed_ranges) {
@@ -908,6 +914,11 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
+
+        if (is_written) {
+            MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+        }
+
         if constexpr (NEEDS_BIND_STORAGE_INDEX) {
             runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written);
             ++binding_index;
@@ -925,6 +936,11 @@ void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) {
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
+        const bool is_written = ((channel_state->written_texture_buffers[stage] >> index) & 1) != 0;
+        if (is_written) {
+            MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+        }
+
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const PixelFormat format = binding.format;
         if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
@@ -956,6 +972,8 @@ void BufferCache<P>::BindHostTransformFeedbackBuffers() {
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
+        MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+
         const u32 offset = buffer.Offset(binding.cpu_addr);
         host_bindings.buffers.push_back(&buffer);
         host_bindings.offsets.push_back(offset);
@@ -1005,6 +1023,11 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const bool is_written =
             ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
+
+        if (is_written) {
+            MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+        }
+
         if constexpr (NEEDS_BIND_STORAGE_INDEX) {
             runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written);
             ++binding_index;
@@ -1022,6 +1045,12 @@ void BufferCache<P>::BindHostComputeTextureBuffers() {
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
+        const bool is_written =
+            ((channel_state->written_compute_texture_buffers >> index) & 1) != 0;
+        if (is_written) {
+            MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+        }
+
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const PixelFormat format = binding.format;
         if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
@@ -1195,16 +1224,11 @@ void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
 
 template <class P>
 void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
-    const u32 written_mask = channel_state->written_storage_buffers[stage];
     ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
         // Resolve buffer
         Binding& binding = channel_state->storage_buffers[stage][index];
         const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size);
         binding.buffer_id = buffer_id;
-        // Mark buffer as written if needed
-        if (((written_mask >> index) & 1) != 0) {
-            MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size);
-        }
     });
 }
 
@@ -1213,10 +1237,6 @@ void BufferCache<P>::UpdateTextureBuffers(size_t stage) {
     ForEachEnabledBit(channel_state->enabled_texture_buffers[stage], [&](u32 index) {
         Binding& binding = channel_state->texture_buffers[stage][index];
         binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
-        // Mark buffer as written if needed
-        if (((channel_state->written_texture_buffers[stage] >> index) & 1) != 0) {
-            MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
-        }
     });
 }
 
@@ -1246,7 +1266,6 @@ void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) {
         .size = size,
         .buffer_id = buffer_id,
     };
-    MarkWrittenBuffer(buffer_id, *cpu_addr, size);
 }
 
 template <class P>
@@ -1273,10 +1292,6 @@ void BufferCache<P>::UpdateComputeStorageBuffers() {
         // Resolve buffer
         Binding& binding = channel_state->compute_storage_buffers[index];
         binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
-        // Mark as written if needed
-        if (((channel_state->written_compute_storage_buffers >> index) & 1) != 0) {
-            MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
-        }
     });
 }
 
@@ -1285,18 +1300,11 @@ void BufferCache<P>::UpdateComputeTextureBuffers() {
     ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) {
         Binding& binding = channel_state->compute_texture_buffers[index];
         binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
-        // Mark as written if needed
-        if (((channel_state->written_compute_texture_buffers >> index) & 1) != 0) {
-            MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
-        }
     });
 }
 
 template <class P>
 void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 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};
diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h
index 0b7135d49..c4f6e8d12 100644
--- a/src/video_core/buffer_cache/buffer_cache_base.h
+++ b/src/video_core/buffer_cache/buffer_cache_base.h
@@ -295,6 +295,10 @@ public:
     [[nodiscard]] std::pair<Buffer*, u32> ObtainBuffer(GPUVAddr gpu_addr, u32 size,
                                                        ObtainBufferSynchronize sync_info,
                                                        ObtainBufferOperation post_op);
+
+    [[nodiscard]] std::pair<Buffer*, u32> ObtainCPUBuffer(VAddr gpu_addr, u32 size,
+                                                          ObtainBufferSynchronize sync_info,
+                                                          ObtainBufferOperation post_op);
     void FlushCachedWrites();
 
     /// Return true when there are uncommitted buffers to be downloaded
@@ -335,6 +339,14 @@ public:
 
     [[nodiscard]] std::pair<Buffer*, u32> GetDrawIndirectBuffer();
 
+    template <typename Func>
+    void BufferOperations(Func&& func) {
+        do {
+            channel_state->has_deleted_buffers = false;
+            func();
+        } while (channel_state->has_deleted_buffers);
+    }
+
     std::recursive_mutex mutex;
     Runtime& runtime;