Sync the whole buffer insteed of only the range

src/video_core/buffer_cache/buffer_cache.cpp

@@ -909,18 +909,18 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr,
     return true;
 }
 
-void BufferCache::SynchronizeRange(VAddr device_addr, u64 size) {
+void BufferCache::SynchronizeBuffersInRange(VAddr device_addr, u64 size) {
     if (device_addr == 0) {
         return;
     }
     VAddr device_addr_end = device_addr + size;
     ForEachBufferInRange(device_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
-        VAddr buffer_start = buffer.CpuAddr();
-        VAddr buffer_end = buffer_start + buffer.SizeBytes();
-        VAddr start = std::max(buffer_start, device_addr);
-        VAddr end = std::min(buffer_end, device_addr_end);
-        u32 size = static_cast<u32>(end - start);
-        SynchronizeBuffer(buffer, start, size, false);
+        // Note that this function synchronizes the whole buffer, not just the range.
+        // This is because this function is used to sync buffers before using a
+        // shader that uses DMA.
+        // The ideal solution would be to sync all the mapped regions but it is
+        // very slow.
+        SynchronizeBuffer(buffer, buffer.CpuAddr(), buffer.SizeBytes(), false);
     });
 }
 

src/video_core/buffer_cache/buffer_cache.h

@@ -137,7 +137,7 @@ public:
     void ProcessFaultBuffer();
 
     /// Synchronizes all buffers in the specified range.
-    void SynchronizeRange(VAddr device_addr, u64 size);
+    void SynchronizeBuffersInRange(VAddr device_addr, u64 size);
 
     /// Record memory barrier. Used for buffers when accessed via BDA.
     void MemoryBarrier();

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -474,14 +474,15 @@ bool Rasterizer::BindResources(const Pipeline* pipeline) {
 
     pipeline->BindResources(set_writes, buffer_barriers, push_data);
 
-    if (uses_dma && !fault_process_pending) {
+    if (uses_dma) {
         // We only use fault buffer for DMA right now.
         {
             std::shared_lock lock{dma_sync_mapped_ranges_mutex};
             for (const auto& range : dma_sync_mapped_ranges) {
-                buffer_cache.SynchronizeRange(range.lower(), range.upper() - range.lower());
+                buffer_cache.SynchronizeBuffersInRange(range.lower(), range.upper() - range.lower());
             }
         }
+        buffer_cache.MemoryBarrier();
     }
 
     fault_process_pending |= uses_dma;

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -91,11 +91,17 @@ void Scheduler::Wait(u64 tick) {
     }
     master_semaphore.Wait(tick);
 
-    // Apply pending operations until the wait tick
-    while (!pending_ops.empty() && pending_ops.front().gpu_tick <= tick) {
-        pending_ops.front().callback();
-        pending_ops.pop();
-    }
+    // TODO: We should be applyting pending operations here because that gives us
+    // the ability to use mapped regions on stream buffers in deferred operations.
+    // We don't do that right now because it might introduce varioations in the
+    // timing and, since we don't sync the GPU some games might be affected by that.
+    // It shouldn't be an issue right now, because we only use mapped regions in
+    // deferred operations to download faulted addresses. That is only 8KB every tick
+    // and the stream buffer is 256MB. GPU doesn't go that behind.
+    // while (!pending_ops.empty() && pending_ops.front().gpu_tick <= tick) {
+    //     pending_ops.front().callback();
+    //     pending_ops.pop();
+    // }
 }
 
 void Scheduler::AllocateWorkerCommandBuffers() {