video_core: dma_pusher: Add support for prefetched command lists.

src/video_core/dma_pusher.cpp

@@ -45,18 +45,28 @@ bool DmaPusher::Step() {
         return false;
     }
 
-    const CommandList& command_list{dma_pushbuffer.front()};
-    ASSERT_OR_EXECUTE(!command_list.empty(), {
+    CommandList& command_list{dma_pushbuffer.front()};
+
+    ASSERT_OR_EXECUTE(
+        command_list.command_lists.size() || command_list.prefetch_command_list.size(), {
             // Somehow the command_list is empty, in order to avoid a crash
             // We ignore it and assume its size is 0.
             dma_pushbuffer.pop();
             dma_pushbuffer_subindex = 0;
             return true;
         });
-    const CommandListHeader command_list_header{command_list[dma_pushbuffer_subindex++]};
+
+    if (command_list.prefetch_command_list.size()) {
+        // Prefetched command list from nvdrv, used for things like synchronization
+        command_headers = std::move(command_list.prefetch_command_list);
+        dma_pushbuffer.pop();
+    } else {
+        const CommandListHeader command_list_header{
+            command_list.command_lists[dma_pushbuffer_subindex]};
+        const u64 next_hash = command_list.command_list_hashes[dma_pushbuffer_subindex++];
         const GPUVAddr dma_get = command_list_header.addr;
 
-    if (dma_pushbuffer_subindex >= command_list.size()) {
+        if (dma_pushbuffer_subindex >= command_list.command_lists.size()) {
             // We've gone through the current list, remove it from the queue
             dma_pushbuffer.pop();
             dma_pushbuffer_subindex = 0;
@@ -70,7 +80,7 @@ bool DmaPusher::Step() {
         command_headers.resize(command_list_header.size);
         gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(),
                                             command_list_header.size * sizeof(u32));
-
+    }
     for (std::size_t index = 0; index < command_headers.size();) {
         const CommandHeader& command_header = command_headers[index];
 

src/video_core/dma_pusher.h

@@ -74,9 +74,26 @@ union CommandHeader {
 static_assert(std::is_standard_layout_v<CommandHeader>, "CommandHeader is not standard layout");
 static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!");
 
+static constexpr CommandHeader BuildCommandHeader(BufferMethods method, u32 arg_count,
+                                                  SubmissionMode mode) {
+    CommandHeader result{};
+    result.method.Assign(static_cast<u32>(method));
+    result.arg_count.Assign(arg_count);
+    result.mode.Assign(mode);
+    return result;
+}
+
 class GPU;
 
-using CommandList = std::vector<Tegra::CommandListHeader>;
+struct CommandList final {
+    CommandList() = default;
+    explicit CommandList(std::size_t size) : command_lists(size) {}
+    explicit CommandList(std::vector<Tegra::CommandHeader>&& prefetch_command_list)
+        : prefetch_command_list{std::move(prefetch_command_list)} {}
+
+    std::vector<Tegra::CommandListHeader> command_lists;
+    std::vector<Tegra::CommandHeader> prefetch_command_list;
+};
 
 /**
  * The DmaPusher class implements DMA submission to FIFOs, providing an area of memory that the
@@ -85,7 +102,7 @@ using CommandList = std::vector<Tegra::CommandListHeader>;
  * See https://envytools.readthedocs.io/en/latest/hw/fifo/dma-pusher.html#fifo-dma-pusher for
  * details on this implementation.
  */
-class DmaPusher {
+class DmaPusher final {
 public:
     explicit DmaPusher(Core::System& system, GPU& gpu);
     ~DmaPusher();