From 3476140ba2a1bd0e157eb73425caec6b325227c6 Mon Sep 17 00:00:00 2001
From: Stenzek <stenzek@gmail.com>
Date: Mon, 13 Jan 2025 16:08:20 +1000
Subject: [PATCH] GPUThread: Add BeginASyncBufferCall()

---
 src/common/types.h             |  1 +
 src/core/gpu_thread.cpp        | 30 ++++++++++++++++++++++++------
 src/core/gpu_thread.h          |  3 +++
 src/core/gpu_thread_commands.h | 10 +---------
 4 files changed, 29 insertions(+), 15 deletions(-)

diff --git a/src/common/types.h b/src/common/types.h
index 85b7d3024..b14915bb6 100644
--- a/src/common/types.h
+++ b/src/common/types.h
@@ -103,6 +103,7 @@ char (&__countof_ArraySizeHelper(T (&array)[N]))[N];
 // disable warnings that show up at warning level 4
 // TODO: Move to build system instead
 #ifdef _MSC_VER
+#pragma warning(disable : 4200) // warning C4200: nonstandard extension used: zero-sized array in struct/union
 #pragma warning(disable : 4201) // warning C4201: nonstandard extension used : nameless struct/union
 #pragma warning(disable : 4100) // warning C4100: 'Platform' : unreferenced formal parameter
 #pragma warning(disable : 4355) // warning C4355: 'this' : used in base member initializer list
diff --git a/src/core/gpu_thread.cpp b/src/core/gpu_thread.cpp
index f1e48690d..6834f10a0 100644
--- a/src/core/gpu_thread.cpp
+++ b/src/core/gpu_thread.cpp
@@ -61,6 +61,8 @@ static bool Reconfigure(std::string serial, std::optional<GPURenderer> renderer,
 
 // NOTE: Use with care! The handler needs to manually run the destructor.
 template<class T, typename... Args>
+T* AllocateCommand(u32 size, GPUBackendCommandType type, Args... args);
+template<class T, typename... Args>
 T* AllocateCommand(GPUBackendCommandType type, Args... args);
 
 static u32 GetPendingCommandSize();
@@ -248,21 +250,27 @@ GPUThreadCommand* GPUThread::AllocateCommand(GPUBackendCommandType command, u32
 }
 
 template<class T, typename... Args>
-T* GPUThread::AllocateCommand(GPUBackendCommandType command, Args... args)
+T* GPUThread::AllocateCommand(u32 size, GPUBackendCommandType type, Args... args)
 {
-  const u32 size = GPUThreadCommand::AlignCommandSize(sizeof(T));
-  GPUThreadCommand* cmd = AllocateCommand(command, size);
-  DebugAssert(cmd->size == size);
+  const u32 alloc_size = GPUThreadCommand::AlignCommandSize(size);
+  GPUThreadCommand* cmd = AllocateCommand(type, alloc_size);
+  DebugAssert(cmd->size == alloc_size);
 
   new (cmd) T(std::forward<Args>(args)...);
 
   // constructor may overwrite the fields, need to reset them
-  cmd->type = command;
-  cmd->size = size;
+  cmd->type = type;
+  cmd->size = alloc_size;
 
   return static_cast<T*>(cmd);
 }
 
+template<class T, typename... Args>
+T* GPUThread::AllocateCommand(GPUBackendCommandType type, Args... args)
+{
+  return AllocateCommand<T>(sizeof(T), type, std::forward<Args>(args)...);
+}
+
 u32 GPUThread::GetPendingCommandSize()
 {
   const u32 read_ptr = s_state.command_fifo_read_ptr.load(std::memory_order_acquire);
@@ -995,6 +1003,16 @@ void GPUThread::RunOnBackend(AsyncBackendCallType func, bool sync, bool spin_or_
     PushCommand(cmd);
 }
 
+std::pair<GPUThreadCommand*, void*> GPUThread::BeginASyncBufferCall(AsyncBufferCallType func, u32 buffer_size)
+{
+  // this is less than optimal, but it's only used for input osd updates currently, so whatever
+  GPUThreadAsyncCallCommand* const cmd = AllocateCommand<GPUThreadAsyncCallCommand>(
+    sizeof(GPUThreadAsyncCallCommand) + buffer_size, GPUBackendCommandType::AsyncCall);
+  void* const buffer = static_cast<void*>(cmd + 1);
+  cmd->func = [func, buffer]() { func(buffer); };
+  return std::make_pair(static_cast<GPUThreadCommand*>(cmd), buffer);
+}
+
 void GPUThread::UpdateSettings(bool gpu_settings_changed, bool device_settings_changed)
 {
   if (device_settings_changed)
diff --git a/src/core/gpu_thread.h b/src/core/gpu_thread.h
index d8debf55b..1737a5d1c 100644
--- a/src/core/gpu_thread.h
+++ b/src/core/gpu_thread.h
@@ -7,6 +7,7 @@
 
 #include <functional>
 #include <optional>
+#include <utility>
 
 class Error;
 struct WindowInfo;
@@ -28,6 +29,7 @@ struct GPUBackendUpdateDisplayCommand;
 namespace GPUThread {
 using AsyncCallType = std::function<void()>;
 using AsyncBackendCallType = std::function<void(GPUBackend*)>;
+using AsyncBufferCallType = void(*)(void*);
 
 enum class RunIdleReason : u8
 {
@@ -70,6 +72,7 @@ bool IsOnThread();
 bool IsUsingThread();
 void RunOnThread(AsyncCallType func);
 void RunOnBackend(AsyncBackendCallType func, bool sync, bool spin_or_wake);
+std::pair<GPUThreadCommand*, void*> BeginASyncBufferCall(AsyncBufferCallType func, u32 buffer_size);
 void SetVSync(GPUVSyncMode mode, bool allow_present_throttle);
 
 // Should only be called on the GPU thread.
diff --git a/src/core/gpu_thread_commands.h b/src/core/gpu_thread_commands.h
index 9fda31e14..7b289fda5 100644
--- a/src/core/gpu_thread_commands.h
+++ b/src/core/gpu_thread_commands.h
@@ -23,11 +23,6 @@ namespace System {
 struct MemorySaveState;
 }
 
-#ifdef _MSC_VER
-#pragma warning(push)
-#pragma warning(disable : 4200) // warning C4200: nonstandard extension used: zero-sized array in struct/union
-#endif
-
 enum class GPUBackendCommandType : u8
 {
   Wraparound,
@@ -87,6 +82,7 @@ struct GPUThreadReconfigureCommand : public GPUThreadCommand
 
 struct GPUThreadAsyncCallCommand : public GPUThreadCommand
 {
+  GPUThreadAsyncCallCommand() = default;
   GPUThreadAsyncCallCommand(std::function<void()> func_) : func(std::move(func_)) {}
 
   std::function<void()> func;
@@ -329,7 +325,3 @@ struct GPUBackendDrawPreciseLineCommand : public GPUBackendDrawCommand
 
   Vertex vertices[0];
 };
-
-#ifdef _MSC_VER
-#pragma warning(pop)
-#endif