CommandQueueMT: Reduce lock contention in cases of single flusher

Author: RandomShaper

core/templates/command_queue_mt.cpp

@@ -30,7 +30,8 @@
 
 #include "command_queue_mt.h"
 
-CommandQueueMT::CommandQueueMT() {
+CommandQueueMT::CommandQueueMT(bool p_unique_flusher) :
+		unique_flusher(p_unique_flusher) {
 	command_mem.reserve(DEFAULT_COMMAND_MEM_SIZE_KB * 1024);
 }
 

core/templates/command_queue_mt.h

@@ -107,6 +107,7 @@ class CommandQueueMT {
 
 	static const uint32_t DEFAULT_COMMAND_MEM_SIZE_KB = 64;
 
+	bool unique_flusher = false;
 	BinaryMutex mutex;
 	LocalVector<uint8_t> command_mem;
 	ConditionVariable sync_cond_var;
@@ -176,9 +177,17 @@ class CommandQueueMT {
 			// invalidating the pointer due to reallocs.
 			memcpy(cmd_backup, (char *)cmd, size);
 
-			uint32_t allowance_id = WorkerThreadPool::thread_enter_unlock_allowance_zone(lock);
-			((CommandBase *)cmd_backup)->call();
-			WorkerThreadPool::thread_exit_unlock_allowance_zone(allowance_id);
+			if (unique_flusher) {
+				// A single thread will pump; the lock is only needed for the command queue itself.
+				lock.temp_unlock();
+				((CommandBase *)cmd_backup)->call();
+				lock.temp_relock();
+			} else {
+				// At least we can unlock during WTP operations.
+				uint32_t allowance_id = WorkerThreadPool::thread_enter_unlock_allowance_zone(lock);
+				((CommandBase *)cmd_backup)->call();
+				WorkerThreadPool::thread_exit_unlock_allowance_zone(allowance_id);
+			}
 
 			// Handle potential realloc due to the command and unlock allowance.
 			cmd = reinterpret_cast<CommandBase *>(&command_mem[flush_read_ptr]);
@@ -266,6 +275,6 @@ class CommandQueueMT {
 		pump_task_id = p_task_id;
 	}
 
-	CommandQueueMT();
+	CommandQueueMT(bool p_unique_flusher = false);
 	~CommandQueueMT();
 };

modules/betsy/image_compress_betsy.h

@@ -103,7 +103,7 @@ Error _betsy_compress_s3tc(Image *r_img, Image::UsedChannels p_channels);
 class BetsyCompressor : public Object {
 	GDSOFTCLASS(BetsyCompressor, Object);
 
-	mutable CommandQueueMT command_queue;
+	mutable CommandQueueMT command_queue = CommandQueueMT(true);
 	bool exit = false;
 	WorkerThreadPool::TaskID task_id = WorkerThreadPool::INVALID_TASK_ID;
 

servers/rendering/rendering_server_default.h

@@ -74,7 +74,7 @@ class RenderingServerDefault : public RenderingServer {
 	uint64_t print_frame_profile_ticks_from = 0;
 	uint32_t print_frame_profile_frame_count = 0;
 
-	mutable CommandQueueMT command_queue;
+	mutable CommandQueueMT command_queue = CommandQueueMT(true);
 
 	Thread::ID server_thread = Thread::MAIN_ID;
 	WorkerThreadPool::TaskID server_task_id = WorkerThreadPool::INVALID_TASK_ID;