Add new accuracy control for PUTLLUC accuracy setting (non-TSX)

With the option enabled GET commands are blocked until the current PUTLLC/PUTLLUC executer on that address finishes

Additional improvements:
- Minor race fix of sys_ppu_thread_exit (wait until the writer finishes)
- Max number of ppu threads bumped to 8

rpcs3/Emu/Cell/SPUThread.cpp

@@ -1315,7 +1315,12 @@ void spu_thread::do_dma_transfer(const spu_mfc_cmd& args)
 	u8* dst = vm::_ptr<u8>(eal);
 	u8* src = vm::_ptr<u8>(offset + lsa);
 
-	if (UNLIKELY(!is_get && !g_use_rtm))
+	if (is_get)
+	{
+		std::swap(dst, src);
+	}
+
+	if (UNLIKELY(!g_use_rtm && (!is_get || g_cfg.core.spu_accurate_putlluc)))
 	{
 		switch (u32 size = args.size)
 		{
@@ -1323,28 +1328,28 @@ void spu_thread::do_dma_transfer(const spu_mfc_cmd& args)
 		{
 			auto& res = vm::reservation_lock(eal, 1);
 			*reinterpret_cast<u8*>(dst) = *reinterpret_cast<const u8*>(src);
-			res.release(res.load() + 127);
+			res.release(res.load() - 1);
 			break;
 		}
 		case 2:
 		{
 			auto& res = vm::reservation_lock(eal, 2);
 			*reinterpret_cast<u16*>(dst) = *reinterpret_cast<const u16*>(src);
-			res.release(res.load() + 127);
+			res.release(res.load() - 1);
 			break;
 		}
 		case 4:
 		{
 			auto& res = vm::reservation_lock(eal, 4);
 			*reinterpret_cast<u32*>(dst) = *reinterpret_cast<const u32*>(src);
-			res.release(res.load() + 127);
+			res.release(res.load() - 1);
 			break;
 		}
 		case 8:
 		{
 			auto& res = vm::reservation_lock(eal, 8);
 			*reinterpret_cast<u64*>(dst) = *reinterpret_cast<const u64*>(src);
-			res.release(res.load() + 127);
+			res.release(res.load() - 1);
 			break;
 		}
 		default:
@@ -1363,7 +1368,7 @@ void spu_thread::do_dma_transfer(const spu_mfc_cmd& args)
 					size -= 16;
 				}
 
-				res.release(res.load() + 127);
+				res.release(res.load() - 1);
 				break;
 			}
 
@@ -1395,11 +1400,6 @@ void spu_thread::do_dma_transfer(const spu_mfc_cmd& args)
 		return;
 	}
 
-	if (is_get)
-	{
-		std::swap(dst, src);
-	}
-
 	switch (u32 size = args.size)
 	{
 	case 1:

rpcs3/Emu/Cell/lv2/sys_ppu_thread.cpp

@@ -15,6 +15,9 @@ void _sys_ppu_thread_exit(ppu_thread& ppu, u64 errorcode)
 {
 	vm::temporary_unlock(ppu);
 
+	// Need to wait until the current writer finish
+	if (ppu.state & cpu_flag::memory) vm::g_mutex.lock_unlock();
+
 	sys_ppu_thread.trace("_sys_ppu_thread_exit(errorcode=0x%llx)", errorcode);
 
 	ppu.state += cpu_flag::exit;

rpcs3/Emu/Memory/vm.cpp

@@ -62,7 +62,7 @@ namespace vm
 	// Memory mutex acknowledgement
 	thread_local atomic_t<cpu_thread*>* g_tls_locked = nullptr;
 
-	// Currently locked address
+	// Currently locked cache line
 	atomic_t<u32> g_addr_lock = 0;
 
 	// Memory mutex: passive locks
@@ -71,13 +71,15 @@ namespace vm
 
 	static void _register_lock(cpu_thread* _cpu)
 	{
-		for (u32 i = 0;; i = (i + 1) % g_locks.size())
+		for (u32 i = 0, max = g_cfg.core.ppu_threads;;)
 		{
 			if (!g_locks[i] && g_locks[i].compare_and_swap_test(nullptr, _cpu))
 			{
 				g_tls_locked = g_locks.data() + i;
 				return;
 			}
+
+			if (++i == max) i = 0;
 		}
 	}
 
@@ -165,7 +167,7 @@ namespace vm
 
 	void cleanup_unlock(cpu_thread& cpu) noexcept
 	{
-		for (u32 i = 0; i < g_locks.size(); i++)
+		for (u32 i = 0, max = g_cfg.core.ppu_threads; i < max; i++)
 		{
 			if (g_locks[i] == &cpu)
 			{
@@ -247,9 +249,9 @@ namespace vm
 
 		if (addr)
 		{
-			for (auto& lock : g_locks)
+			for (auto lock = g_locks.cbegin(), end = lock + g_cfg.core.ppu_threads; lock != end; lock++)
 			{
-				if (cpu_thread* ptr = lock)
+				if (cpu_thread* ptr = *lock)
 				{
 					ptr->state.test_and_set(cpu_flag::memory);
 				}
@@ -279,15 +281,10 @@ namespace vm
 				}
 			}
 
-			for (auto& lock : g_locks)
+			for (auto lock = g_locks.cbegin(), end = lock + g_cfg.core.ppu_threads; lock != end; lock++)
 			{
-				while (cpu_thread* ptr = lock)
+				while (*lock)
 				{
-					if (ptr->is_stopped())
-					{
-						break;
-					}
-
 					_mm_pause();
 				}
 			}

rpcs3/Emu/System.h

@@ -388,7 +388,7 @@ struct cfg_root : cfg::node
 		node_core(cfg::node* _this) : cfg::node(_this, "Core") {}
 
 		cfg::_enum<ppu_decoder_type> ppu_decoder{this, "PPU Decoder", ppu_decoder_type::llvm};
-		cfg::_int<1, 4> ppu_threads{this, "PPU Threads", 2}; // Amount of PPU threads running simultaneously (must be 2)
+		cfg::_int<1, 8> ppu_threads{this, "PPU Threads", 2}; // Amount of PPU threads running simultaneously (must be 2)
 		cfg::_bool ppu_debug{this, "PPU Debug"};
 		cfg::_bool llvm_logs{this, "Save LLVM logs"};
 		cfg::string llvm_cpu{this, "Use LLVM CPU"};