Initial support for Apple GPUs

rpcs3/Emu/RSX/VK/VKCompute.cpp

@@ -79,8 +79,8 @@ namespace vk
 			case vk::driver_vendor::NVIDIA:
 				// Warps are multiples of 32. Increasing kernel depth seems to hurt performance (Nier, Big Duck sample)
 				unroll_loops = true;
-				optimal_group_size = 32;
 				optimal_kernel_size = 1;
+				optimal_group_size = 32;
 				break;
 			case vk::driver_vendor::AMD:
 			case vk::driver_vendor::RADV:
@@ -89,6 +89,11 @@ namespace vk
 				optimal_kernel_size = 1;
 				optimal_group_size = 64;
 				break;
+			case vk::driver_vendor::MVK:
+				unroll_loops = true;
+				optimal_kernel_size = 1;
+				optimal_group_size = 256;
+				break;
 			}
 
 			const auto& gpu = vk::g_render_device->gpu();

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -603,6 +603,11 @@ VKGSRender::VKGSRender() : GSRender()
 			}
 			break;
 #endif
+		case vk::driver_vendor::MVK:
+			// Async compute crashes immediately on Apple GPUs
+			rsx_log.error("Apple GPUs are incompatible with the current implementation of asynchronous texture decoding.");
+			backend_config.supports_asynchronous_compute = false;
+			break;
 		default: break;
 		}
 

rpcs3/Emu/RSX/VK/VKHelpers.cpp

@@ -124,6 +124,9 @@ namespace vk
 		case driver_vendor::ANV:
 			// INTEL vulkan drivers are mostly OK, workarounds are applied when creating the device
 			break;
+		case driver_vendor::MVK:
+			// Apple GPUs / moltenVK need more testing
+			break;
 		default:
 			rsx_log.warning("Unsupported device: %s", gpu_name);
 		}

rpcs3/Emu/RSX/VK/VKRenderTargets.h

@@ -115,6 +115,7 @@ namespace vk
 				[[ fallthrough ]];
 			case driver_vendor::NVIDIA:
 			case driver_vendor::INTEL:
+			case driver_vendor::MVK:
 				break;
 			}
 
@@ -453,4 +454,4 @@ namespace vk
 		void free_invalidated(vk::command_buffer& cmd, rsx::problem_severity memory_pressure);
 	};
 }
-//h
+//h

rpcs3/Emu/RSX/VK/vkutils/chip_class.h

@@ -20,7 +20,8 @@ namespace vk
 		NV_pascal,
 		NV_volta,
 		NV_turing,
-		NV_ampere
+		NV_ampere,
+		MVK_apple
 	};
 
 	enum class driver_vendor
@@ -30,7 +31,8 @@ namespace vk
 		NVIDIA,
 		RADV,
 		INTEL,
-		ANV
+		ANV,
+		MVK
 	};
 
 	driver_vendor get_driver_vendor();

rpcs3/Emu/RSX/VK/vkutils/device.cpp

@@ -183,9 +183,16 @@ namespace vk
 
 	driver_vendor physical_device::get_driver_vendor() const
 	{
+#ifdef __APPLE__
+		// moltenVK currently returns DRIVER_ID_MOLTENVK (0).
+		// For now, assume the vendor is moltenVK on Apple devices.
+		return driver_vendor::MVK;
+#endif
+
 		if (!driver_properties.driverID)
 		{
 			const auto gpu_name = get_name();
+
 			if (gpu_name.find("Radeon") != umax)
 			{
 				return driver_vendor::AMD;

rpcs3/Emu/RSX/VK/vkutils/swapchain.hpp

@@ -522,6 +522,7 @@ namespace vk
 				case driver_vendor::AMD:
 				case driver_vendor::INTEL:
 				case driver_vendor::RADV:
+				case driver_vendor::MVK:
 					break;
 				case driver_vendor::ANV:
 				case driver_vendor::NVIDIA:

rpcs3/Emu/system_config.h

@@ -73,11 +73,10 @@ struct cfg_root : cfg::node
 		cfg::uint64 tx_limit2_ns{this, "TSX Transaction Second Limit", 2000}; // In nanoseconds
 
 		cfg::_int<10, 3000> clocks_scale{ this, "Clocks scale", 100 }; // Changing this from 100 (percentage) may affect game speed in unexpected ways
-		cfg::_enum<sleep_timers_accuracy_level> sleep_timers_accuracy{ this, "Sleep Timers Accuracy",
+#if defined (__linux__) || defined (__APPLE__)
-#ifdef __linux__
+		cfg::_enum<sleep_timers_accuracy_level> sleep_timers_accuracy{ this, "Sleep Timers Accuracy", sleep_timers_accuracy_level::_as_host, true };
-			sleep_timers_accuracy_level::_as_host, true };
 #else
-			sleep_timers_accuracy_level::_usleep, true };
+		cfg::_enum<sleep_timers_accuracy_level> sleep_timers_accuracy{ this, "Sleep Timers Accuracy", sleep_timers_accuracy_level::_usleep, true };
 #endif
 
 		cfg::uint64 perf_report_threshold{this, "Performance Report Threshold", 500, true}; // In µs, 0.5ms = default, 0 = everything
@@ -134,7 +133,11 @@ struct cfg_root : cfg::node
 		cfg::_bool disable_vulkan_mem_allocator{ this, "Disable Vulkan Memory Allocator", false };
 		cfg::_bool full_rgb_range_output{ this, "Use full RGB output range", true, true }; // Video out dynamic range
 		cfg::_bool strict_texture_flushing{ this, "Strict Texture Flushing", false };
+#ifdef __APPLE__
+		cfg::_bool disable_native_float16{ this, "Disable native float16 support", true };
+#else
 		cfg::_bool disable_native_float16{ this, "Disable native float16 support", false };
+#endif
 		cfg::_bool multithreaded_rsx{ this, "Multithreaded RSX", false };
 		cfg::_bool relaxed_zcull_sync{ this, "Relaxed ZCULL Sync", false };
 		cfg::_bool enable_3d{ this, "Enable 3D", false };