rsx: Implement Atomic Vertex Upload (via setting)

RPCS3 · Sep 1, 2022 · ed60ccd · ed60ccd
1 parent 2cd7238
commit ed60ccd
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 2 deletions.
diff --git a/rpcs3/Emu/Cell/SPURecompiler.cpp b/rpcs3/Emu/Cell/SPURecompiler.cpp
@@ -6563,7 +6563,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
 					break;
 				}
 
-				if (u64 cmdh = ci->getZExtValue() & ~(MFC_BARRIER_MASK | MFC_FENCE_MASK | MFC_RESULT_MASK); g_cfg.core.rsx_fifo_accuracy || !g_use_rtm)
+				if (u64 cmdh = ci->getZExtValue() & ~(MFC_BARRIER_MASK | MFC_FENCE_MASK | MFC_RESULT_MASK); g_cfg.core.rsx_fifo_accuracy || g_cfg.core.rsx_atomic_vertex_upload || !g_use_rtm)
 				{
 					// TODO: don't require TSX (current implementation is TSX-only)
 					if (cmdh == MFC_PUT_CMD || cmdh == MFC_SNDSIG_CMD)

diff --git a/rpcs3/Emu/Cell/SPUThread.cpp b/rpcs3/Emu/Cell/SPUThread.cpp
@@ -2063,7 +2063,7 @@ void spu_thread::do_dma_transfer(spu_thread* _this, const spu_mfc_cmd& args, u8*
 		src = zero_buf;
 	}
 
-	rsx::reservation_lock<false, 1> rsx_lock(eal, args.size, !is_get && g_cfg.core.rsx_fifo_accuracy && !g_cfg.core.spu_accurate_dma);
+	rsx::reservation_lock<false, 1> rsx_lock(eal, args.size, !is_get && (g_cfg.core.rsx_fifo_accuracy || g_cfg.core.rsx_atomic_vertex_upload) && !g_cfg.core.spu_accurate_dma);
 
 	if ((!g_use_rtm && !is_get) || g_cfg.core.spu_accurate_dma)  [[unlikely]]
 	{

diff --git a/rpcs3/Emu/RSX/RSXOffload.cpp b/rpcs3/Emu/RSX/RSXOffload.cpp
@@ -1,5 +1,6 @@
 #include "stdafx.h"
 
+#include "Emu/Memory/vm.h"
 #include "Common/BufferUtils.h"
 #include "RSXOffload.h"
 #include "RSXThread.h"
@@ -44,6 +45,8 @@ namespace rsx
 					{
 					case raw_copy:
 					{
+						const u32 vm_addr = vm::try_get_addr(job.src).first;
+						rsx::reservation_lock<true, 1> rsx_lock(vm_addr, job.length, g_cfg.core.rsx_atomic_vertex_upload && vm_addr);
 						std::memcpy(job.dst, job.src, job.length);
 						break;
 					}
@@ -108,6 +111,8 @@ namespace rsx
 	{
 		if (length <= max_immediate_transfer_size || !g_cfg.video.multithreaded_rsx)
 		{
+			const u32 vm_addr = vm::try_get_addr(src).first;
+			rsx::reservation_lock<true, 1> rsx_lock(vm_addr, length, g_cfg.core.rsx_atomic_vertex_upload && vm_addr);
 			std::memcpy(dst, src, length);
 		}
 		else

diff --git a/rpcs3/Emu/system_config.h b/rpcs3/Emu/system_config.h
@@ -45,6 +45,7 @@ struct cfg_root : cfg::node
 		cfg::_bool spu_accurate_reservations{ this, "Accurate SPU Reservations", true };
 		cfg::_bool accurate_cache_line_stores{ this, "Accurate Cache Line Stores", false };
 		cfg::_bool rsx_accurate_res_access{this, "Accurate RSX reservation access", false, true};
+		cfg::_bool rsx_atomic_vertex_upload{this, "Atomic RSX Vertex Upload", false };
 
 		struct fifo_setting : public cfg::_enum<rsx_fifo_mode>
 		{