From 64e77a60bfef4e6ed3a9650b198f98f8fd6976ea Mon Sep 17 00:00:00 2001
From: pineappleEA <pineaea@gmail.com>
Date: Tue, 29 Mar 2022 07:26:45 +0200
Subject: [PATCH] early-access version 2617

---
 README.md                                       |  2 +-
 src/common/atomic_ops.h                         | 17 +++++++++++++++++
 src/common/x64/native_clock.cpp                 |  4 ++--
 src/video_core/gpu_thread.cpp                   |  4 ++++
 .../renderer_opengl/gl_rasterizer.cpp           |  6 +++---
 5 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/README.md b/README.md
index dd6590bbd..9b87c976b 100755
--- a/README.md
+++ b/README.md
@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 2616.
+This is the source code for early-access 2617.
 
 ## Legal Notice
 
diff --git a/src/common/atomic_ops.h b/src/common/atomic_ops.h
index b94d73c7a..b963e7b99 100755
--- a/src/common/atomic_ops.h
+++ b/src/common/atomic_ops.h
@@ -46,6 +46,13 @@ namespace Common {
                                           reinterpret_cast<__int64*>(expected.data())) != 0;
 }
 
+[[nodiscard]] inline u128 AtomicLoad128(volatile u64* pointer) {
+    u128 result{};
+    _InterlockedCompareExchange128(reinterpret_cast<volatile __int64*>(pointer), result[1],
+                                   result[0], reinterpret_cast<__int64*>(result.data()));
+    return result;
+}
+
 #else
 
 [[nodiscard]] inline bool AtomicCompareAndSwap(volatile u8* pointer, u8 value, u8 expected) {
@@ -72,6 +79,16 @@ namespace Common {
     return __sync_bool_compare_and_swap((unsigned __int128*)pointer, expected_a, value_a);
 }
 
+[[nodiscard]] inline u128 AtomicLoad128(volatile u64* pointer) {
+    unsigned __int128 zeros_a = 0;
+    unsigned __int128 result_a =
+        __sync_val_compare_and_swap((unsigned __int128*)pointer, zeros_a, zeros_a);
+
+    u128 result;
+    std::memcpy(result.data(), &result_a, sizeof(u128));
+    return result;
+}
+
 #endif
 
 } // namespace Common
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
index 347e41efc..2a2664e5d 100755
--- a/src/common/x64/native_clock.cpp
+++ b/src/common/x64/native_clock.cpp
@@ -56,7 +56,7 @@ u64 NativeClock::GetRTSC() {
     TimePoint new_time_point{};
     TimePoint current_time_point{};
     do {
-        current_time_point.pack = time_point.pack;
+        current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
         _mm_mfence();
         const u64 current_measure = __rdtsc();
         u64 diff = current_measure - current_time_point.inner.last_measure;
@@ -76,7 +76,7 @@ void NativeClock::Pause(bool is_paused) {
         TimePoint current_time_point{};
         TimePoint new_time_point{};
         do {
-            current_time_point.pack = time_point.pack;
+            current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
             new_time_point.pack = current_time_point.pack;
             _mm_mfence();
             new_time_point.inner.last_measure = __rdtsc();
diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 9547f277a..78d391cd9 100755
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -90,6 +90,10 @@ void ThreadManager::FlushRegion(VAddr addr, u64 size) {
         return;
     }
     if (!Settings::IsGPULevelExtreme()) {
+        // Push a command and block here before proceeding, addresses a synchronization
+        // bug causing an SVC break in Kirby and the Forgotten Land
+        // GPUTickCommand is essentially a no-op if we don't RequestFlush()
+        PushCommand(GPUTickCommand(), true);
         return;
     }
     auto& gpu = system.GPU();
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 4d632d211..e6f9ece8b 100755
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -15,8 +15,9 @@
 #include "common/logging/log.h"
 #include "common/math_util.h"
 #include "common/microprofile.h"
+#include "common/scope_exit.h"
 #include "common/settings.h"
-#include "core/memory.h"
+
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
@@ -210,6 +211,7 @@ void RasterizerOpenGL::Clear() {
 void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
     MICROPROFILE_SCOPE(OpenGL_Drawing);
 
+    SCOPE_EXIT({ gpu.TickWork(); });
     query_cache.UpdateCounters();
 
     GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()};
@@ -265,8 +267,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
 
     ++num_queued_commands;
     has_written_global_memory |= pipeline->WritesGlobalMemory();
-
-    gpu.TickWork();
 }
 
 void RasterizerOpenGL::DispatchCompute() {