5 files changed, 24 insertions, 12 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index de971041f..9e6b87960 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -596,7 +596,7 @@ void BufferCache<P>::PopAsyncFlushes() {
             runtime.CopyBuffer(download_staging.buffer, slot_buffers[buffer_id], copies);
         }
         runtime.Finish();
-        for (const auto [copy, buffer_id] : downloads) {
+        for (const auto& [copy, buffer_id] : downloads) {
             const Buffer& buffer = slot_buffers[buffer_id];
             const VAddr cpu_addr = buffer.CpuAddr() + copy.src_offset;
             // Undo the modified offset
@@ -606,7 +606,7 @@ void BufferCache<P>::PopAsyncFlushes() {
         }
     } else {
         const std::span<u8> immediate_buffer = ImmediateBuffer(largest_copy);
-        for (const auto [copy, buffer_id] : downloads) {
+        for (const auto& [copy, buffer_id] : downloads) {
             Buffer& buffer = slot_buffers[buffer_id];
             buffer.ImmediateDownload(copy.src_offset, immediate_buffer.subspan(0, copy.size));
             const VAddr cpu_addr = buffer.CpuAddr() + copy.src_offset;
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 37f7b24e1..35cc561be 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -104,7 +104,13 @@ void GPU::WaitFence(u32 syncpoint_id, u32 value) {
     }
     MICROPROFILE_SCOPE(GPU_wait);
     std::unique_lock lock{sync_mutex};
-    sync_cv.wait(lock, [=, this] { return syncpoints.at(syncpoint_id).load() >= value; });
+    sync_cv.wait(lock, [=, this] {
+        if (shutting_down.load(std::memory_order_relaxed)) {
+            // We're shutting down, ensure no threads continue to wait for the next syncpoint
+            return true;
+        }
+        return syncpoints.at(syncpoint_id).load() >= value;
+    });
 }
 
 void GPU::IncrementSyncPoint(const u32 syncpoint_id) {
@@ -523,6 +529,10 @@ void GPU::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const {
 }
 
 void GPU::ShutDown() {
+    // Signal that threads should no longer block on syncpoint fences
+    shutting_down.store(true, std::memory_order_relaxed);
+    sync_cv.notify_all();
+
     gpu_thread.ShutDown();
 }
 
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 29a867863..a8e98e51b 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -389,6 +389,8 @@ private:
     std::unique_ptr<Engines::KeplerMemory> kepler_memory;
     /// Shader build notifier
     std::unique_ptr<VideoCore::ShaderNotify> shader_notify;
+    /// When true, we are about to shut down emulation session, so terminate outstanding tasks
+    std::atomic_bool shutting_down{};
 
     std::array<std::atomic<u32>, Service::Nvidia::MaxSyncPoints> syncpoints{};
 
diff --git a/src/video_core/rasterizer_accelerated.cpp b/src/video_core/rasterizer_accelerated.cpp
index 62d84c0f8..6decd2546 100644
--- a/src/video_core/rasterizer_accelerated.cpp
+++ b/src/video_core/rasterizer_accelerated.cpp
@@ -18,10 +18,10 @@ RasterizerAccelerated::~RasterizerAccelerated() = default;
 void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {
     const auto page_end = Common::DivCeil(addr + size, Core::Memory::PAGE_SIZE);
     for (auto page = addr >> Core::Memory::PAGE_BITS; page != page_end; ++page) {
-        auto& count = cached_pages.at(page >> 3).Count(page);
+        auto& count = cached_pages.at(page >> 2).Count(page);
 
         if (delta > 0) {
-            ASSERT_MSG(count < UINT8_MAX, "Count may overflow!");
+            ASSERT_MSG(count < UINT16_MAX, "Count may overflow!");
         } else if (delta < 0) {
             ASSERT_MSG(count > 0, "Count may underflow!");
         } else {
@@ -29,7 +29,7 @@ void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int del
         }
 
         // Adds or subtracts 1, as count is a unsigned 8-bit value
-        count += static_cast<u8>(delta);
+        count += static_cast<u16>(delta);
 
         // Assume delta is either -1 or 1
         if (count == 0) {
diff --git a/src/video_core/rasterizer_accelerated.h b/src/video_core/rasterizer_accelerated.h
index 9227a4adc..ea879bfdd 100644
--- a/src/video_core/rasterizer_accelerated.h
+++ b/src/video_core/rasterizer_accelerated.h
@@ -29,20 +29,20 @@ private:
     public:
         CacheEntry() = default;
 
-        std::atomic_uint8_t& Count(std::size_t page) {
-            return values[page & 7];
+        std::atomic_uint16_t& Count(std::size_t page) {
+            return values[page & 3];
         }
 
-        const std::atomic_uint8_t& Count(std::size_t page) const {
-            return values[page & 7];
+        const std::atomic_uint16_t& Count(std::size_t page) const {
+            return values[page & 3];
         }
 
     private:
-        std::array<std::atomic_uint8_t, 8> values{};
+        std::array<std::atomic_uint16_t, 4> values{};
     };
     static_assert(sizeof(CacheEntry) == 8, "CacheEntry should be 8 bytes!");
 
-    std::array<CacheEntry, 0x800000> cached_pages;
+    std::array<CacheEntry, 0x1000000> cached_pages;
     Core::Memory::Memory& cpu_memory;
 };