qt: warn on inoperable keys

vk_pipeline_cache: Skip compute pipelines on Intel proprietary drivers

src/core/hle/service/audio/audout_u.cpp

@@ -49,12 +49,6 @@ public:
         };
         // clang-format on
         RegisterHandlers(functions);
-
-        if (impl->GetSystem()
-                .Initialize(device_name, in_params, handle, applet_resource_user_id)
-                .IsError()) {
-            LOG_ERROR(Service_Audio, "Failed to initialize the AudioOut System!");
-        }
     }
 
     ~IAudioOut() override {
@@ -287,6 +281,14 @@ void AudOutU::OpenAudioOut(HLERequestContext& ctx) {
 
     auto audio_out = std::make_shared<IAudioOut>(system, *impl, new_session_id, device_name,
                                                  in_params, handle, applet_resource_user_id);
+    result = audio_out->GetImpl()->GetSystem().Initialize(device_name, in_params, handle,
+                                                          applet_resource_user_id);
+    if (result.IsError()) {
+        LOG_ERROR(Service_Audio, "Failed to initialize the AudioOut System!");
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(result);
+        return;
+    }
 
     impl->sessions[new_session_id] = audio_out->GetImpl();
     impl->applet_resource_user_ids[new_session_id] = applet_resource_user_id;

src/core/memory.cpp

@@ -462,7 +462,7 @@ struct Memory::Impl {
         }
 
         if (Settings::IsFastmemEnabled()) {
-            const bool is_read_enable = !Settings::IsGPULevelExtreme() || !cached;
+            const bool is_read_enable = Settings::IsGPULevelHigh() || !cached;
             system.DeviceMemory().buffer.Protect(vaddr, size, is_read_enable, !cached);
         }
 

src/video_core/engines/maxwell_3d.cpp

@@ -4,6 +4,7 @@
 #include <cstring>
 #include <optional>
 #include "common/assert.h"
+#include "common/bit_util.h"
 #include "common/scope_exit.h"
 #include "common/settings.h"
 #include "core/core.h"
@@ -222,6 +223,9 @@ void Maxwell3D::ProcessMacro(u32 method, const u32* base_start, u32 amount, bool
 }
 
 void Maxwell3D::RefreshParametersImpl() {
+    if (!Settings::IsGPULevelHigh()) {
+        return;
+    }
     size_t current_index = 0;
     for (auto& segment : macro_segments) {
         if (segment.first == 0) {
@@ -259,12 +263,13 @@ u32 Maxwell3D::GetMaxCurrentVertices() {
 size_t Maxwell3D::EstimateIndexBufferSize() {
     GPUVAddr start_address = regs.index_buffer.StartAddress();
     GPUVAddr end_address = regs.index_buffer.EndAddress();
-    static constexpr std::array<size_t, 4> max_sizes = {
-        std::numeric_limits<u8>::max(), std::numeric_limits<u16>::max(),
-        std::numeric_limits<u32>::max(), std::numeric_limits<u32>::max()};
+    static constexpr std::array<size_t, 3> max_sizes = {std::numeric_limits<u8>::max(),
+                                                        std::numeric_limits<u16>::max(),
+                                                        std::numeric_limits<u32>::max()};
     const size_t byte_size = regs.index_buffer.FormatSizeInBytes();
+    const size_t log2_byte_size = Common::Log2Ceil64(byte_size);
     return std::min<size_t>(
-        memory_manager.GetMemoryLayoutSize(start_address, byte_size * max_sizes[byte_size]) /
+        memory_manager.GetMemoryLayoutSize(start_address, byte_size * max_sizes[log2_byte_size]) /
             byte_size,
         static_cast<size_t>(end_address - start_address));
 }

src/video_core/fence_manager.h

@@ -4,20 +4,13 @@
 #pragma once
 
 #include <algorithm>
-#include <condition_variable>
 #include <cstring>
 #include <deque>
 #include <functional>
 #include <memory>
-#include <mutex>
-#include <thread>
 #include <queue>
 
 #include "common/common_types.h"
-#include "common/microprofile.h"
-#include "common/scope_exit.h"
-#include "common/settings.h"
-#include "common/thread.h"
 #include "video_core/delayed_destruction_ring.h"
 #include "video_core/gpu.h"
 #include "video_core/host1x/host1x.h"
@@ -30,26 +23,15 @@ class FenceBase {
 public:
     explicit FenceBase(bool is_stubbed_) : is_stubbed{is_stubbed_} {}
 
-    bool IsStubbed() const {
-        return is_stubbed;
-    }
-
 protected:
     bool is_stubbed;
 };
 
-template <typename Traits>
+template <typename TFence, typename TTextureCache, typename TTBufferCache, typename TQueryCache>
 class FenceManager {
-    using TFence = typename Traits::FenceType;
-    using TTextureCache = typename Traits::TextureCacheType;
-    using TBufferCache = typename Traits::BufferCacheType;
-    using TQueryCache = typename Traits::QueryCacheType;
-    static constexpr bool can_async_check = Traits::HAS_ASYNC_CHECK;
-
 public:
     /// Notify the fence manager about a new frame
     void TickFrame() {
-        std::unique_lock lock(ring_guard);
         delayed_destruction_ring.Tick();
     }
 
@@ -64,33 +46,17 @@ public:
     }
 
     void SignalFence(std::function<void()>&& func) {
-        rasterizer.InvalidateGPUCache();
-        bool delay_fence = Settings::IsGPULevelHigh();
-        if constexpr (!can_async_check) {
-            TryReleasePendingFences<false>();
-        }
+        TryReleasePendingFences();
         const bool should_flush = ShouldFlush();
         CommitAsyncFlushes();
+        uncommitted_operations.emplace_back(std::move(func));
+        CommitOperations();
         TFence new_fence = CreateFence(!should_flush);
-        if constexpr (can_async_check) {
-            guard.lock();
-        }
-        if (delay_fence) {
-            uncommitted_operations.emplace_back(std::move(func));
-        }
-        pending_operations.emplace_back(std::move(uncommitted_operations));
+        fences.push(new_fence);
         QueueFence(new_fence);
-        if (!delay_fence) {
-            func();
-        }
-        fences.push(std::move(new_fence));
         if (should_flush) {
             rasterizer.FlushCommands();
         }
-        if constexpr (can_async_check) {
-            guard.unlock();
-            cv.notify_all();
-        }
     }
 
     void SignalSyncPoint(u32 value) {
@@ -100,30 +66,29 @@ public:
     }
 
     void WaitPendingFences() {
-        if constexpr (!can_async_check) {
-            TryReleasePendingFences<true>();
+        while (!fences.empty()) {
+            TFence& current_fence = fences.front();
+            if (ShouldWait()) {
+                WaitFence(current_fence);
+            }
+            PopAsyncFlushes();
+            auto operations = std::move(pending_operations.front());
+            pending_operations.pop_front();
+            for (auto& operation : operations) {
+                operation();
+            }
+            PopFence();
         }
     }
 
 protected:
     explicit FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                          TTextureCache& texture_cache_, TBufferCache& buffer_cache_,
+                          TTextureCache& texture_cache_, TTBufferCache& buffer_cache_,
                           TQueryCache& query_cache_)
         : rasterizer{rasterizer_}, gpu{gpu_}, syncpoint_manager{gpu.Host1x().GetSyncpointManager()},
-          texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, query_cache{query_cache_} {
-        if constexpr (can_async_check) {
-            fence_thread =
-                std::jthread([this](std::stop_token token) { ReleaseThreadFunc(token); });
-        }
-    }
+          texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, query_cache{query_cache_} {}
 
-    virtual ~FenceManager() {
-        if constexpr (can_async_check) {
-            fence_thread.request_stop();
-            cv.notify_all();
-            fence_thread.join();
-        }
-    }
+    virtual ~FenceManager() = default;
 
     /// Creates a Fence Interface, does not create a backend fence if 'is_stubbed' is
     /// true
@@ -139,20 +104,15 @@ protected:
     Tegra::GPU& gpu;
     Tegra::Host1x::SyncpointManager& syncpoint_manager;
     TTextureCache& texture_cache;
-    TBufferCache& buffer_cache;
+    TTBufferCache& buffer_cache;
     TQueryCache& query_cache;
 
 private:
-    template <bool force_wait>
     void TryReleasePendingFences() {
         while (!fences.empty()) {
             TFence& current_fence = fences.front();
             if (ShouldWait() && !IsFenceSignaled(current_fence)) {
-                if constexpr (force_wait) {
-                    WaitFence(current_fence);
-                } else {
-                    return;
-                }
+                return;
             }
             PopAsyncFlushes();
             auto operations = std::move(pending_operations.front());
@@ -160,49 +120,7 @@ private:
             for (auto& operation : operations) {
                 operation();
             }
-            {
-                std::unique_lock lock(ring_guard);
-                delayed_destruction_ring.Push(std::move(current_fence));
-            }
-            fences.pop();
-        }
-    }
-
-    void ReleaseThreadFunc(std::stop_token stop_token) {
-        std::string name = "GPUFencingThread";
-        MicroProfileOnThreadCreate(name.c_str());
-
-        // Cleanup
-        SCOPE_EXIT({ MicroProfileOnThreadExit(); });
-
-        Common::SetCurrentThreadName(name.c_str());
-        Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
-
-        TFence current_fence;
-        std::deque<std::function<void()>> current_operations;
-        while (!stop_token.stop_requested()) {
-            {
-                std::unique_lock lock(guard);
-                cv.wait(lock, [&] { return stop_token.stop_requested() || !fences.empty(); });
-                if (stop_token.stop_requested()) [[unlikely]] {
-                    return;
-                }
-                current_fence = std::move(fences.front());
-                current_operations = std::move(pending_operations.front());
-                fences.pop();
-                pending_operations.pop_front();
-            }
-            if (!current_fence->IsStubbed()) {
-                WaitFence(current_fence);
-            }
-            PopAsyncFlushes();
-            for (auto& operation : current_operations) {
-                operation();
-            }
-            {
-                std::unique_lock lock(ring_guard);
-                delayed_destruction_ring.Push(std::move(current_fence));
-            }
+            PopFence();
         }
     }
 
@@ -236,16 +154,19 @@ private:
         query_cache.CommitAsyncFlushes();
     }
 
+    void PopFence() {
+        delayed_destruction_ring.Push(std::move(fences.front()));
+        fences.pop();
+    }
+
+    void CommitOperations() {
+        pending_operations.emplace_back(std::move(uncommitted_operations));
+    }
+
     std::queue<TFence> fences;
     std::deque<std::function<void()>> uncommitted_operations;
     std::deque<std::deque<std::function<void()>>> pending_operations;
 
-    std::mutex guard;
-    std::mutex ring_guard;
-    std::condition_variable cv;
-
-    std::jthread fence_thread;
-
     DelayedDestructionRing<TFence, 6> delayed_destruction_ring;
 };
 

src/video_core/memory_manager.cpp

@@ -170,7 +170,6 @@ void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_)
 
 GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size, PTEKind kind,
                             bool is_big_pages) {
-    std::unique_lock<std::mutex> lock(guard);
     if (is_big_pages) [[likely]] {
         return BigPageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size, kind);
     }
@@ -178,7 +177,6 @@ GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size,
 }
 
 GPUVAddr MemoryManager::MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages) {
-    std::unique_lock<std::mutex> lock(guard);
     if (is_big_pages) [[likely]] {
         return BigPageTableOp<EntryType::Reserved>(gpu_addr, 0, size, PTEKind::INVALID);
     }
@@ -189,7 +187,6 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
     if (size == 0) {
         return;
     }
-    std::unique_lock<std::mutex> lock(guard);
     GetSubmappedRangeImpl<false>(gpu_addr, size, page_stash);
 
     for (const auto& [map_addr, map_size] : page_stash) {
@@ -556,7 +553,6 @@ size_t MemoryManager::MaxContinuousRange(GPUVAddr gpu_addr, size_t size) const {
 }
 
 size_t MemoryManager::GetMemoryLayoutSize(GPUVAddr gpu_addr, size_t max_size) const {
-    std::unique_lock<std::mutex> lock(guard);
     return kind_map.GetContinuousSizeFrom(gpu_addr);
 }
 
@@ -749,10 +745,10 @@ void MemoryManager::FlushCaching() {
         return;
     }
     accumulator->Callback([this](GPUVAddr addr, size_t size) {
-        GetSubmappedRangeImpl<false>(addr, size, page_stash2);
+        GetSubmappedRangeImpl<false>(addr, size, page_stash);
     });
-    rasterizer->InnerInvalidation(page_stash2);
-    page_stash2.clear();
+    rasterizer->InnerInvalidation(page_stash);
+    page_stash.clear();
     accumulator->Clear();
 }
 

src/video_core/memory_manager.h

@@ -5,7 +5,6 @@
 
 #include <atomic>
 #include <map>
-#include <mutex>
 #include <optional>
 #include <vector>
 
@@ -216,9 +215,6 @@ private:
 
     std::vector<u64> big_page_continuous;
     std::vector<std::pair<VAddr, std::size_t>> page_stash{};
-    std::vector<std::pair<VAddr, std::size_t>> page_stash2{};
-
-    mutable std::mutex guard;
 
     static constexpr size_t continuous_bits = 64;
 

src/video_core/query_cache.h

@@ -6,7 +6,6 @@
 #include <algorithm>
 #include <array>
 #include <cstring>
-#include <functional>
 #include <iterator>
 #include <list>
 #include <memory>
@@ -18,19 +17,13 @@
 
 #include "common/assert.h"
 #include "common/settings.h"
-#include "core/memory.h"
 #include "video_core/control/channel_state_cache.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
-#include "video_core/texture_cache/slot_vector.h"
 
 namespace VideoCommon {
 
-using AsyncJobId = SlotId;
-
-static constexpr AsyncJobId NULL_ASYNC_JOB_ID{0};
-
 template <class QueryCache, class HostCounter>
 class CounterStreamBase {
 public:
@@ -100,13 +93,9 @@ private:
 template <class QueryCache, class CachedQuery, class CounterStream, class HostCounter>
 class QueryCacheBase : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
 public:
-    explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_,
-                            Core::Memory::Memory& cpu_memory_)
-        : rasterizer{rasterizer_},
-          cpu_memory{cpu_memory_}, streams{{CounterStream{static_cast<QueryCache&>(*this),
-                                                          VideoCore::QueryType::SamplesPassed}}} {
-        (void)slot_async_jobs.insert(); // Null value
-    }
+    explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_)
+        : rasterizer{rasterizer_}, streams{{CounterStream{static_cast<QueryCache&>(*this),
+                                                          VideoCore::QueryType::SamplesPassed}}} {}
 
     void InvalidateRegion(VAddr addr, std::size_t size) {
         std::unique_lock lock{mutex};
@@ -137,15 +126,10 @@ public:
             query = Register(type, *cpu_addr, host_ptr, timestamp.has_value());
         }
 
-        auto result = query->BindCounter(Stream(type).Current(), timestamp);
-        if (result) {
-            auto async_job_id = query->GetAsyncJob();
-            auto& async_job = slot_async_jobs[async_job_id];
-            async_job.collected = true;
-            async_job.value = *result;
-            query->SetAsyncJob(NULL_ASYNC_JOB_ID);
+        query->BindCounter(Stream(type).Current(), timestamp);
+        if (Settings::values.use_asynchronous_gpu_emulation.GetValue()) {
+            AsyncFlushQuery(*cpu_addr);
         }
-        AsyncFlushQuery(query, timestamp, lock);
     }
 
     /// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch.
@@ -189,18 +173,15 @@ public:
     }
 
     void CommitAsyncFlushes() {
-        std::unique_lock lock{mutex};
         committed_flushes.push_back(uncommitted_flushes);
         uncommitted_flushes.reset();
     }
 
     bool HasUncommittedFlushes() const {
-        std::unique_lock lock{mutex};
         return uncommitted_flushes != nullptr;
     }
 
     bool ShouldWaitAsyncFlushes() const {
-        std::unique_lock lock{mutex};
         if (committed_flushes.empty()) {
             return false;
         }
@@ -208,7 +189,6 @@ public:
     }
 
     void PopAsyncFlushes() {
-        std::unique_lock lock{mutex};
         if (committed_flushes.empty()) {
             return;
         }
@@ -217,25 +197,15 @@ public:
             committed_flushes.pop_front();
             return;
         }
-        for (AsyncJobId async_job_id : *flush_list) {
-            AsyncJob& async_job = slot_async_jobs[async_job_id];
-            if (!async_job.collected) {
-                FlushAndRemoveRegion(async_job.query_location, 2, true);
-            }
+        for (VAddr query_address : *flush_list) {
+            FlushAndRemoveRegion(query_address, 4);
         }
         committed_flushes.pop_front();
     }
 
 private:
-    struct AsyncJob {
-        bool collected = false;
-        u64 value = 0;
-        VAddr query_location = 0;
-        std::optional<u64> timestamp{};
-    };
-
     /// Flushes a memory range to guest memory and removes it from the cache.
-    void FlushAndRemoveRegion(VAddr addr, std::size_t size, bool async = false) {
+    void FlushAndRemoveRegion(VAddr addr, std::size_t size) {
         const u64 addr_begin = addr;
         const u64 addr_end = addr_begin + size;
         const auto in_range = [addr_begin, addr_end](const CachedQuery& query) {
@@ -256,16 +226,7 @@ private:
                     continue;
                 }
                 rasterizer.UpdatePagesCachedCount(query.GetCpuAddr(), query.SizeInBytes(), -1);
-                AsyncJobId async_job_id = query.GetAsyncJob();
-                auto flush_result = query.Flush(async);
-                if (async_job_id == NULL_ASYNC_JOB_ID) {
-                    ASSERT_MSG(false, "This should not be reachable at all");
-                    continue;
-                }
-                AsyncJob& async_job = slot_async_jobs[async_job_id];
-                async_job.collected = true;
-                async_job.value = flush_result;
-                query.SetAsyncJob(NULL_ASYNC_JOB_ID);
+                query.Flush();
             }
             std::erase_if(contents, in_range);
         }
@@ -292,60 +253,26 @@ private:
         return found != std::end(contents) ? &*found : nullptr;
     }
 
-    void AsyncFlushQuery(CachedQuery* query, std::optional<u64> timestamp,
-                         std::unique_lock<std::recursive_mutex>& lock) {
-        const AsyncJobId new_async_job_id = slot_async_jobs.insert();
-        {
-            AsyncJob& async_job = slot_async_jobs[new_async_job_id];
-            query->SetAsyncJob(new_async_job_id);
-            async_job.query_location = query->GetCpuAddr();
-            async_job.collected = false;
-
-            if (!uncommitted_flushes) {
-                uncommitted_flushes = std::make_shared<std::vector<AsyncJobId>>();
-            }
-            uncommitted_flushes->push_back(new_async_job_id);
+    void AsyncFlushQuery(VAddr addr) {
+        if (!uncommitted_flushes) {
+            uncommitted_flushes = std::make_shared<std::vector<VAddr>>();
         }
-        lock.unlock();
-        std::function<void()> operation([this, new_async_job_id, timestamp] {
-            std::unique_lock local_lock{mutex};
-            AsyncJob& async_job = slot_async_jobs[new_async_job_id];
-            u64 value = async_job.value;
-            VAddr address = async_job.query_location;
-            slot_async_jobs.erase(new_async_job_id);
-            local_lock.unlock();
-            if (timestamp) {
-                u64 timestamp_value = *timestamp;
-                cpu_memory.WriteBlockUnsafe(address + sizeof(u64), &timestamp_value, sizeof(u64));
-                cpu_memory.WriteBlockUnsafe(address, &value, sizeof(u64));
-                rasterizer.InvalidateRegion(address, sizeof(u64) * 2,
-                                            VideoCommon::CacheType::NoQueryCache);
-            } else {
-                u32 small_value = static_cast<u32>(value);
-                cpu_memory.WriteBlockUnsafe(address, &small_value, sizeof(u32));
-                rasterizer.InvalidateRegion(address, sizeof(u32),
-                                            VideoCommon::CacheType::NoQueryCache);
-            }
-        });
-        rasterizer.SyncOperation(std::move(operation));
+        uncommitted_flushes->push_back(addr);
     }
 
     static constexpr std::uintptr_t YUZU_PAGESIZE = 4096;
     static constexpr unsigned YUZU_PAGEBITS = 12;
 
-    SlotVector<AsyncJob> slot_async_jobs;
-
     VideoCore::RasterizerInterface& rasterizer;
-    Core::Memory::Memory& cpu_memory;
 
-    mutable std::recursive_mutex mutex;
+    std::recursive_mutex mutex;
 
     std::unordered_map<u64, std::vector<CachedQuery>> cached_queries;
 
     std::array<CounterStream, VideoCore::NumQueryTypes> streams;
 
-    std::shared_ptr<std::vector<AsyncJobId>> uncommitted_flushes{};
-    std::list<std::shared_ptr<std::vector<AsyncJobId>>> committed_flushes;
+    std::shared_ptr<std::vector<VAddr>> uncommitted_flushes{};
+    std::list<std::shared_ptr<std::vector<VAddr>>> committed_flushes;
 };
 
 template <class QueryCache, class HostCounter>
@@ -364,12 +291,12 @@ public:
     virtual ~HostCounterBase() = default;
 
     /// Returns the current value of the query.
-    u64 Query(bool async = false) {
+    u64 Query() {
         if (result) {
             return *result;
         }
 
-        u64 value = BlockingQuery(async) + base_result;
+        u64 value = BlockingQuery() + base_result;
         if (dependency) {
             value += dependency->Query();
             dependency = nullptr;
@@ -390,7 +317,7 @@ public:
 
 protected:
     /// Returns the value of query from the backend API blocking as needed.
-    virtual u64 BlockingQuery(bool async = false) const = 0;
+    virtual u64 BlockingQuery() const = 0;
 
 private:
     std::shared_ptr<HostCounter> dependency; ///< Counter to add to this value.
@@ -413,33 +340,26 @@ public:
     CachedQueryBase& operator=(const CachedQueryBase&) = delete;
 
     /// Flushes the query to guest memory.
-    virtual u64 Flush(bool async = false) {
+    virtual void Flush() {
         // When counter is nullptr it means that it's just been reset. We are supposed to write a
         // zero in these cases.
-        const u64 value = counter ? counter->Query(async) : 0;
-        if (async) {
-            return value;
-        }
+        const u64 value = counter ? counter->Query() : 0;
         std::memcpy(host_ptr, &value, sizeof(u64));
 
         if (timestamp) {
             std::memcpy(host_ptr + TIMESTAMP_OFFSET, &*timestamp, sizeof(u64));
         }
-        return value;
     }
 
     /// Binds a counter to this query.
-    std::optional<u64> BindCounter(std::shared_ptr<HostCounter> counter_,
-                                   std::optional<u64> timestamp_) {
-        std::optional<u64> result{};
+    void BindCounter(std::shared_ptr<HostCounter> counter_, std::optional<u64> timestamp_) {
         if (counter) {
             // If there's an old counter set it means the query is being rewritten by the game.
             // To avoid losing the data forever, flush here.
-            result = std::make_optional(Flush());
+            Flush();
         }
         counter = std::move(counter_);
         timestamp = timestamp_;
-        return result;
     }
 
     VAddr GetCpuAddr() const noexcept {
@@ -454,14 +374,6 @@ public:
         return with_timestamp ? LARGE_QUERY_SIZE : SMALL_QUERY_SIZE;
     }
 
-    void SetAsyncJob(AsyncJobId assigned_async_job_) {
-        assigned_async_job = assigned_async_job_;
-    }
-
-    AsyncJobId GetAsyncJob() const {
-        return assigned_async_job;
-    }
-
 protected:
     /// Returns true when querying the counter may potentially block.
     bool WaitPending() const noexcept {
@@ -477,7 +389,6 @@ private:
     u8* host_ptr;                         ///< Writable host pointer.
     std::shared_ptr<HostCounter> counter; ///< Host counter to query, owns the dependency tree.
     std::optional<u64> timestamp;         ///< Timestamp to flush to guest memory.
-    AsyncJobId assigned_async_job;
 };
 
 } // namespace VideoCommon

src/video_core/renderer_opengl/gl_fence_manager.h

@@ -30,17 +30,7 @@ private:
 };
 
 using Fence = std::shared_ptr<GLInnerFence>;
-
-struct FenceManagerParams {
-    using FenceType = Fence;
-    using BufferCacheType = BufferCache;
-    using TextureCacheType = TextureCache;
-    using QueryCacheType = QueryCache;
-
-    static constexpr bool HAS_ASYNC_CHECK = false;
-};
-
-using GenericFenceManager = VideoCommon::FenceManager<FenceManagerParams>;
+using GenericFenceManager = VideoCommon::FenceManager<Fence, TextureCache, BufferCache, QueryCache>;
 
 class FenceManagerOpenGL final : public GenericFenceManager {
 public:

src/video_core/renderer_opengl/gl_query_cache.cpp

@@ -26,8 +26,8 @@ constexpr GLenum GetTarget(VideoCore::QueryType type) {
 
 } // Anonymous namespace
 
-QueryCache::QueryCache(RasterizerOpenGL& rasterizer_, Core::Memory::Memory& cpu_memory_)
-    : QueryCacheBase(rasterizer_, cpu_memory_), gl_rasterizer{rasterizer_} {}
+QueryCache::QueryCache(RasterizerOpenGL& rasterizer_)
+    : QueryCacheBase(rasterizer_), gl_rasterizer{rasterizer_} {}
 
 QueryCache::~QueryCache() = default;
 
@@ -74,7 +74,7 @@ void HostCounter::EndQuery() {
     glEndQuery(GetTarget(type));
 }
 
-u64 HostCounter::BlockingQuery([[maybe_unused]] bool async) const {
+u64 HostCounter::BlockingQuery() const {
     GLint64 value;
     glGetQueryObjecti64v(query.handle, GL_QUERY_RESULT, &value);
     return static_cast<u64>(value);
@@ -96,7 +96,7 @@ CachedQuery& CachedQuery::operator=(CachedQuery&& rhs) noexcept {
     return *this;
 }
 
-u64 CachedQuery::Flush([[maybe_unused]] bool async) {
+void CachedQuery::Flush() {
     // Waiting for a query while another query of the same target is enabled locks Nvidia's driver.
     // To avoid this disable and re-enable keeping the dependency stream.
     // But we only have to do this if we have pending waits to be done.
@@ -106,13 +106,11 @@ u64 CachedQuery::Flush([[maybe_unused]] bool async) {
         stream.Update(false);
     }
 
-    auto result = VideoCommon::CachedQueryBase<HostCounter>::Flush();
+    VideoCommon::CachedQueryBase<HostCounter>::Flush();
 
     if (slice_counter) {
         stream.Update(true);
     }
-
-    return result;
 }
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_query_cache.h

@@ -28,7 +28,7 @@ using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>;
 class QueryCache final
     : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> {
 public:
-    explicit QueryCache(RasterizerOpenGL& rasterizer_, Core::Memory::Memory& cpu_memory_);
+    explicit QueryCache(RasterizerOpenGL& rasterizer_);
     ~QueryCache();
 
     OGLQuery AllocateQuery(VideoCore::QueryType type);
@@ -51,7 +51,7 @@ public:
     void EndQuery();
 
 private:
-    u64 BlockingQuery(bool async = false) const override;
+    u64 BlockingQuery() const override;
 
     QueryCache& cache;
     const VideoCore::QueryType type;
@@ -70,7 +70,7 @@ public:
     CachedQuery(const CachedQuery&) = delete;
     CachedQuery& operator=(const CachedQuery&) = delete;
 
-    u64 Flush(bool async = false) override;
+    void Flush() override;
 
 private:
     QueryCache* cache;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -63,7 +63,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
       buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
       shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager,
                    state_tracker, gpu.ShaderNotify()),
-      query_cache(*this, cpu_memory_), accelerate_dma(buffer_cache, texture_cache),
+      query_cache(*this), accelerate_dma(buffer_cache, texture_cache),
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
       blit_image(program_manager_) {}
 
@@ -1287,8 +1287,7 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     }
     const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height);
     static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize;
-    const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing
-                                         : VideoCommon::ObtainBufferOperation::MarkAsWritten;
+    const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing;
     const auto [buffer, offset] =
         buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op);
 
@@ -1299,7 +1298,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     if constexpr (IS_IMAGE_UPLOAD) {
         image->UploadMemory(buffer->Handle(), offset, copy_span);
     } else {
-        image->DownloadMemory(buffer->Handle(), offset, copy_span);
+        texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span,
+                                              buffer_operand.address, buffer_size);
     }
     return true;
 }

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -803,30 +803,40 @@ void Image::UploadMemory(const ImageBufferMap& map,
 
 void Image::DownloadMemory(GLuint buffer_handle, size_t buffer_offset,
                            std::span<const VideoCommon::BufferImageCopy> copies) {
+    std::array buffer_handles{buffer_handle};
+    std::array buffer_offsets{buffer_offset};
+    DownloadMemory(buffer_handles, buffer_offsets, copies);
+}
+
+void Image::DownloadMemory(std::span<GLuint> buffer_handles, std::span<size_t> buffer_offsets,
+                           std::span<const VideoCommon::BufferImageCopy> copies) {
     const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
     if (is_rescaled) {
         ScaleDown();
     }
     glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API
-    glBindBuffer(GL_PIXEL_PACK_BUFFER, buffer_handle);
-    glPixelStorei(GL_PACK_ALIGNMENT, 1);
+    for (size_t i = 0; i < buffer_handles.size(); i++) {
+        auto& buffer_handle = buffer_handles[i];
+        glBindBuffer(GL_PIXEL_PACK_BUFFER, buffer_handle);
+        glPixelStorei(GL_PACK_ALIGNMENT, 1);
 
-    u32 current_row_length = std::numeric_limits<u32>::max();
-    u32 current_image_height = std::numeric_limits<u32>::max();
+        u32 current_row_length = std::numeric_limits<u32>::max();
+        u32 current_image_height = std::numeric_limits<u32>::max();
 
-    for (const VideoCommon::BufferImageCopy& copy : copies) {
-        if (copy.image_subresource.base_level >= gl_num_levels) {
-            continue;
+        for (const VideoCommon::BufferImageCopy& copy : copies) {
+            if (copy.image_subresource.base_level >= gl_num_levels) {
+                continue;
+            }
+            if (current_row_length != copy.buffer_row_length) {
+                current_row_length = copy.buffer_row_length;
+                glPixelStorei(GL_PACK_ROW_LENGTH, current_row_length);
+            }
+            if (current_image_height != copy.buffer_image_height) {
+                current_image_height = copy.buffer_image_height;
+                glPixelStorei(GL_PACK_IMAGE_HEIGHT, current_image_height);
+            }
+            CopyImageToBuffer(copy, buffer_offsets[i]);
         }
-        if (current_row_length != copy.buffer_row_length) {
-            current_row_length = copy.buffer_row_length;
-            glPixelStorei(GL_PACK_ROW_LENGTH, current_row_length);
-        }
-        if (current_image_height != copy.buffer_image_height) {
-            current_image_height = copy.buffer_image_height;
-            glPixelStorei(GL_PACK_IMAGE_HEIGHT, current_image_height);
-        }
-        CopyImageToBuffer(copy, buffer_offset);
     }
     if (is_rescaled) {
         ScaleUp(true);

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -215,6 +215,9 @@ public:
     void DownloadMemory(GLuint buffer_handle, size_t buffer_offset,
                         std::span<const VideoCommon::BufferImageCopy> copies);
 
+    void DownloadMemory(std::span<GLuint> buffer_handle, std::span<size_t> buffer_offset,
+                        std::span<const VideoCommon::BufferImageCopy> copies);
+
     void DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies);
 
     GLuint StorageHandle() noexcept;
@@ -376,6 +379,7 @@ struct TextureCacheParams {
     using Sampler = OpenGL::Sampler;
     using Framebuffer = OpenGL::Framebuffer;
     using AsyncBuffer = u32;
+    using BufferType = GLuint;
 };
 
 using TextureCache = VideoCommon::TextureCache<TextureCacheParams>;

src/video_core/renderer_vulkan/vk_fence_manager.cpp

@@ -5,7 +5,6 @@
 
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_fence_manager.h"
-#include "video_core/renderer_vulkan/vk_query_cache.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/vulkan_common/vulkan_device.h"

src/video_core/renderer_vulkan/vk_fence_manager.h

@@ -40,16 +40,7 @@ private:
 };
 using Fence = std::shared_ptr<InnerFence>;
 
-struct FenceManagerParams {
-    using FenceType = Fence;
-    using BufferCacheType = BufferCache;
-    using TextureCacheType = TextureCache;
-    using QueryCacheType = QueryCache;
-
-    static constexpr bool HAS_ASYNC_CHECK = true;
-};
-
-using GenericFenceManager = VideoCommon::FenceManager<FenceManagerParams>;
+using GenericFenceManager = VideoCommon::FenceManager<Fence, TextureCache, BufferCache, QueryCache>;
 
 class FenceManager final : public GenericFenceManager {
 public:

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -696,6 +696,13 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
 std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env,
     PipelineStatistics* statistics, bool build_in_parallel) try {
+    // TODO: Remove this when Intel fixes their shader compiler.
+    //       https://github.com/IGCIT/Intel-GPU-Community-Issue-Tracker-IGCIT/issues/159
+    if (device.GetDriverID() == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS) {
+        LOG_ERROR(Render_Vulkan, "Skipping 0x{:016x}", key.Hash());
+        return nullptr;
+    }
+
     LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -66,10 +66,9 @@ void QueryPool::Reserve(std::pair<VkQueryPool, u32> query) {
     }
 }
 
-QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_,
-                       Core::Memory::Memory& cpu_memory_, const Device& device_,
+QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_,
                        Scheduler& scheduler_)
-    : QueryCacheBase{rasterizer_, cpu_memory_}, device{device_}, scheduler{scheduler_},
+    : QueryCacheBase{rasterizer_}, device{device_}, scheduler{scheduler_},
       query_pools{
           QueryPool{device_, scheduler_, QueryType::SamplesPassed},
       } {}
@@ -99,10 +98,8 @@ HostCounter::HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> depend
       query{cache_.AllocateQuery(type_)}, tick{cache_.GetScheduler().CurrentTick()} {
     const vk::Device* logical = &cache.GetDevice().GetLogical();
     cache.GetScheduler().Record([logical, query = query](vk::CommandBuffer cmdbuf) {
-        const bool use_precise = Settings::IsGPULevelHigh();
         logical->ResetQueryPool(query.first, query.second, 1);
-        cmdbuf.BeginQuery(query.first, query.second,
-                          use_precise ? VK_QUERY_CONTROL_PRECISE_BIT : 0);
+        cmdbuf.BeginQuery(query.first, query.second, VK_QUERY_CONTROL_PRECISE_BIT);
     });
 }
 
@@ -115,10 +112,8 @@ void HostCounter::EndQuery() {
         [query = query](vk::CommandBuffer cmdbuf) { cmdbuf.EndQuery(query.first, query.second); });
 }
 
-u64 HostCounter::BlockingQuery(bool async) const {
-    if (!async) {
-        cache.GetScheduler().Wait(tick);
-    }
+u64 HostCounter::BlockingQuery() const {
+    cache.GetScheduler().Wait(tick);
     u64 data;
     const VkResult query_result = cache.GetDevice().GetLogical().GetQueryResults(
         query.first, query.second, 1, sizeof(data), &data, sizeof(data),

src/video_core/renderer_vulkan/vk_query_cache.h

@@ -52,8 +52,7 @@ private:
 class QueryCache final
     : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> {
 public:
-    explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_,
-                        Core::Memory::Memory& cpu_memory_, const Device& device_,
+    explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_,
                         Scheduler& scheduler_);
     ~QueryCache();
 
@@ -84,7 +83,7 @@ public:
     void EndQuery();
 
 private:
-    u64 BlockingQuery(bool async = false) const override;
+    u64 BlockingQuery() const override;
 
     QueryCache& cache;
     const VideoCore::QueryType type;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -172,8 +172,7 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
       buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
       pipeline_cache(*this, device, scheduler, descriptor_pool, update_descriptor_queue,
                      render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()),
-      query_cache{*this, cpu_memory_, device, scheduler},
-      accelerate_dma(buffer_cache, texture_cache, scheduler),
+      query_cache{*this, device, scheduler}, accelerate_dma(buffer_cache, texture_cache, scheduler),
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
       wfi_event(device.GetLogical().CreateEvent()) {
     scheduler.SetQueryCache(query_cache);
@@ -676,8 +675,7 @@ bool RasterizerVulkan::AccelerateConditionalRendering() {
     const GPUVAddr condition_address{maxwell3d->regs.render_enable.Address()};
     Maxwell::ReportSemaphore::Compare cmp;
     if (gpu_memory->IsMemoryDirty(condition_address, sizeof(cmp),
-                                  VideoCommon::CacheType::BufferCache |
-                                      VideoCommon::CacheType::QueryCache)) {
+                                  VideoCommon::CacheType::BufferCache)) {
         return true;
     }
     return false;
@@ -783,8 +781,7 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     }
     const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height);
     static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize;
-    const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing
-                                         : VideoCommon::ObtainBufferOperation::MarkAsWritten;
+    const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing;
     const auto [buffer, offset] =
         buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op);
 
@@ -795,7 +792,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     if constexpr (IS_IMAGE_UPLOAD) {
         image->UploadMemory(buffer->Handle(), offset, copy_span);
     } else {
-        image->DownloadMemory(buffer->Handle(), offset, copy_span);
+        texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span,
+                                              buffer_operand.address, buffer_size);
     }
     return true;
 }

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -65,6 +65,18 @@ VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 wi
     return extent;
 }
 
+VkCompositeAlphaFlagBitsKHR ChooseAlphaFlags(const VkSurfaceCapabilitiesKHR& capabilities) {
+    if (capabilities.supportedCompositeAlpha & VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR) {
+        return VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR;
+    } else if (capabilities.supportedCompositeAlpha & VK_COMPOSITE_ALPHA_INHERIT_BIT_KHR) {
+        return VK_COMPOSITE_ALPHA_INHERIT_BIT_KHR;
+    } else {
+        LOG_ERROR(Render_Vulkan, "Unknown composite alpha flags value {:#x}",
+                  capabilities.supportedCompositeAlpha);
+        return VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR;
+    }
+}
+
 } // Anonymous namespace
 
 Swapchain::Swapchain(VkSurfaceKHR surface_, const Device& device_, Scheduler& scheduler_,
@@ -155,6 +167,7 @@ void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bo
     const auto formats{physical_device.GetSurfaceFormatsKHR(surface)};
     const auto present_modes{physical_device.GetSurfacePresentModesKHR(surface)};
 
+    const VkCompositeAlphaFlagBitsKHR alpha_flags{ChooseAlphaFlags(capabilities)};
     const VkSurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats)};
     present_mode = ChooseSwapPresentMode(present_modes);
 
@@ -185,7 +198,7 @@ void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bo
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,
         .preTransform = capabilities.currentTransform,
-        .compositeAlpha = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR,
+        .compositeAlpha = alpha_flags,
         .presentMode = present_mode,
         .clipped = VK_FALSE,
         .oldSwapchain = nullptr,

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -1,10 +1,11 @@
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
+// SPDX-License-Identifier: GPL-3.0-or-later
 
 #include <algorithm>
 #include <array>
 #include <span>
 #include <vector>
+#include <boost/container/small_vector.hpp>
 
 #include "common/bit_cast.h"
 #include "common/bit_util.h"
@@ -1343,14 +1344,31 @@ void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImag
 
 void Image::DownloadMemory(VkBuffer buffer, VkDeviceSize offset,
                            std::span<const VideoCommon::BufferImageCopy> copies) {
+    std::array buffer_handles{
+        buffer,
+    };
+    std::array buffer_offsets{
+        offset,
+    };
+    DownloadMemory(buffer_handles, buffer_offsets, copies);
+}
+
+void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<VkDeviceSize> offsets_span,
+                           std::span<const VideoCommon::BufferImageCopy> copies) {
     const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
     if (is_rescaled) {
         ScaleDown();
     }
-    std::vector vk_copies = TransformBufferImageCopies(copies, offset, aspect_mask);
+    boost::container::small_vector<VkBuffer, 1> buffers_vector{};
+    boost::container::small_vector<std::vector<VkBufferImageCopy>, 1> vk_copies;
+    for (size_t index = 0; index < buffers_span.size(); index++) {
+        buffers_vector.emplace_back(buffers_span[index]);
+        vk_copies.emplace_back(
+            TransformBufferImageCopies(copies, offsets_span[index], aspect_mask));
+    }
     scheduler->RequestOutsideRenderPassOperationContext();
-    scheduler->Record([buffer, image = *original_image, aspect_mask = aspect_mask,
-                       vk_copies](vk::CommandBuffer cmdbuf) {
+    scheduler->Record([buffers = std::move(buffers_vector), image = *original_image,
+                       aspect_mask = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) {
         const VkImageMemoryBarrier read_barrier{
             .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
             .pNext = nullptr,
@@ -1369,6 +1387,20 @@ void Image::DownloadMemory(VkBuffer buffer, VkDeviceSize offset,
                 .layerCount = VK_REMAINING_ARRAY_LAYERS,
             },
         };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                               0, read_barrier);
+
+        for (size_t index = 0; index < buffers.size(); index++) {
+            cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffers[index],
+                                     vk_copies[index]);
+        }
+
+        const VkMemoryBarrier memory_write_barrier{
+            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+        };
         const VkImageMemoryBarrier image_write_barrier{
             .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
             .pNext = nullptr,
@@ -1387,15 +1419,6 @@ void Image::DownloadMemory(VkBuffer buffer, VkDeviceSize offset,
                 .layerCount = VK_REMAINING_ARRAY_LAYERS,
             },
         };
-        const VkMemoryBarrier memory_write_barrier{
-            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
-        };
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               0, read_barrier);
-        cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, vk_copies);
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                0, memory_write_barrier, nullptr, image_write_barrier);
     });
@@ -1405,7 +1428,13 @@ void Image::DownloadMemory(VkBuffer buffer, VkDeviceSize offset,
 }
 
 void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
-    DownloadMemory(map.buffer, map.offset, copies);
+    std::array buffers{
+        map.buffer,
+    };
+    std::array offsets{
+        map.offset,
+    };
+    DownloadMemory(buffers, offsets, copies);
 }
 
 bool Image::IsRescaled() const noexcept {

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -1,5 +1,5 @@
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
+// SPDX-License-Identifier: GPL-3.0-or-later
 
 #pragma once
 
@@ -141,6 +141,9 @@ public:
     void DownloadMemory(VkBuffer buffer, VkDeviceSize offset,
                         std::span<const VideoCommon::BufferImageCopy> copies);
 
+    void DownloadMemory(std::span<VkBuffer> buffers, std::span<VkDeviceSize> offsets,
+                        std::span<const VideoCommon::BufferImageCopy> copies);
+
     void DownloadMemory(const StagingBufferRef& map,
                         std::span<const VideoCommon::BufferImageCopy> copies);
 
@@ -371,6 +374,7 @@ struct TextureCacheParams {
     using Sampler = Vulkan::Sampler;
     using Framebuffer = Vulkan::Framebuffer;
     using AsyncBuffer = Vulkan::StagingBufferRef;
+    using BufferType = VkBuffer;
 };
 
 using TextureCache = VideoCommon::TextureCache<TextureCacheParams>;

src/video_core/texture_cache/texture_cache.h

@@ -1,9 +1,10 @@
-// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
+// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #pragma once
 
 #include <unordered_set>
+#include <boost/container/small_vector.hpp>
 
 #include "common/alignment.h"
 #include "common/settings.h"
@@ -17,15 +18,10 @@
 
 namespace VideoCommon {
 
-using Tegra::Texture::SwizzleSource;
-using Tegra::Texture::TextureType;
 using Tegra::Texture::TICEntry;
 using Tegra::Texture::TSCEntry;
 using VideoCore::Surface::GetFormatType;
-using VideoCore::Surface::IsCopyCompatible;
 using VideoCore::Surface::PixelFormat;
-using VideoCore::Surface::PixelFormatFromDepthFormat;
-using VideoCore::Surface::PixelFormatFromRenderTargetFormat;
 using VideoCore::Surface::SurfaceType;
 using namespace Common::Literals;
 
@@ -143,6 +139,13 @@ void TextureCache<P>::TickFrame() {
     runtime.TickFrame();
     critical_gc = 0;
     ++frame_tick;
+
+    if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+        for (auto& buffer : async_buffers_death_ring) {
+            runtime.FreeDeferredStagingBuffer(buffer);
+        }
+        async_buffers_death_ring.clear();
+    }
 }
 
 template <class P>
@@ -661,25 +664,39 @@ template <class P>
 void TextureCache<P>::CommitAsyncFlushes() {
     // This is intentionally passing the value by copy
     if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
-        const std::span<const ImageId> download_ids = uncommitted_downloads;
+        auto& download_ids = uncommitted_downloads;
         if (download_ids.empty()) {
             committed_downloads.emplace_back(std::move(uncommitted_downloads));
             uncommitted_downloads.clear();
-            async_buffers.emplace_back(std::optional<AsyncBuffer>{});
+            async_buffers.emplace_back(std::move(uncommitted_async_buffers));
+            uncommitted_async_buffers.clear();
             return;
         }
         size_t total_size_bytes = 0;
-        for (const ImageId image_id : download_ids) {
-            total_size_bytes += slot_images[image_id].unswizzled_size_bytes;
+        size_t last_async_buffer_id = uncommitted_async_buffers.size();
+        bool any_none_dma = false;
+        for (PendingDownload& download_info : download_ids) {
+            if (download_info.is_swizzle) {
+                total_size_bytes +=
+                    Common::AlignUp(slot_images[download_info.object_id].unswizzled_size_bytes, 64);
+                any_none_dma = true;
+                download_info.async_buffer_id = last_async_buffer_id;
+            }
         }
-        auto download_map = runtime.DownloadStagingBuffer(total_size_bytes, true);
-        for (const ImageId image_id : download_ids) {
-            Image& image = slot_images[image_id];
-            const auto copies = FullDownloadCopies(image.info);
-            image.DownloadMemory(download_map, copies);
-            download_map.offset += Common::AlignUp(image.unswizzled_size_bytes, 64);
+        if (any_none_dma) {
+            auto download_map = runtime.DownloadStagingBuffer(total_size_bytes, true);
+            for (const PendingDownload& download_info : download_ids) {
+                if (download_info.is_swizzle) {
+                    Image& image = slot_images[download_info.object_id];
+                    const auto copies = FullDownloadCopies(image.info);
+                    image.DownloadMemory(download_map, copies);
+                    download_map.offset += Common::AlignUp(image.unswizzled_size_bytes, 64);
+                }
+            }
+            uncommitted_async_buffers.emplace_back(download_map);
         }
-        async_buffers.emplace_back(download_map);
+        async_buffers.emplace_back(std::move(uncommitted_async_buffers));
+        uncommitted_async_buffers.clear();
     }
     committed_downloads.emplace_back(std::move(uncommitted_downloads));
     uncommitted_downloads.clear();
@@ -691,39 +708,57 @@ void TextureCache<P>::PopAsyncFlushes() {
         return;
     }
     if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
-        const std::span<const ImageId> download_ids = committed_downloads.front();
+        const auto& download_ids = committed_downloads.front();
         if (download_ids.empty()) {
             committed_downloads.pop_front();
             async_buffers.pop_front();
             return;
         }
-        auto download_map = *async_buffers.front();
-        std::span<u8> download_span = download_map.mapped_span;
+        auto download_map = std::move(async_buffers.front());
         for (size_t i = download_ids.size(); i > 0; i--) {
-            const ImageBase& image = slot_images[download_ids[i - 1]];
-            const auto copies = FullDownloadCopies(image.info);
-            download_map.offset -= Common::AlignUp(image.unswizzled_size_bytes, 64);
-            std::span<u8> download_span_alt = download_span.subspan(download_map.offset);
-            SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span_alt,
-                         swizzle_data_buffer);
+            auto& download_info = download_ids[i - 1];
+            auto& download_buffer = download_map[download_info.async_buffer_id];
+            if (download_info.is_swizzle) {
+                const ImageBase& image = slot_images[download_info.object_id];
+                const auto copies = FullDownloadCopies(image.info);
+                download_buffer.offset -= Common::AlignUp(image.unswizzled_size_bytes, 64);
+                std::span<u8> download_span =
+                    download_buffer.mapped_span.subspan(download_buffer.offset);
+                SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span,
+                             swizzle_data_buffer);
+            } else {
+                const BufferDownload& buffer_info = slot_buffer_downloads[download_info.object_id];
+                std::span<u8> download_span =
+                    download_buffer.mapped_span.subspan(download_buffer.offset);
+                gpu_memory->WriteBlockUnsafe(buffer_info.address, download_span.data(),
+                                             buffer_info.size);
+                slot_buffer_downloads.erase(download_info.object_id);
+            }
+        }
+        for (auto& download_buffer : download_map) {
+            async_buffers_death_ring.emplace_back(download_buffer);
         }
-        runtime.FreeDeferredStagingBuffer(download_map);
         committed_downloads.pop_front();
         async_buffers.pop_front();
     } else {
-        const std::span<const ImageId> download_ids = committed_downloads.front();
+        const auto& download_ids = committed_downloads.front();
         if (download_ids.empty()) {
             committed_downloads.pop_front();
             return;
         }
         size_t total_size_bytes = 0;
-        for (const ImageId image_id : download_ids) {
-            total_size_bytes += slot_images[image_id].unswizzled_size_bytes;
+        for (const PendingDownload& download_info : download_ids) {
+            if (download_info.is_swizzle) {
+                total_size_bytes += slot_images[download_info.object_id].unswizzled_size_bytes;
+            }
         }
         auto download_map = runtime.DownloadStagingBuffer(total_size_bytes);
         const size_t original_offset = download_map.offset;
-        for (const ImageId image_id : download_ids) {
-            Image& image = slot_images[image_id];
+        for (const PendingDownload& download_info : download_ids) {
+            if (!download_info.is_swizzle) {
+                continue;
+            }
+            Image& image = slot_images[download_info.object_id];
             const auto copies = FullDownloadCopies(image.info);
             image.DownloadMemory(download_map, copies);
             download_map.offset += image.unswizzled_size_bytes;
@@ -732,8 +767,11 @@ void TextureCache<P>::PopAsyncFlushes() {
         runtime.Finish();
         download_map.offset = original_offset;
         std::span<u8> download_span = download_map.mapped_span;
-        for (const ImageId image_id : download_ids) {
-            const ImageBase& image = slot_images[image_id];
+        for (const PendingDownload& download_info : download_ids) {
+            if (!download_info.is_swizzle) {
+                continue;
+            }
+            const ImageBase& image = slot_images[download_info.object_id];
             const auto copies = FullDownloadCopies(image.info);
             SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span,
                          swizzle_data_buffer);
@@ -833,6 +871,33 @@ std::pair<typename TextureCache<P>::Image*, BufferImageCopy> TextureCache<P>::Dm
     return {image, copy};
 }
 
+template <class P>
+void TextureCache<P>::DownloadImageIntoBuffer(typename TextureCache<P>::Image* image,
+                                              typename TextureCache<P>::BufferType buffer,
+                                              size_t buffer_offset,
+                                              std::span<const VideoCommon::BufferImageCopy> copies,
+                                              GPUVAddr address, size_t size) {
+    if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+        const BufferDownload new_buffer_download{address, size};
+        auto slot = slot_buffer_downloads.insert(new_buffer_download);
+        const PendingDownload new_download{false, uncommitted_async_buffers.size(), slot};
+        uncommitted_downloads.emplace_back(new_download);
+        auto download_map = runtime.DownloadStagingBuffer(size, true);
+        uncommitted_async_buffers.emplace_back(download_map);
+        std::array buffers{
+            buffer,
+            download_map.buffer,
+        };
+        std::array buffer_offsets{
+            buffer_offset,
+            download_map.offset,
+        };
+        image->DownloadMemory(buffers, buffer_offsets, copies);
+    } else {
+        image->DownloadMemory(buffer, buffer_offset, copies);
+    }
+}
+
 template <class P>
 void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {
     if (False(image.flags & ImageFlagBits::CpuModified)) {
@@ -2209,7 +2274,8 @@ void TextureCache<P>::BindRenderTarget(ImageViewId* old_id, ImageViewId new_id)
     if (new_id) {
         const ImageViewBase& old_view = slot_image_views[new_id];
         if (True(old_view.flags & ImageViewFlagBits::PreemtiveDownload)) {
-            uncommitted_downloads.push_back(old_view.image_id);
+            const PendingDownload new_download{true, 0, old_view.image_id};
+            uncommitted_downloads.emplace_back(new_download);
         }
     }
     *old_id = new_id;

src/video_core/texture_cache/texture_cache_base.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
+// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #pragma once
@@ -40,14 +40,9 @@ struct ChannelState;
 
 namespace VideoCommon {
 
-using Tegra::Texture::SwizzleSource;
 using Tegra::Texture::TICEntry;
 using Tegra::Texture::TSCEntry;
-using VideoCore::Surface::GetFormatType;
-using VideoCore::Surface::IsCopyCompatible;
 using VideoCore::Surface::PixelFormat;
-using VideoCore::Surface::PixelFormatFromDepthFormat;
-using VideoCore::Surface::PixelFormatFromRenderTargetFormat;
 using namespace Common::Literals;
 
 struct ImageViewInOut {
@@ -119,6 +114,7 @@ class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelI
     using Sampler = typename P::Sampler;
     using Framebuffer = typename P::Framebuffer;
     using AsyncBuffer = typename P::AsyncBuffer;
+    using BufferType = typename P::BufferType;
 
     struct BlitImages {
         ImageId dst_id;
@@ -215,6 +211,10 @@ public:
         const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand,
         const Tegra::DMA::ImageOperand& image_operand, ImageId image_id, bool modifies_image);
 
+    void DownloadImageIntoBuffer(Image* image, BufferType buffer, size_t buffer_offset,
+                                 std::span<const VideoCommon::BufferImageCopy> copies,
+                                 GPUVAddr address = 0, size_t size = 0);
+
     /// Return true when a CPU region is modified from the GPU
     [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
 
@@ -424,17 +424,32 @@ private:
     u64 critical_memory;
     size_t critical_gc;
 
+    struct BufferDownload {
+        GPUVAddr address;
+        size_t size;
+    };
+
+    struct PendingDownload {
+        bool is_swizzle;
+        size_t async_buffer_id;
+        SlotId object_id;
+    };
+
     SlotVector<Image> slot_images;
     SlotVector<ImageMapView> slot_map_views;
     SlotVector<ImageView> slot_image_views;
     SlotVector<ImageAlloc> slot_image_allocs;
     SlotVector<Sampler> slot_samplers;
     SlotVector<Framebuffer> slot_framebuffers;
+    SlotVector<BufferDownload> slot_buffer_downloads;
 
     // TODO: This data structure is not optimal and it should be reworked
-    std::vector<ImageId> uncommitted_downloads;
-    std::deque<std::vector<ImageId>> committed_downloads;
-    std::deque<std::optional<AsyncBuffer>> async_buffers;
+
+    std::vector<PendingDownload> uncommitted_downloads;
+    std::deque<std::vector<PendingDownload>> committed_downloads;
+    std::vector<AsyncBuffer> uncommitted_async_buffers;
+    std::deque<std::vector<AsyncBuffer>> async_buffers;
+    std::deque<AsyncBuffer> async_buffers_death_ring;
 
     struct LRUItemParams {
         using ObjectType = ImageId;

src/yuzu/main.cpp

@@ -27,6 +27,7 @@
 #include "configuration/configure_input.h"
 #include "configuration/configure_per_game.h"
 #include "configuration/configure_tas.h"
+#include "core/file_sys/romfs_factory.h"
 #include "core/file_sys/vfs.h"
 #include "core/file_sys/vfs_real.h"
 #include "core/frontend/applets/cabinet.h"
@@ -4171,6 +4172,8 @@ void GMainWindow::OnReinitializeKeys(ReinitializeKeyBehavior behavior) {
     }
 
     Core::Crypto::KeyManager& keys = Core::Crypto::KeyManager::Instance();
+    bool all_keys_present{true};
+
     if (keys.BaseDeriveNecessary()) {
         Core::Crypto::PartitionDataManager pdm{vfs->OpenDirectory("", FileSys::Mode::Read)};
 
@@ -4195,6 +4198,7 @@ void GMainWindow::OnReinitializeKeys(ReinitializeKeyBehavior behavior) {
             errors += tr(" - Missing PRODINFO");
         }
         if (!errors.isEmpty()) {
+            all_keys_present = false;
             QMessageBox::warning(
                 this, tr("Derivation Components Missing"),
                 tr("Encryption keys are missing. "
@@ -4222,11 +4226,40 @@ void GMainWindow::OnReinitializeKeys(ReinitializeKeyBehavior behavior) {
 
     system->GetFileSystemController().CreateFactories(*vfs);
 
+    if (all_keys_present && !this->CheckSystemArchiveDecryption()) {
+        LOG_WARNING(Frontend, "Mii model decryption failed");
+        QMessageBox::warning(
+            this, tr("System Archive Decryption Failed"),
+            tr("Encryption keys failed to decrypt firmware. "
+               "<br>Please follow <a href='https://yuzu-emu.org/help/quickstart/'>the yuzu "
+               "quickstart guide</a> to get all your keys, firmware and "
+               "games."));
+    }
+
     if (behavior == ReinitializeKeyBehavior::Warning) {
         game_list->PopulateAsync(UISettings::values.game_dirs);
     }
 }
 
+bool GMainWindow::CheckSystemArchiveDecryption() {
+    constexpr u64 MiiModelId = 0x0100000000000802;
+
+    auto bis_system = system->GetFileSystemController().GetSystemNANDContents();
+    if (!bis_system) {
+        // Not having system BIS files is not an error.
+        return true;
+    }
+
+    auto mii_nca = bis_system->GetEntry(MiiModelId, FileSys::ContentRecordType::Data);
+    if (!mii_nca) {
+        // Not having the Mii model is not an error.
+        return true;
+    }
+
+    // Return whether we are able to decrypt the RomFS of the Mii model.
+    return mii_nca->GetRomFS().get() != nullptr;
+}
+
 std::optional<u64> GMainWindow::SelectRomFSDumpTarget(const FileSys::ContentProvider& installed,
                                                       u64 program_id) {
     const auto dlc_entries =

src/yuzu/main.h

@@ -392,6 +392,7 @@ private:
     void LoadTranslation();
     void OpenPerGameConfiguration(u64 title_id, const std::string& file_name);
     bool CheckDarkMode();
+    bool CheckSystemArchiveDecryption();
 
     QString GetTasStateDescription() const;
     bool CreateShortcut(const std::string& shortcut_path, const std::string& title,