Android #46

Support masked depthstencil clears

README.md

@@ -1,3 +1,11 @@
+| Pull Request | Commit | Title | Author | Merged? |
+|----|----|----|----|----|
+
+
+End of merge log. You can find the original README.md below the break.
+
+-----
+
 <!--
 SPDX-FileCopyrightText: 2018 yuzu Emulator Project
 SPDX-License-Identifier: GPL-2.0-or-later

src/CMakeLists.txt

@@ -134,7 +134,7 @@ else()
     endif()
 
     # GCC bugs
-    if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL "12" AND CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+    if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL "11" AND CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
         # These diagnostics would be great if they worked, but are just completely broken
         # and produce bogus errors on external libraries like fmt.
         add_compile_options(

src/core/hle/kernel/k_hardware_timer.h

@@ -19,13 +19,7 @@ public:
     void Initialize();
     void Finalize();
 
-    s64 GetCount() const {
+    s64 GetTick() const;
-        return GetTick();
-    }
-
-    void RegisterTask(KTimerTask* task, s64 time_from_now) {
-        this->RegisterAbsoluteTask(task, GetTick() + time_from_now);
-    }
 
     void RegisterAbsoluteTask(KTimerTask* task, s64 task_time) {
         KScopedDisableDispatch dd{m_kernel};
@@ -42,7 +36,6 @@ private:
     void EnableInterrupt(s64 wakeup_time);
     void DisableInterrupt();
     bool GetInterruptEnabled();
-    s64 GetTick() const;
     void DoTask();
 
 private:

src/core/hle/kernel/k_resource_limit.cpp

@@ -5,6 +5,7 @@
 #include "common/overflow.h"
 #include "core/core.h"
 #include "core/core_timing.h"
+#include "core/hle/kernel/k_hardware_timer.h"
 #include "core/hle/kernel/k_resource_limit.h"
 #include "core/hle/kernel/svc_results.h"
 
@@ -15,9 +16,7 @@ KResourceLimit::KResourceLimit(KernelCore& kernel)
     : KAutoObjectWithSlabHeapAndContainer{kernel}, m_lock{m_kernel}, m_cond_var{m_kernel} {}
 KResourceLimit::~KResourceLimit() = default;
 
-void KResourceLimit::Initialize(const Core::Timing::CoreTiming* core_timing) {
+void KResourceLimit::Initialize() {}
-    m_core_timing = core_timing;
-}
 
 void KResourceLimit::Finalize() {}
 
@@ -86,7 +85,7 @@ Result KResourceLimit::SetLimitValue(LimitableResource which, s64 value) {
 }
 
 bool KResourceLimit::Reserve(LimitableResource which, s64 value) {
-    return Reserve(which, value, m_core_timing->GetGlobalTimeNs().count() + DefaultTimeout);
+    return Reserve(which, value, m_kernel.HardwareTimer().GetTick() + DefaultTimeout);
 }
 
 bool KResourceLimit::Reserve(LimitableResource which, s64 value, s64 timeout) {
@@ -117,7 +116,7 @@ bool KResourceLimit::Reserve(LimitableResource which, s64 value, s64 timeout) {
         }
 
         if (m_current_hints[index] + value <= m_limit_values[index] &&
-            (timeout < 0 || m_core_timing->GetGlobalTimeNs().count() < timeout)) {
+            (timeout < 0 || m_kernel.HardwareTimer().GetTick() < timeout)) {
             m_waiter_count++;
             m_cond_var.Wait(std::addressof(m_lock), timeout, false);
             m_waiter_count--;
@@ -154,7 +153,7 @@ void KResourceLimit::Release(LimitableResource which, s64 value, s64 hint) {
 
 KResourceLimit* CreateResourceLimitForProcess(Core::System& system, s64 physical_memory_size) {
     auto* resource_limit = KResourceLimit::Create(system.Kernel());
-    resource_limit->Initialize(std::addressof(system.CoreTiming()));
+    resource_limit->Initialize();
 
     // Initialize default resource limit values.
     // TODO(bunnei): These values are the system defaults, the limits for service processes are

src/core/hle/kernel/k_resource_limit.h

@@ -31,7 +31,7 @@ public:
     explicit KResourceLimit(KernelCore& kernel);
     ~KResourceLimit() override;
 
-    void Initialize(const Core::Timing::CoreTiming* core_timing);
+    void Initialize();
     void Finalize() override;
 
     s64 GetLimitValue(LimitableResource which) const;
@@ -57,7 +57,6 @@ private:
     mutable KLightLock m_lock;
     s32 m_waiter_count{};
     KLightConditionVariable m_cond_var;
-    const Core::Timing::CoreTiming* m_core_timing{};
 };
 
 KResourceLimit* CreateResourceLimitForProcess(Core::System& system, s64 physical_memory_size);

src/core/hle/kernel/k_scoped_scheduler_lock_and_sleep.h

@@ -28,7 +28,7 @@ public:
     ~KScopedSchedulerLockAndSleep() {
         // Register the sleep.
         if (m_timeout_tick > 0) {
-            m_timer->RegisterTask(m_thread, m_timeout_tick);
+            m_timer->RegisterAbsoluteTask(m_thread, m_timeout_tick);
         }
 
         // Unlock the scheduler.

src/core/hle/kernel/kernel.cpp

@@ -231,7 +231,7 @@ struct KernelCore::Impl {
     void InitializeSystemResourceLimit(KernelCore& kernel,
                                        const Core::Timing::CoreTiming& core_timing) {
         system_resource_limit = KResourceLimit::Create(system.Kernel());
-        system_resource_limit->Initialize(&core_timing);
+        system_resource_limit->Initialize();
         KResourceLimit::Register(kernel, system_resource_limit);
 
         const auto sizes{memory_layout->GetTotalAndKernelMemorySizes()};

src/core/hle/kernel/svc/svc_address_arbiter.cpp

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include "core/core.h"
+#include "core/hle/kernel/k_hardware_timer.h"
 #include "core/hle/kernel/k_memory_layout.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/kernel.h"
@@ -52,7 +53,7 @@ Result WaitForAddress(Core::System& system, u64 address, ArbitrationType arb_typ
     if (timeout_ns > 0) {
         const s64 offset_tick(timeout_ns);
         if (offset_tick > 0) {
-            timeout = offset_tick + 2;
+            timeout = system.Kernel().HardwareTimer().GetTick() + offset_tick + 2;
             if (timeout <= 0) {
                 timeout = std::numeric_limits<s64>::max();
             }

src/core/hle/kernel/svc/svc_condition_variable.cpp

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include "core/core.h"
+#include "core/hle/kernel/k_hardware_timer.h"
 #include "core/hle/kernel/k_memory_layout.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/kernel.h"
@@ -25,7 +26,7 @@ Result WaitProcessWideKeyAtomic(Core::System& system, u64 address, u64 cv_key, u
     if (timeout_ns > 0) {
         const s64 offset_tick(timeout_ns);
         if (offset_tick > 0) {
-            timeout = offset_tick + 2;
+            timeout = system.Kernel().HardwareTimer().GetTick() + offset_tick + 2;
             if (timeout <= 0) {
                 timeout = std::numeric_limits<s64>::max();
             }

src/core/hle/kernel/svc/svc_ipc.cpp

@@ -5,6 +5,7 @@
 #include "common/scratch_buffer.h"
 #include "core/core.h"
 #include "core/hle/kernel/k_client_session.h"
+#include "core/hle/kernel/k_hardware_timer.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/k_server_session.h"
 #include "core/hle/kernel/svc.h"
@@ -82,12 +83,29 @@ Result ReplyAndReceive(Core::System& system, s32* out_index, uint64_t handles_ad
         R_TRY(session->SendReply());
     }
 
+    // Convert the timeout from nanoseconds to ticks.
+    // NOTE: Nintendo does not use this conversion logic in WaitSynchronization...
+    s64 timeout;
+    if (timeout_ns > 0) {
+        const s64 offset_tick(timeout_ns);
+        if (offset_tick > 0) {
+            timeout = kernel.HardwareTimer().GetTick() + offset_tick + 2;
+            if (timeout <= 0) {
+                timeout = std::numeric_limits<s64>::max();
+            }
+        } else {
+            timeout = std::numeric_limits<s64>::max();
+        }
+    } else {
+        timeout = timeout_ns;
+    }
+
     // Wait for a message.
     while (true) {
         // Wait for an object.
         s32 index;
         Result result = KSynchronizationObject::Wait(kernel, std::addressof(index), objs.data(),
-                                                     num_handles, timeout_ns);
+                                                     num_handles, timeout);
         if (result == ResultTimedOut) {
             R_RETURN(result);
         }

src/core/hle/kernel/svc/svc_resource_limit.cpp

@@ -21,7 +21,7 @@ Result CreateResourceLimit(Core::System& system, Handle* out_handle) {
     SCOPE_EXIT({ resource_limit->Close(); });
 
     // Initialize the resource limit.
-    resource_limit->Initialize(std::addressof(system.CoreTiming()));
+    resource_limit->Initialize();
 
     // Register the limit.
     KResourceLimit::Register(kernel, resource_limit);

src/core/hle/kernel/svc/svc_synchronization.cpp

@@ -4,6 +4,7 @@
 #include "common/scope_exit.h"
 #include "common/scratch_buffer.h"
 #include "core/core.h"
+#include "core/hle/kernel/k_hardware_timer.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/k_readable_event.h"
 #include "core/hle/kernel/svc.h"
@@ -83,9 +84,20 @@ Result WaitSynchronization(Core::System& system, int32_t* out_index, u64 user_ha
         }
     });
 
+    // Convert the timeout from nanoseconds to ticks.
+    s64 timeout;
+    if (timeout_ns > 0) {
+        u64 ticks = kernel.HardwareTimer().GetTick();
+        ticks += timeout_ns;
+        ticks += 2;
+
+        timeout = ticks;
+    } else {
+        timeout = timeout_ns;
+    }
+
     // Wait on the objects.
-    Result res =
+    Result res = KSynchronizationObject::Wait(kernel, out_index, objs.data(), num_handles, timeout);
-        KSynchronizationObject::Wait(kernel, out_index, objs.data(), num_handles, timeout_ns);
 
     R_SUCCEED_IF(res == ResultSessionClosed);
     R_RETURN(res);

src/core/hle/kernel/svc/svc_thread.cpp

@@ -4,6 +4,7 @@
 #include "common/scope_exit.h"
 #include "core/core.h"
 #include "core/core_timing.h"
+#include "core/hle/kernel/k_hardware_timer.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/k_scoped_resource_reservation.h"
 #include "core/hle/kernel/k_thread.h"
@@ -42,9 +43,9 @@ Result CreateThread(Core::System& system, Handle* out_handle, u64 entry_point, u
     R_UNLESS(process.CheckThreadPriority(priority), ResultInvalidPriority);
 
     // Reserve a new thread from the process resource limit (waiting up to 100ms).
-    KScopedResourceReservation thread_reservation(
+    KScopedResourceReservation thread_reservation(std::addressof(process),
-        std::addressof(process), LimitableResource::ThreadCountMax, 1,
+                                                  LimitableResource::ThreadCountMax, 1,
-        system.CoreTiming().GetGlobalTimeNs().count() + 100000000);
+                                                  kernel.HardwareTimer().GetTick() + 100000000);
     R_UNLESS(thread_reservation.Succeeded(), ResultLimitReached);
 
     // Create the thread.
@@ -102,20 +103,31 @@ void ExitThread(Core::System& system) {
 }
 
 /// Sleep the current thread
-void SleepThread(Core::System& system, s64 nanoseconds) {
+void SleepThread(Core::System& system, s64 ns) {
     auto& kernel = system.Kernel();
-    const auto yield_type = static_cast<Svc::YieldType>(nanoseconds);
+    const auto yield_type = static_cast<Svc::YieldType>(ns);
 
-    LOG_TRACE(Kernel_SVC, "called nanoseconds={}", nanoseconds);
+    LOG_TRACE(Kernel_SVC, "called nanoseconds={}", ns);
 
     // When the input tick is positive, sleep.
-    if (nanoseconds > 0) {
+    if (ns > 0) {
         // Convert the timeout from nanoseconds to ticks.
         // NOTE: Nintendo does not use this conversion logic in WaitSynchronization...
+        s64 timeout;
+
+        const s64 offset_tick(ns);
+        if (offset_tick > 0) {
+            timeout = kernel.HardwareTimer().GetTick() + offset_tick + 2;
+            if (timeout <= 0) {
+                timeout = std::numeric_limits<s64>::max();
+            }
+        } else {
+            timeout = std::numeric_limits<s64>::max();
+        }
 
         // Sleep.
         // NOTE: Nintendo does not check the result of this sleep.
-        static_cast<void>(GetCurrentThread(kernel).Sleep(nanoseconds));
+        static_cast<void>(GetCurrentThread(kernel).Sleep(timeout));
     } else if (yield_type == Svc::YieldType::WithoutCoreMigration) {
         KScheduler::YieldWithoutCoreMigration(kernel);
     } else if (yield_type == Svc::YieldType::WithCoreMigration) {
@@ -124,7 +136,6 @@ void SleepThread(Core::System& system, s64 nanoseconds) {
         KScheduler::YieldToAnyThread(kernel);
     } else {
         // Nintendo does nothing at all if an otherwise invalid value is passed.
-        ASSERT_MSG(false, "Unimplemented sleep yield type '{:016X}'!", nanoseconds);
     }
 }
 

src/core/hle/service/sockets/sockets.h

@@ -18,7 +18,9 @@ enum class Errno : u32 {
     AGAIN = 11,
     INVAL = 22,
     MFILE = 24,
+    PIPE = 32,
     MSGSIZE = 90,
+    CONNABORTED = 103,
     CONNRESET = 104,
     NOTCONN = 107,
     TIMEDOUT = 110,

src/core/hle/service/sockets/sockets_translate.cpp

@@ -23,10 +23,14 @@ Errno Translate(Network::Errno value) {
         return Errno::INVAL;
     case Network::Errno::MFILE:
         return Errno::MFILE;
+    case Network::Errno::PIPE:
+        return Errno::PIPE;
     case Network::Errno::NOTCONN:
         return Errno::NOTCONN;
     case Network::Errno::TIMEDOUT:
         return Errno::TIMEDOUT;
+    case Network::Errno::CONNABORTED:
+        return Errno::CONNABORTED;
     case Network::Errno::CONNRESET:
         return Errno::CONNRESET;
     case Network::Errno::INPROGRESS:

src/core/internal_network/network.cpp

@@ -39,6 +39,11 @@ namespace Network {
 
 namespace {
 
+enum class CallType {
+    Send,
+    Other,
+};
+
 #ifdef _WIN32
 
 using socklen_t = int;
@@ -96,7 +101,7 @@ bool EnableNonBlock(SOCKET fd, bool enable) {
     return ioctlsocket(fd, FIONBIO, &value) != SOCKET_ERROR;
 }
 
-Errno TranslateNativeError(int e) {
+Errno TranslateNativeError(int e, CallType call_type = CallType::Other) {
     switch (e) {
     case 0:
         return Errno::SUCCESS;
@@ -112,6 +117,14 @@ Errno TranslateNativeError(int e) {
         return Errno::AGAIN;
     case WSAECONNREFUSED:
         return Errno::CONNREFUSED;
+    case WSAECONNABORTED:
+        if (call_type == CallType::Send) {
+            // Winsock yields WSAECONNABORTED from `send` in situations where Unix
+            // systems, and actual Switches, yield EPIPE.
+            return Errno::PIPE;
+        } else {
+            return Errno::CONNABORTED;
+        }
     case WSAECONNRESET:
         return Errno::CONNRESET;
     case WSAEHOSTUNREACH:
@@ -198,7 +211,7 @@ bool EnableNonBlock(int fd, bool enable) {
     return fcntl(fd, F_SETFL, flags) == 0;
 }
 
-Errno TranslateNativeError(int e) {
+Errno TranslateNativeError(int e, CallType call_type = CallType::Other) {
     switch (e) {
     case 0:
         return Errno::SUCCESS;
@@ -208,6 +221,10 @@ Errno TranslateNativeError(int e) {
         return Errno::INVAL;
     case EMFILE:
         return Errno::MFILE;
+    case EPIPE:
+        return Errno::PIPE;
+    case ECONNABORTED:
+        return Errno::CONNABORTED;
     case ENOTCONN:
         return Errno::NOTCONN;
     case EAGAIN:
@@ -236,13 +253,13 @@ Errno TranslateNativeError(int e) {
 
 #endif
 
-Errno GetAndLogLastError() {
+Errno GetAndLogLastError(CallType call_type = CallType::Other) {
 #ifdef _WIN32
     int e = WSAGetLastError();
 #else
     int e = errno;
 #endif
-    const Errno err = TranslateNativeError(e);
+    const Errno err = TranslateNativeError(e, call_type);
     if (err == Errno::AGAIN || err == Errno::TIMEDOUT || err == Errno::INPROGRESS) {
         // These happen during normal operation, so only log them at debug level.
         LOG_DEBUG(Network, "Socket operation error: {}", Common::NativeErrorToString(e));
@@ -731,13 +748,17 @@ std::pair<s32, Errno> Socket::Send(std::span<const u8> message, int flags) {
     ASSERT(message.size() < static_cast<size_t>(std::numeric_limits<int>::max()));
     ASSERT(flags == 0);
 
+    int native_flags = 0;
+#if YUZU_UNIX
+    native_flags |= MSG_NOSIGNAL; // do not send us SIGPIPE
+#endif
     const auto result = send(fd, reinterpret_cast<const char*>(message.data()),
-                             static_cast<int>(message.size()), 0);
+                             static_cast<int>(message.size()), native_flags);
     if (result != SOCKET_ERROR) {
         return {static_cast<s32>(result), Errno::SUCCESS};
     }
 
-    return {-1, GetAndLogLastError()};
+    return {-1, GetAndLogLastError(CallType::Send)};
 }
 
 std::pair<s32, Errno> Socket::SendTo(u32 flags, std::span<const u8> message,
@@ -759,7 +780,7 @@ std::pair<s32, Errno> Socket::SendTo(u32 flags, std::span<const u8> message,
         return {static_cast<s32>(result), Errno::SUCCESS};
     }
 
-    return {-1, GetAndLogLastError()};
+    return {-1, GetAndLogLastError(CallType::Send)};
 }
 
 Errno Socket::Close() {

src/core/internal_network/network.h

@@ -33,10 +33,12 @@ enum class Errno {
     BADF,
     INVAL,
     MFILE,
+    PIPE,
     NOTCONN,
     AGAIN,
     CONNREFUSED,
     CONNRESET,
+    CONNABORTED,
     HOSTUNREACH,
     NETDOWN,
     NETUNREACH,

src/video_core/host_shaders/CMakeLists.txt

@@ -50,6 +50,7 @@ set(SHADER_FILES
     vulkan_blit_depth_stencil.frag
     vulkan_color_clear.frag
     vulkan_color_clear.vert
+    vulkan_depthstencil_clear.frag
     vulkan_fidelityfx_fsr_easu_fp16.comp
     vulkan_fidelityfx_fsr_easu_fp32.comp
     vulkan_fidelityfx_fsr_rcas_fp16.comp

src/video_core/host_shaders/vulkan_depthstencil_clear.frag

@@ -0,0 +1,12 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#version 460 core
+
+layout (push_constant) uniform PushConstants {
+    vec4 clear_depth;
+};
+
+void main() {
+    gl_FragDepth = clear_depth.x;
+}

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -1335,7 +1335,8 @@ 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 = VideoCommon::ObtainBufferOperation::DoNothing;
+    const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing
+                                         : VideoCommon::ObtainBufferOperation::MarkAsWritten;
     const auto [buffer, offset] =
         buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op);
 
@@ -1344,8 +1345,12 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     const std::span copy_span{&copy, 1};
 
     if constexpr (IS_IMAGE_UPLOAD) {
+        texture_cache.PrepareImage(image_id, true, false);
         image->UploadMemory(buffer->Handle(), offset, copy_span);
     } else {
+        if (offset % BytesPerBlock(image->info.format)) {
+            return false;
+        }
         texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span,
                                               buffer_operand.address, buffer_size);
     }

src/video_core/renderer_vulkan/blit_image.cpp

@@ -16,6 +16,7 @@
 #include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h"
 #include "video_core/host_shaders/vulkan_color_clear_frag_spv.h"
 #include "video_core/host_shaders/vulkan_color_clear_vert_spv.h"
+#include "video_core/host_shaders/vulkan_depthstencil_clear_frag_spv.h"
 #include "video_core/renderer_vulkan/blit_image.h"
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -428,6 +429,7 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
       blit_depth_stencil_frag(BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV)),
       clear_color_vert(BuildShader(device, VULKAN_COLOR_CLEAR_VERT_SPV)),
       clear_color_frag(BuildShader(device, VULKAN_COLOR_CLEAR_FRAG_SPV)),
+      clear_stencil_frag(BuildShader(device, VULKAN_DEPTHSTENCIL_CLEAR_FRAG_SPV)),
       convert_depth_to_float_frag(BuildShader(device, CONVERT_DEPTH_TO_FLOAT_FRAG_SPV)),
       convert_float_to_depth_frag(BuildShader(device, CONVERT_FLOAT_TO_DEPTH_FRAG_SPV)),
       convert_abgr8_to_d24s8_frag(BuildShader(device, CONVERT_ABGR8_TO_D24S8_FRAG_SPV)),
@@ -593,6 +595,28 @@ void BlitImageHelper::ClearColor(const Framebuffer* dst_framebuffer, u8 color_ma
     scheduler.InvalidateState();
 }
 
+void BlitImageHelper::ClearDepthStencil(const Framebuffer* dst_framebuffer, bool depth_clear,
+                                        f32 clear_depth, u8 stencil_mask, u32 stencil_ref,
+                                        u32 stencil_compare_mask, const Region2D& dst_region) {
+    const BlitDepthStencilPipelineKey key{
+        .renderpass = dst_framebuffer->RenderPass(),
+        .depth_clear = depth_clear,
+        .stencil_mask = stencil_mask,
+        .stencil_compare_mask = stencil_compare_mask,
+        .stencil_ref = stencil_ref,
+    };
+    const VkPipeline pipeline = FindOrEmplaceClearStencilPipeline(key);
+    const VkPipelineLayout layout = *clear_color_pipeline_layout;
+    scheduler.RequestRenderpass(dst_framebuffer);
+    scheduler.Record([pipeline, layout, clear_depth, dst_region](vk::CommandBuffer cmdbuf) {
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
+        BindBlitState(cmdbuf, dst_region);
+        cmdbuf.PushConstants(layout, VK_SHADER_STAGE_FRAGMENT_BIT, clear_depth);
+        cmdbuf.Draw(3, 1, 0, 0);
+    });
+    scheduler.InvalidateState();
+}
+
 void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
                               const ImageView& src_image_view) {
     const VkPipelineLayout layout = *one_texture_pipeline_layout;
@@ -820,6 +844,61 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel
     return *clear_color_pipelines.back();
 }
 
+VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
+    const BlitDepthStencilPipelineKey& key) {
+    const auto it = std::ranges::find(clear_stencil_keys, key);
+    if (it != clear_stencil_keys.end()) {
+        return *clear_stencil_pipelines[std::distance(clear_stencil_keys.begin(), it)];
+    }
+    clear_stencil_keys.push_back(key);
+    const std::array stages = MakeStages(*clear_color_vert, *clear_stencil_frag);
+    const auto stencil = VkStencilOpState{
+        .failOp = VK_STENCIL_OP_KEEP,
+        .passOp = VK_STENCIL_OP_REPLACE,
+        .depthFailOp = VK_STENCIL_OP_KEEP,
+        .compareOp = VK_COMPARE_OP_ALWAYS,
+        .compareMask = key.stencil_compare_mask,
+        .writeMask = key.stencil_mask,
+        .reference = key.stencil_ref,
+    };
+    const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .depthTestEnable = VK_FALSE,
+        .depthWriteEnable = key.depth_clear,
+        .depthCompareOp = VK_COMPARE_OP_ALWAYS,
+        .depthBoundsTestEnable = VK_FALSE,
+        .stencilTestEnable = VK_TRUE,
+        .front = stencil,
+        .back = stencil,
+        .minDepthBounds = 0.0f,
+        .maxDepthBounds = 0.0f,
+    };
+    clear_stencil_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
+        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .stageCount = static_cast<u32>(stages.size()),
+        .pStages = stages.data(),
+        .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
+        .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
+        .pTessellationState = nullptr,
+        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+        .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+        .pDepthStencilState = &depth_stencil_ci,
+        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
+        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+        .layout = *clear_color_pipeline_layout,
+        .renderPass = key.renderpass,
+        .subpass = 0,
+        .basePipelineHandle = VK_NULL_HANDLE,
+        .basePipelineIndex = 0,
+    }));
+    return *clear_stencil_pipelines.back();
+}
+
 void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass,
                                       bool is_target_depth) {
     if (pipeline) {

src/video_core/renderer_vulkan/blit_image.h

@@ -27,6 +27,16 @@ struct BlitImagePipelineKey {
     Tegra::Engines::Fermi2D::Operation operation;
 };
 
+struct BlitDepthStencilPipelineKey {
+    constexpr auto operator<=>(const BlitDepthStencilPipelineKey&) const noexcept = default;
+
+    VkRenderPass renderpass;
+    bool depth_clear;
+    u8 stencil_mask;
+    u32 stencil_compare_mask;
+    u32 stencil_ref;
+};
+
 class BlitImageHelper {
 public:
     explicit BlitImageHelper(const Device& device, Scheduler& scheduler,
@@ -64,6 +74,10 @@ public:
     void ClearColor(const Framebuffer* dst_framebuffer, u8 color_mask,
                     const std::array<f32, 4>& clear_color, const Region2D& dst_region);
 
+    void ClearDepthStencil(const Framebuffer* dst_framebuffer, bool depth_clear, f32 clear_depth,
+                           u8 stencil_mask, u32 stencil_ref, u32 stencil_compare_mask,
+                           const Region2D& dst_region);
+
 private:
     void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
                  const ImageView& src_image_view);
@@ -76,6 +90,8 @@ private:
     [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key);
 
     [[nodiscard]] VkPipeline FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key);
+    [[nodiscard]] VkPipeline FindOrEmplaceClearStencilPipeline(
+        const BlitDepthStencilPipelineKey& key);
 
     void ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass, bool is_target_depth);
 
@@ -108,6 +124,7 @@ private:
     vk::ShaderModule blit_depth_stencil_frag;
     vk::ShaderModule clear_color_vert;
     vk::ShaderModule clear_color_frag;
+    vk::ShaderModule clear_stencil_frag;
     vk::ShaderModule convert_depth_to_float_frag;
     vk::ShaderModule convert_float_to_depth_frag;
     vk::ShaderModule convert_abgr8_to_d24s8_frag;
@@ -122,6 +139,8 @@ private:
     std::vector<vk::Pipeline> blit_depth_stencil_pipelines;
     std::vector<BlitImagePipelineKey> clear_color_keys;
     std::vector<vk::Pipeline> clear_color_pipelines;
+    std::vector<BlitDepthStencilPipelineKey> clear_stencil_keys;
+    std::vector<vk::Pipeline> clear_stencil_pipelines;
     vk::Pipeline convert_d32_to_r32_pipeline;
     vk::Pipeline convert_r32_to_d32_pipeline;
     vk::Pipeline convert_d16_to_r16_pipeline;

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -126,7 +126,7 @@ struct FormatTuple {
     {VK_FORMAT_A1R5G5B5_UNORM_PACK16, Attachable},              // A1R5G5B5_UNORM
     {VK_FORMAT_A2B10G10R10_UNORM_PACK32, Attachable | Storage}, // A2B10G10R10_UNORM
     {VK_FORMAT_A2B10G10R10_UINT_PACK32, Attachable | Storage},  // A2B10G10R10_UINT
-    {VK_FORMAT_A2R10G10B10_UNORM_PACK32, Attachable | Storage}, // A2R10G10B10_UNORM
+    {VK_FORMAT_A2R10G10B10_UNORM_PACK32, Attachable},           // A2R10G10B10_UNORM
     {VK_FORMAT_A1R5G5B5_UNORM_PACK16, Attachable},         // A1B5G5R5_UNORM (flipped with swizzle)
     {VK_FORMAT_R5G5B5A1_UNORM_PACK16},                     // A5B5G5R1_UNORM (specially swizzled)
     {VK_FORMAT_R8_UNORM, Attachable | Storage},            // R8_UNORM

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -428,6 +428,17 @@ void RasterizerVulkan::Clear(u32 layer_count) {
     if (aspect_flags == 0) {
         return;
     }
+
+    if (use_stencil && regs.stencil_front_mask != 0xFF && regs.stencil_front_mask != 0) {
+        Region2D dst_region = {
+            Offset2D{.x = clear_rect.rect.offset.x, .y = clear_rect.rect.offset.y},
+            Offset2D{.x = clear_rect.rect.offset.x + static_cast<s32>(clear_rect.rect.extent.width),
+                     .y = clear_rect.rect.offset.y +
+                          static_cast<s32>(clear_rect.rect.extent.height)}};
+        blit_image.ClearDepthStencil(framebuffer, use_depth, regs.clear_depth,
+                                     static_cast<u8>(regs.stencil_front_mask), regs.clear_stencil,
+                                     regs.stencil_front_func_mask, dst_region);
+    } else {
         scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil,
                           clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) {
             VkClearAttachment attachment;
@@ -437,6 +448,7 @@ void RasterizerVulkan::Clear(u32 layer_count) {
             attachment.clearValue.depthStencil.stencil = clear_stencil;
             cmdbuf.ClearAttachments(attachment, clear_rect);
         });
+    }
 }
 
 void RasterizerVulkan::DispatchCompute() {
@@ -830,7 +842,8 @@ 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 = VideoCommon::ObtainBufferOperation::DoNothing;
+    const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing
+                                         : VideoCommon::ObtainBufferOperation::MarkAsWritten;
     const auto [buffer, offset] =
         buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op);
 
@@ -839,8 +852,12 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     const std::span copy_span{&copy, 1};
 
     if constexpr (IS_IMAGE_UPLOAD) {
+        texture_cache.PrepareImage(image_id, true, false);
         image->UploadMemory(buffer->Handle(), offset, copy_span);
     } else {
+        if (offset % BytesPerBlock(image->info.format)) {
+            return false;
+        }
         texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span,
                                               buffer_operand.address, buffer_size);
     }

src/video_core/texture_cache/texture_cache_base.h

@@ -243,6 +243,9 @@ public:
     /// Create channel state.
     void CreateChannel(Tegra::Control::ChannelState& channel) final override;
 
+    /// Prepare an image to be used
+    void PrepareImage(ImageId image_id, bool is_modification, bool invalidate);
+
     std::recursive_mutex mutex;
 
 private:
@@ -387,9 +390,6 @@ private:
     /// Synchronize image aliases, copying data if needed
     void SynchronizeAliases(ImageId image_id);
 
-    /// Prepare an image to be used
-    void PrepareImage(ImageId image_id, bool is_modification, bool invalidate);
-
     /// Prepare an image view to be used
     void PrepareImageView(ImageViewId image_view_id, bool is_modification, bool invalidate);
 

src/video_core/vulkan_common/vulkan_device.cpp

@@ -71,6 +71,11 @@ constexpr std::array R8G8B8_SSCALED{
     VK_FORMAT_UNDEFINED,
 };
 
+constexpr std::array VK_FORMAT_R32G32B32_SFLOAT{
+    VK_FORMAT_R32G32B32A32_SFLOAT,
+    VK_FORMAT_UNDEFINED,
+};
+
 } // namespace Alternatives
 
 enum class NvidiaArchitecture {
@@ -103,6 +108,8 @@ constexpr const VkFormat* GetFormatAlternatives(VkFormat format) {
         return Alternatives::R16G16B16_SSCALED.data();
     case VK_FORMAT_R8G8B8_SSCALED:
         return Alternatives::R8G8B8_SSCALED.data();
+    case VK_FORMAT_R32G32B32_SFLOAT:
+        return Alternatives::VK_FORMAT_R32G32B32_SFLOAT.data();
     default:
         return nullptr;
     }
@@ -130,6 +137,7 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(vk::Physica
         VK_FORMAT_A2B10G10R10_UINT_PACK32,
         VK_FORMAT_A2B10G10R10_UNORM_PACK32,
         VK_FORMAT_A2B10G10R10_USCALED_PACK32,
+        VK_FORMAT_A2R10G10B10_UNORM_PACK32,
         VK_FORMAT_A8B8G8R8_SINT_PACK32,
         VK_FORMAT_A8B8G8R8_SNORM_PACK32,
         VK_FORMAT_A8B8G8R8_SRGB_PACK32,