Reimplement the invalidate_texture_data_cache register

settings: added missing graphical settings to RestoreGlobalState()

src/core/hid/emulated_controller.cpp

@@ -957,7 +957,7 @@ void EmulatedController::SetMotion(const Common::Input::CallbackStatus& callback
         raw_status.gyro.y.value,
         raw_status.gyro.z.value,
     });
-    emulated.SetGyroThreshold(raw_status.gyro.x.properties.threshold);
+    emulated.SetUserGyroThreshold(raw_status.gyro.x.properties.threshold);
     emulated.UpdateRotation(raw_status.delta_timestamp);
     emulated.UpdateOrientation(raw_status.delta_timestamp);
     force_update_motion = raw_status.force_update;
@@ -1284,6 +1284,26 @@ void EmulatedController::SetLedPattern() {
     }
 }
 
+void EmulatedController::SetGyroscopeZeroDriftMode(GyroscopeZeroDriftMode mode) {
+    for (auto& motion : controller.motion_values) {
+        switch (mode) {
+        case GyroscopeZeroDriftMode::Loose:
+            motion_sensitivity = motion.emulated.IsAtRestLoose;
+            motion.emulated.SetGyroThreshold(motion.emulated.ThresholdLoose);
+            break;
+        case GyroscopeZeroDriftMode::Tight:
+            motion_sensitivity = motion.emulated.IsAtRestThight;
+            motion.emulated.SetGyroThreshold(motion.emulated.ThresholdThight);
+            break;
+        case GyroscopeZeroDriftMode::Standard:
+        default:
+            motion_sensitivity = motion.emulated.IsAtRestStandard;
+            motion.emulated.SetGyroThreshold(motion.emulated.ThresholdStandard);
+            break;
+        }
+    }
+}
+
 void EmulatedController::SetSupportedNpadStyleTag(NpadStyleTag supported_styles) {
     supported_style_tag = supported_styles;
     if (!is_connected) {

src/core/hid/emulated_controller.h

@@ -398,6 +398,9 @@ public:
     /// Asks the output device to change the player led pattern
     void SetLedPattern();
 
+    /// Changes sensitivity of the motion sensor
+    void SetGyroscopeZeroDriftMode(GyroscopeZeroDriftMode mode);
+
     /**
      * Adds a callback to the list of events
      * @param update_callback A ConsoleUpdateCallback that will be triggered
@@ -523,7 +526,7 @@ private:
     bool is_connected{false};
     bool is_configuring{false};
     bool system_buttons_enabled{true};
-    f32 motion_sensitivity{0.01f};
+    f32 motion_sensitivity{Core::HID::MotionInput::IsAtRestStandard};
     bool force_update_motion{false};
     u32 turbo_button_state{0};
 

src/core/hid/hid_types.h

@@ -282,6 +282,13 @@ enum class VibrationGcErmCommand : u64 {
     StopHard = 2,
 };
 
+// This is nn::hid::GyroscopeZeroDriftMode
+enum class GyroscopeZeroDriftMode : u32 {
+    Loose = 0,
+    Standard = 1,
+    Tight = 2,
+};
+
 // This is nn::hid::NpadStyleTag
 struct NpadStyleTag {
     union {

src/core/hid/motion_input.cpp

@@ -9,7 +9,7 @@ namespace Core::HID {
 MotionInput::MotionInput() {
     // Initialize PID constants with default values
     SetPID(0.3f, 0.005f, 0.0f);
-    SetGyroThreshold(0.007f);
+    SetGyroThreshold(ThresholdStandard);
 }
 
 void MotionInput::SetPID(f32 new_kp, f32 new_ki, f32 new_kd) {
@@ -26,11 +26,11 @@ void MotionInput::SetGyroscope(const Common::Vec3f& gyroscope) {
     gyro = gyroscope - gyro_bias;
 
     // Auto adjust drift to minimize drift
-    if (!IsMoving(0.1f)) {
+    if (!IsMoving(IsAtRestRelaxed)) {
         gyro_bias = (gyro_bias * 0.9999f) + (gyroscope * 0.0001f);
     }
 
-    if (gyro.Length() < gyro_threshold) {
+    if (gyro.Length() < gyro_threshold * user_gyro_threshold) {
         gyro = {};
     } else {
         only_accelerometer = false;
@@ -49,6 +49,10 @@ void MotionInput::SetGyroThreshold(f32 threshold) {
     gyro_threshold = threshold;
 }
 
+void MotionInput::SetUserGyroThreshold(f32 threshold) {
+    user_gyro_threshold = threshold / ThresholdStandard;
+}
+
 void MotionInput::EnableReset(bool reset) {
     reset_enabled = reset;
 }
@@ -208,7 +212,7 @@ void MotionInput::ResetOrientation() {
     if (!reset_enabled || only_accelerometer) {
         return;
     }
-    if (!IsMoving(0.5f) && accel.z <= -0.9f) {
+    if (!IsMoving(IsAtRestRelaxed) && accel.z <= -0.9f) {
         ++reset_counter;
         if (reset_counter > 900) {
             quat.w = 0;

src/core/hid/motion_input.h

@@ -11,6 +11,15 @@ namespace Core::HID {
 
 class MotionInput {
 public:
+    static constexpr float ThresholdLoose = 0.01f;
+    static constexpr float ThresholdStandard = 0.007f;
+    static constexpr float ThresholdThight = 0.002f;
+
+    static constexpr float IsAtRestRelaxed = 0.05f;
+    static constexpr float IsAtRestLoose = 0.02f;
+    static constexpr float IsAtRestStandard = 0.01f;
+    static constexpr float IsAtRestThight = 0.005f;
+
     explicit MotionInput();
 
     MotionInput(const MotionInput&) = default;
@@ -26,6 +35,9 @@ public:
     void SetGyroBias(const Common::Vec3f& bias);
     void SetGyroThreshold(f32 threshold);
 
+    /// Applies a modifier on top of the normal gyro threshold
+    void SetUserGyroThreshold(f32 threshold);
+
     void EnableReset(bool reset);
     void ResetRotations();
 
@@ -74,6 +86,9 @@ private:
     // Minimum gyro amplitude to detect if the device is moving
     f32 gyro_threshold = 0.0f;
 
+    // Multiplies gyro_threshold by this value
+    f32 user_gyro_threshold = 0.0f;
+
     // Number of invalid sequential data
     u32 reset_counter = 0;
 

src/core/hle/service/hid/controllers/npad.cpp

@@ -1132,7 +1132,8 @@ Result Controller_NPad::DisconnectNpad(Core::HID::NpadIdType npad_id) {
     return ResultSuccess;
 }
 Result Controller_NPad::SetGyroscopeZeroDriftMode(
-    const Core::HID::SixAxisSensorHandle& sixaxis_handle, GyroscopeZeroDriftMode drift_mode) {
+    const Core::HID::SixAxisSensorHandle& sixaxis_handle,
+    Core::HID::GyroscopeZeroDriftMode drift_mode) {
     const auto is_valid = VerifyValidSixAxisSensorHandle(sixaxis_handle);
     if (is_valid.IsError()) {
         LOG_ERROR(Service_HID, "Invalid handle, error_code={}", is_valid.raw);
@@ -1140,14 +1141,16 @@ Result Controller_NPad::SetGyroscopeZeroDriftMode(
     }
 
     auto& sixaxis = GetSixaxisState(sixaxis_handle);
+    auto& controller = GetControllerFromHandle(sixaxis_handle);
     sixaxis.gyroscope_zero_drift_mode = drift_mode;
+    controller.device->SetGyroscopeZeroDriftMode(drift_mode);
 
     return ResultSuccess;
 }
 
 Result Controller_NPad::GetGyroscopeZeroDriftMode(
     const Core::HID::SixAxisSensorHandle& sixaxis_handle,
-    GyroscopeZeroDriftMode& drift_mode) const {
+    Core::HID::GyroscopeZeroDriftMode& drift_mode) const {
     const auto is_valid = VerifyValidSixAxisSensorHandle(sixaxis_handle);
     if (is_valid.IsError()) {
         LOG_ERROR(Service_HID, "Invalid handle, error_code={}", is_valid.raw);

src/core/hle/service/hid/controllers/npad.h

@@ -52,13 +52,6 @@ public:
     // When the controller is requesting a motion update for the shared memory
     void OnMotionUpdate(const Core::Timing::CoreTiming& core_timing) override;
 
-    // This is nn::hid::GyroscopeZeroDriftMode
-    enum class GyroscopeZeroDriftMode : u32 {
-        Loose = 0,
-        Standard = 1,
-        Tight = 2,
-    };
-
     // This is nn::hid::NpadJoyHoldType
     enum class NpadJoyHoldType : u64 {
         Vertical = 0,
@@ -146,9 +139,9 @@ public:
     Result DisconnectNpad(Core::HID::NpadIdType npad_id);
 
     Result SetGyroscopeZeroDriftMode(const Core::HID::SixAxisSensorHandle& sixaxis_handle,
-                                     GyroscopeZeroDriftMode drift_mode);
+                                     Core::HID::GyroscopeZeroDriftMode drift_mode);
     Result GetGyroscopeZeroDriftMode(const Core::HID::SixAxisSensorHandle& sixaxis_handle,
-                                     GyroscopeZeroDriftMode& drift_mode) const;
+                                     Core::HID::GyroscopeZeroDriftMode& drift_mode) const;
     Result IsSixAxisSensorAtRest(const Core::HID::SixAxisSensorHandle& sixaxis_handle,
                                  bool& is_at_rest) const;
     Result IsFirmwareUpdateAvailableForSixAxisSensor(
@@ -489,7 +482,8 @@ private:
         Core::HID::SixAxisSensorFusionParameters fusion{};
         Core::HID::SixAxisSensorCalibrationParameter calibration{};
         Core::HID::SixAxisSensorIcInformation ic_information{};
-        GyroscopeZeroDriftMode gyroscope_zero_drift_mode{GyroscopeZeroDriftMode::Standard};
+        Core::HID::GyroscopeZeroDriftMode gyroscope_zero_drift_mode{
+            Core::HID::GyroscopeZeroDriftMode::Standard};
     };
 
     struct NpadControllerData {

src/core/hle/service/hid/hid.cpp

@@ -712,7 +712,7 @@ void Hid::ResetSixAxisSensorFusionParameters(Kernel::HLERequestContext& ctx) {
 void Hid::SetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     const auto sixaxis_handle{rp.PopRaw<Core::HID::SixAxisSensorHandle>()};
-    const auto drift_mode{rp.PopEnum<Controller_NPad::GyroscopeZeroDriftMode>()};
+    const auto drift_mode{rp.PopEnum<Core::HID::GyroscopeZeroDriftMode>()};
     const auto applet_resource_user_id{rp.Pop<u64>()};
 
     auto& controller = GetAppletResource()->GetController<Controller_NPad>(HidController::NPad);
@@ -739,7 +739,7 @@ void Hid::GetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx) {
 
     const auto parameters{rp.PopRaw<Parameters>()};
 
-    auto drift_mode{Controller_NPad::GyroscopeZeroDriftMode::Standard};
+    auto drift_mode{Core::HID::GyroscopeZeroDriftMode::Standard};
     auto& controller = GetAppletResource()->GetController<Controller_NPad>(HidController::NPad);
     const auto result = controller.GetGyroscopeZeroDriftMode(parameters.sixaxis_handle, drift_mode);
 
@@ -764,7 +764,7 @@ void Hid::ResetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx) {
 
     const auto parameters{rp.PopRaw<Parameters>()};
 
-    const auto drift_mode{Controller_NPad::GyroscopeZeroDriftMode::Standard};
+    const auto drift_mode{Core::HID::GyroscopeZeroDriftMode::Standard};
     auto& controller = GetAppletResource()->GetController<Controller_NPad>(HidController::NPad);
     const auto result = controller.SetGyroscopeZeroDriftMode(parameters.sixaxis_handle, drift_mode);
 

src/video_core/engines/maxwell_3d.cpp

@@ -186,6 +186,7 @@ bool Maxwell3D::IsMethodExecutable(u32 method) {
     case MAXWELL3D_REG_INDEX(launch_dma):
     case MAXWELL3D_REG_INDEX(inline_data):
     case MAXWELL3D_REG_INDEX(fragment_barrier):
+    case MAXWELL3D_REG_INDEX(invalidate_texture_data_cache):
     case MAXWELL3D_REG_INDEX(tiled_cache_barrier):
         return true;
     default:
@@ -375,6 +376,9 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
         return;
     case MAXWELL3D_REG_INDEX(fragment_barrier):
         return rasterizer->FragmentBarrier();
+    case MAXWELL3D_REG_INDEX(invalidate_texture_data_cache):
+        rasterizer->InvalidateGPUCache();
+        return rasterizer->WaitForIdle();
     case MAXWELL3D_REG_INDEX(tiled_cache_barrier):
         return rasterizer->TiledCacheBarrier();
     default:

src/video_core/host_shaders/CMakeLists.txt

@@ -22,6 +22,8 @@ set(SHADER_FILES
     convert_d24s8_to_abgr8.frag
     convert_depth_to_float.frag
     convert_float_to_depth.frag
+    convert_msaa_to_non_msaa.comp
+    convert_non_msaa_to_msaa.comp
     convert_s8d24_to_abgr8.frag
     full_screen_triangle.vert
     fxaa.frag

src/video_core/host_shaders/convert_msaa_to_non_msaa.comp

@@ -0,0 +1,30 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#version 450 core
+layout (local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+layout (binding = 0, rgba8) uniform readonly restrict image2DMSArray msaa_in;
+layout (binding = 1, rgba8) uniform writeonly restrict image2DArray output_img;
+
+void main() {
+    const ivec3 coords = ivec3(gl_GlobalInvocationID);
+    if (any(greaterThanEqual(coords, imageSize(msaa_in)))) {
+        return;
+    }
+
+    // TODO: Specialization constants for num_samples?
+    const int num_samples = imageSamples(msaa_in);
+    for (int curr_sample = 0; curr_sample < num_samples; ++curr_sample) {
+        const vec4 pixel = imageLoad(msaa_in, coords, curr_sample);
+
+        const int single_sample_x = 2 * coords.x + (curr_sample & 1);
+        const int single_sample_y = 2 * coords.y + ((curr_sample / 2) & 1);
+        const ivec3 dest_coords = ivec3(single_sample_x, single_sample_y, coords.z);
+
+        if (any(greaterThanEqual(dest_coords, imageSize(output_img)))) {
+            continue;
+        }
+        imageStore(output_img, dest_coords, pixel);
+    }
+}

src/video_core/host_shaders/convert_non_msaa_to_msaa.comp

@@ -0,0 +1,29 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#version 450 core
+layout (local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+layout (binding = 0, rgba8) uniform readonly restrict image2DArray img_in;
+layout (binding = 1, rgba8) uniform writeonly restrict image2DMSArray output_msaa;
+
+void main() {
+    const ivec3 coords = ivec3(gl_GlobalInvocationID);
+    if (any(greaterThanEqual(coords, imageSize(output_msaa)))) {
+        return;
+    }
+
+    // TODO: Specialization constants for num_samples?
+    const int num_samples = imageSamples(output_msaa);
+    for (int curr_sample = 0; curr_sample < num_samples; ++curr_sample) {
+        const int single_sample_x = 2 * coords.x + (curr_sample & 1);
+        const int single_sample_y = 2 * coords.y + ((curr_sample / 2) & 1);
+        const ivec3 single_coords = ivec3(single_sample_x, single_sample_y, coords.z);
+
+        if (any(greaterThanEqual(single_coords, imageSize(img_in)))) {
+            continue;
+        }
+        const vec4 pixel = imageLoad(img_in, single_coords);
+        imageStore(output_msaa, coords, curr_sample, pixel);
+    }
+}

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -557,6 +557,14 @@ void TextureCacheRuntime::CopyImage(Image& dst_image, Image& src_image,
     }
 }
 
+void TextureCacheRuntime::CopyImageMSAA(Image& dst_image, Image& src_image,
+                                        std::span<const VideoCommon::ImageCopy> copies) {
+    LOG_DEBUG(Render_OpenGL, "Copying from {} samples to {} samples", src_image.info.num_samples,
+              dst_image.info.num_samples);
+    // TODO: Leverage the format conversion pass if possible/accurate.
+    util_shaders.CopyMSAA(dst_image, src_image, copies);
+}
+
 void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src,
                                            std::span<const VideoCommon::ImageCopy> copies) {
     LOG_DEBUG(Render_OpenGL, "Converting {} to {}", src.info.format, dst.info.format);

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -93,12 +93,19 @@ public:
         return device.CanReportMemoryUsage();
     }
 
-    bool ShouldReinterpret([[maybe_unused]] Image& dst, [[maybe_unused]] Image& src) {
+    bool ShouldReinterpret([[maybe_unused]] Image& dst,
+                           [[maybe_unused]] Image& src) const noexcept {
+        return true;
+    }
+
+    bool CanUploadMSAA() const noexcept {
         return true;
     }
 
     void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
 
+    void CopyImageMSAA(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
+
     void ReinterpretImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
 
     void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view) {

src/video_core/renderer_opengl/util_shaders.cpp

@@ -12,6 +12,8 @@
 #include "video_core/host_shaders/astc_decoder_comp.h"
 #include "video_core/host_shaders/block_linear_unswizzle_2d_comp.h"
 #include "video_core/host_shaders/block_linear_unswizzle_3d_comp.h"
+#include "video_core/host_shaders/convert_msaa_to_non_msaa_comp.h"
+#include "video_core/host_shaders/convert_non_msaa_to_msaa_comp.h"
 #include "video_core/host_shaders/opengl_convert_s8d24_comp.h"
 #include "video_core/host_shaders/opengl_copy_bc4_comp.h"
 #include "video_core/host_shaders/pitch_unswizzle_comp.h"
@@ -51,7 +53,9 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_)
       block_linear_unswizzle_3d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_3D_COMP)),
       pitch_unswizzle_program(MakeProgram(PITCH_UNSWIZZLE_COMP)),
       copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)),
-      convert_s8d24_program(MakeProgram(OPENGL_CONVERT_S8D24_COMP)) {
+      convert_s8d24_program(MakeProgram(OPENGL_CONVERT_S8D24_COMP)),
+      convert_ms_to_nonms_program(MakeProgram(CONVERT_MSAA_TO_NON_MSAA_COMP)),
+      convert_nonms_to_ms_program(MakeProgram(CONVERT_NON_MSAA_TO_MSAA_COMP)) {
     const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
     swizzle_table_buffer.Create();
     glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
@@ -269,6 +273,33 @@ void UtilShaders::ConvertS8D24(Image& dst_image, std::span<const ImageCopy> copi
     program_manager.RestoreGuestCompute();
 }
 
+void UtilShaders::CopyMSAA(Image& dst_image, Image& src_image,
+                           std::span<const VideoCommon::ImageCopy> copies) {
+    const bool is_ms_to_non_ms = src_image.info.num_samples > 1 && dst_image.info.num_samples == 1;
+    const auto program_handle =
+        is_ms_to_non_ms ? convert_ms_to_nonms_program.handle : convert_nonms_to_ms_program.handle;
+    program_manager.BindComputeProgram(program_handle);
+
+    for (const ImageCopy& copy : copies) {
+        ASSERT(copy.src_subresource.base_layer == 0);
+        ASSERT(copy.src_subresource.num_layers == 1);
+        ASSERT(copy.dst_subresource.base_layer == 0);
+        ASSERT(copy.dst_subresource.num_layers == 1);
+
+        glBindImageTexture(0, src_image.StorageHandle(), copy.src_subresource.base_level, GL_TRUE,
+                           0, GL_READ_ONLY, GL_RGBA8);
+        glBindImageTexture(1, dst_image.StorageHandle(), copy.dst_subresource.base_level, GL_TRUE,
+                           0, GL_WRITE_ONLY, GL_RGBA8);
+
+        const u32 num_dispatches_x = Common::DivCeil(copy.extent.width, 8U);
+        const u32 num_dispatches_y = Common::DivCeil(copy.extent.height, 8U);
+        const u32 num_dispatches_z = copy.extent.depth;
+
+        glDispatchCompute(num_dispatches_x, num_dispatches_y, num_dispatches_z);
+    }
+    program_manager.RestoreGuestCompute();
+}
+
 GLenum StoreFormat(u32 bytes_per_block) {
     switch (bytes_per_block) {
     case 1:

src/video_core/renderer_opengl/util_shaders.h

@@ -40,6 +40,9 @@ public:
 
     void ConvertS8D24(Image& dst_image, std::span<const VideoCommon::ImageCopy> copies);
 
+    void CopyMSAA(Image& dst_image, Image& src_image,
+                  std::span<const VideoCommon::ImageCopy> copies);
+
 private:
     ProgramManager& program_manager;
 
@@ -51,6 +54,8 @@ private:
     OGLProgram pitch_unswizzle_program;
     OGLProgram copy_bc4_program;
     OGLProgram convert_s8d24_program;
+    OGLProgram convert_ms_to_nonms_program;
+    OGLProgram convert_nonms_to_ms_program;
 };
 
 GLenum StoreFormat(u32 bytes_per_block);

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -1230,6 +1230,11 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
     });
 }
 
+void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
+                                        std::span<const VideoCommon::ImageCopy> copies) {
+    UNIMPLEMENTED_MSG("Copying images with different samples is not implemented in Vulkan.");
+}
+
 u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
     return device.GetDeviceLocalMemory();
 }

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -70,6 +70,8 @@ public:
 
     void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
 
+    void CopyImageMSAA(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
+
     bool ShouldReinterpret(Image& dst, Image& src);
 
     void ReinterpretImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
@@ -80,6 +82,11 @@ public:
         return false;
     }
 
+    bool CanUploadMSAA() const noexcept {
+        // TODO: Implement buffer to MSAA uploads
+        return false;
+    }
+
     void AccelerateImageUpload(Image&, const StagingBufferRef&,
                                std::span<const VideoCommon::SwizzleParameters>);
 

src/video_core/texture_cache/formatter.cpp

@@ -22,6 +22,9 @@ std::string Name(const ImageBase& image) {
     const u32 num_layers = image.info.resources.layers;
     const u32 num_levels = image.info.resources.levels;
     std::string resource;
+    if (image.info.num_samples > 1) {
+        resource += fmt::format(":{}xMSAA", image.info.num_samples);
+    }
     if (num_layers > 1) {
         resource += fmt::format(":L{}", num_layers);
     }

src/video_core/texture_cache/texture_cache.h

@@ -773,7 +773,7 @@ void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {
     image.flags &= ~ImageFlagBits::CpuModified;
     TrackImage(image, image_id);
 
-    if (image.info.num_samples > 1) {
+    if (image.info.num_samples > 1 && !runtime.CanUploadMSAA()) {
         LOG_WARNING(HW_GPU, "MSAA image uploads are not implemented");
         return;
     }
@@ -1167,14 +1167,14 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
         if (True(overlap.flags & ImageFlagBits::GpuModified)) {
             new_image.flags |= ImageFlagBits::GpuModified;
         }
+        const auto& resolution = Settings::values.resolution_info;
+        const SubresourceBase base = new_image.TryFindBase(overlap.gpu_addr).value();
+        const u32 up_scale = can_rescale ? resolution.up_scale : 1;
+        const u32 down_shift = can_rescale ? resolution.down_shift : 0;
+        auto copies = MakeShrinkImageCopies(new_info, overlap.info, base, up_scale, down_shift);
         if (overlap.info.num_samples != new_image.info.num_samples) {
-            LOG_WARNING(HW_GPU, "Copying between images with different samples is not implemented");
+            runtime.CopyImageMSAA(new_image, overlap, std::move(copies));
         } else {
-            const auto& resolution = Settings::values.resolution_info;
-            const SubresourceBase base = new_image.TryFindBase(overlap.gpu_addr).value();
-            const u32 up_scale = can_rescale ? resolution.up_scale : 1;
-            const u32 down_shift = can_rescale ? resolution.down_shift : 0;
-            auto copies = MakeShrinkImageCopies(new_info, overlap.info, base, up_scale, down_shift);
             runtime.CopyImage(new_image, overlap, std::move(copies));
         }
         if (True(overlap.flags & ImageFlagBits::Tracked)) {

src/video_core/texture_cache/util.cpp

@@ -573,10 +573,6 @@ u32 CalculateUnswizzledSizeBytes(const ImageInfo& info) noexcept {
     if (info.type == ImageType::Buffer) {
         return info.size.width * BytesPerBlock(info.format);
     }
-    if (info.num_samples > 1) {
-        // Multisample images can't be uploaded or downloaded to the host
-        return 0;
-    }
     if (info.type == ImageType::Linear) {
         return info.pitch * Common::DivCeil(info.size.height, DefaultBlockHeight(info.format));
     }
@@ -703,7 +699,6 @@ ImageViewType RenderTargetImageViewType(const ImageInfo& info) noexcept {
 std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageInfo& src,
                                              SubresourceBase base, u32 up_scale, u32 down_shift) {
     ASSERT(dst.resources.levels >= src.resources.levels);
-    ASSERT(dst.num_samples == src.num_samples);
 
     const bool is_dst_3d = dst.type == ImageType::e3D;
     if (is_dst_3d) {