[TEST] Quick change on depth's handling

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt

@@ -30,6 +30,7 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
     RENDERER_REACTIVE_FLUSHING("use_reactive_flushing"),
     ENABLE_BUFFER_HISTORY("enable_buffer_history"),
     USE_OPTIMIZED_VERTEX_BUFFERS("use_optimized_vertex_buffers"),
+    ENABLE_GPU_BUFFER_READBACK("enable_gpu_buffer_readback"),
     SYNC_MEMORY_OPERATIONS("sync_memory_operations"),
     BUFFER_REORDER_DISABLE("disable_buffer_reorder"),
     RENDERER_DEBUG("debug"),

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt

@@ -586,6 +586,7 @@ abstract class SettingsItem(
                     IntSetting.FSR_SHARPENING_SLIDER,
                     titleId = R.string.fsr_sharpness,
                     descriptionId = R.string.fsr_sharpness_description,
+                    max = 200,
                     units = "%"
                 )
             )
@@ -808,6 +809,13 @@ abstract class SettingsItem(
                     descriptionId = R.string.enable_buffer_history_description
                 )
             )
+            put(
+                SwitchSetting(
+                    BooleanSetting.ENABLE_GPU_BUFFER_READBACK,
+                    titleId = R.string.enable_gpu_buffer_readback,
+                    descriptionId = R.string.enable_gpu_buffer_readback_description
+                )
+            )
             put(
                 SwitchSetting(
                     BooleanSetting.USE_OPTIMIZED_VERTEX_BUFFERS,

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt

@@ -294,6 +294,7 @@ class SettingsFragmentPresenter(
             add(BooleanSetting.RENDERER_FORCE_MAX_CLOCK.key)
             add(BooleanSetting.RENDERER_REACTIVE_FLUSHING.key)
             add(BooleanSetting.ENABLE_BUFFER_HISTORY.key)
+            add(BooleanSetting.ENABLE_GPU_BUFFER_READBACK.key)
             add(BooleanSetting.USE_OPTIMIZED_VERTEX_BUFFERS.key)
 
             add(HeaderSetting(R.string.hacks))

src/android/app/src/main/java/org/yuzu/yuzu_emu/fragments/EmulationFragment.kt

@@ -1182,7 +1182,7 @@ class EmulationFragment : Fragment(), SurfaceHolder.Callback {
                     container,
                     IntSetting.FSR_SHARPENING_SLIDER,
                     minValue = 0,
-                    maxValue = 100,
+                    maxValue = 200,
                     units = "%"
                 )
             }

src/android/app/src/main/res/values-ar/strings.xml

@@ -499,6 +499,8 @@
     <string name="renderer_reactive_flushing_description">يحسن دقة العرض في بعض الألعاب على حساب الأداء.</string>
     <string name="enable_buffer_history">تمكين سجل التخزين المؤقت</string>
     <string name="enable_buffer_history_description">يُتيح هذا الخيار الوصول إلى حالات التخزين المؤقت السابقة. وقد يُحسّن جودة العرض وثبات الأداء في بعض الألعاب.</string>
+    <string name="enable_gpu_buffer_readback">تفعيل قراءة مخزن وحدة معالجة الرسومات</string>
+    <string name="enable_gpu_buffer_readback_description">يحافظ هذا النظام على بيانات المخزن المؤقت المُعدّلة بواسطة وحدة معالجة الرسومات عن طريق قراءتها مرة أخرى قبل التحميل. تتطلب بعض الألعاب ذلك لعرض بعض التأثيرات بشكل صحيح. قد يُسبب ذلك مشاكل إذا لم يتمكن الجهاز من التعامل مع عبء العمل الإضافي.</string>
     <string name="use_optimized_vertex_buffers">مخازن الرؤوس المُحسّنة</string>
     <string name="use_optimized_vertex_buffers_description">يُتيح ربطًا مُحسَّنًا لمخازن الرؤوس لتحسين الأداء. يتطلب برامج تشغيل Mesa 26.0+ Turnip/ برامج تشغيل QCOM. قد يتعطل على برامج تشغيل Turnip القديمة (25.3 وما دون).</string>
 

src/android/app/src/main/res/values-es/strings.xml

@@ -491,6 +491,8 @@
     <string name="renderer_reactive_flushing_description">Mejora la precisión de renderizado en algunos juegos, pero reduce el rendimiento.</string>
     <string name="enable_buffer_history">Activar el historial del búfer</string>
     <string name="enable_buffer_history_description">Permite el acceso al estado del búfer anterior. Esta opción puede mejorar la calidad de renderizado y la consistencia en el rendimiento de algunos juegos.</string>
+    <string name="enable_gpu_buffer_readback">Activar la lectura del buffer de la GPU</string>
+    <string name="enable_gpu_buffer_readback_description">Conserva los datos del búfer modificados por la GPU leyéndolos antes de subirlos.\nAlgunos juegos requieren esto para renderizar correctamente ciertos efectos.\nPuede causar problemas si el hardware no puede soportar la carga de trabajo adicional.</string>
     <string name="use_optimized_vertex_buffers">Búferes de vértices optimizados</string>
     <string name="use_optimized_vertex_buffers_description">Permite la optimización del enlace del búfer de vértices para un mejor rendimiento. Requiere controladores Mesa 26.0+ Turnip/ controladores QCOM. Fallará con controladores Turnip más antiguos (versión 25.3 o inferior).</string>
 

src/android/app/src/main/res/values-zh-rCN/strings.xml

@@ -489,6 +489,8 @@
     <string name="renderer_reactive_flushing_description">通过牺牲性能来提升某些游戏的渲染精度。</string>
     <string name="enable_buffer_history">启用缓冲区历史</string>
     <string name="enable_buffer_history_description">启用对先前缓冲区状态的访问。此选项可在某些游戏中提升渲染质量并保持性能的一致性。</string>
+    <string name="enable_gpu_buffer_readback">启用 GPU 缓冲区回读</string>
+    <string name="enable_gpu_buffer_readback_description">在上传前回读经由 GPU 修改过的缓冲区数据，以将其保留。一些游戏会用到这项设定以正确渲染某些效果。如果硬件无法处理额外的工作负载，则可能会导致问题。</string>
     <string name="use_optimized_vertex_buffers">优化顶点缓冲区</string>
     <string name="use_optimized_vertex_buffers_description">启用经过优化的顶点缓冲区绑定以提升性能。需要 Mesa 26.0 及以上版本的 Turnip 或 QCOM 驱动程序。若使用较旧版本的 Turnip 驱动 (25.3 及以下版本) 则会导致崩溃。</string>
 

src/android/app/src/main/res/values/strings.xml

@@ -505,6 +505,8 @@
     <string name="renderer_reactive_flushing_description">Improves rendering accuracy in some games at the cost of performance.</string>
     <string name="enable_buffer_history">Enable buffer history</string>
     <string name="enable_buffer_history_description">Enables access to previous buffer states. This option may improve rendering quality and performance consistency in some games.</string>
+    <string name="enable_gpu_buffer_readback">Enable GPU Buffer Readback</string>
+    <string name="enable_gpu_buffer_readback_description">Preserves GPU-modified buffer data by reading it back before uploads. Some games require this to render certain effects properly. May cause issues if the hardware cannot handle the additional workload.</string>
     <string name="use_optimized_vertex_buffers">Optimized Vertex Buffers</string>
     <string name="use_optimized_vertex_buffers_description">Enables optimized vertex buffer binding for improved performance. Requires Mesa 26.0+ Turnip drivers/ QCOM drivers. Will crash on older Turnip drivers (25.3 and below).</string>
 

src/common/settings.h

@@ -576,6 +576,13 @@ struct Values {
         false,
 #endif
         "rescale_hack", Category::RendererHacks};
+    SwitchableSetting<bool> enable_gpu_buffer_readback{linkage,
+                                                       false,
+                                                       "enable_gpu_buffer_readback",
+                                                       Category::RendererAdvanced,
+                                                       Specialization::Default,
+                                                       true,
+                                                       true};
 
     SwitchableSetting<bool> use_asynchronous_shaders{linkage, false, "use_asynchronous_shaders",
                                                      Category::RendererHacks};

src/core/hle/service/vi/application_display_service.cpp

@@ -88,25 +88,10 @@ Result IApplicationDisplayService::GetIndirectDisplayTransactionService(
 }
 
 Result IApplicationDisplayService::OpenDisplay(Out<u64> out_display_id, DisplayName display_name) {
-    LOG_DEBUG(Service_VI, "called with display_name={}", display_name.data());
-
     // Ensure the display name is null-terminated
     display_name[display_name.size() - 1] = '\0';
 
-    // According to switchbrew, only "Default", "External", "Edid", "Internal" and "Null" are valid
-    const std::array<std::string_view, 5> valid_names = {
-        "Default", "External", "Edid", "Internal", "Null"
-    };
-
-    bool valid_name = false;
-    for (const auto& name : valid_names) {
-        if (name == display_name.data()) {
-            valid_name = true;
-            break;
-        }
-    }
-
-    R_UNLESS(valid_name, ResultOperationFailed);
+    LOG_DEBUG(Service_VI, "called with display_name={}", display_name.data());
 
     R_RETURN(m_container->OpenDisplay(out_display_id, display_name));
 }

src/qt_common/config/shared_translation.cpp

@@ -223,6 +223,8 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
            tr("Controls the DMA read mode.\nUnsafe is faster, while Safe is more stable and can fix issues in some games.\nDefault follows the GPU Accuracy setting."));
     INSERT(Settings, gpu_fence_behavior, tr("GPU Fence Behavior:"),
            tr("Controls the GPU fence synchronization behavior.\nImmediate is the fastest option, but can introduce some issues.\nBalanced offers better compatibility and may fix issues in some games.\nAccurate further improves compatibility at the cost of some performance.\nStrict is the slowest option, but can fix issues that require stricter synchronization.\nDefault follows the GPU Accuracy setting."));
+    INSERT(Settings, enable_gpu_buffer_readback, tr("Enable GPU buffer readback"),
+           tr("Preserves GPU-modified data by reading it back before uploading.\nSome games require this to render certain effects properly."));
     INSERT(Settings, use_asynchronous_shaders, tr("Enable asynchronous shader compilation"),
            tr("May reduce shader stutter."));
     INSERT(Settings, gpu_clock, tr("GPU Clocks"),

src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp

@@ -50,12 +50,6 @@ constexpr std::array RGBA_LUT{
     R | G | B | A, //
 };
 
-void CheckAlignment(IR::Reg reg, size_t alignment) {
-    if (!IR::IsAligned(reg, alignment)) {
-        throw NotImplementedException("Unaligned source register {}", reg);
-    }
-}
-
 template <typename... Args>
 IR::Value Composite(TranslatorVisitor& v, Args... regs) {
     return v.ir.CompositeConstruct(v.F(regs)...);
@@ -86,67 +80,53 @@ IR::Value Sample(TranslatorVisitor& v, u64 insn) {
         info.type.Assign(TextureType::Color2D);
         return v.ir.ImageSampleExplicitLod(handle, Composite(v, reg_a, reg_b), zero, {}, info);
     case 3: // 2D.LL
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::Color2D);
         return v.ir.ImageSampleExplicitLod(handle, Composite(v, reg_a, reg_a + 1), v.F(reg_b), {},
                                            info);
     case 4: // 2D.DC
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::Color2D);
         info.is_depth.Assign(1);
         return v.ir.ImageSampleDrefImplicitLod(handle, Composite(v, reg_a, reg_a + 1), v.F(reg_b),
                                                {}, {}, {}, info);
     case 5: // 2D.LL.DC
-        CheckAlignment(reg_a, 2);
-        CheckAlignment(reg_b, 2);
         info.type.Assign(TextureType::Color2D);
         info.is_depth.Assign(1);
         return v.ir.ImageSampleDrefExplicitLod(handle, Composite(v, reg_a, reg_a + 1),
                                                v.F(reg_b + 1), v.F(reg_b), {}, info);
     case 6: // 2D.LZ.DC
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::Color2D);
         info.is_depth.Assign(1);
         return v.ir.ImageSampleDrefExplicitLod(handle, Composite(v, reg_a, reg_a + 1), v.F(reg_b),
                                                zero, {}, info);
     case 7: // ARRAY_2D
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::ColorArray2D);
         return v.ir.ImageSampleImplicitLod(
             handle, v.ir.CompositeConstruct(v.F(reg_a + 1), v.F(reg_b), ReadArray(v, v.X(reg_a))),
             {}, {}, {}, info);
     case 8: // ARRAY_2D.LZ
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::ColorArray2D);
         return v.ir.ImageSampleExplicitLod(
             handle, v.ir.CompositeConstruct(v.F(reg_a + 1), v.F(reg_b), ReadArray(v, v.X(reg_a))),
             zero, {}, info);
     case 9: // ARRAY_2D.LZ.DC
-        CheckAlignment(reg_a, 2);
-        CheckAlignment(reg_b, 2);
         info.type.Assign(TextureType::ColorArray2D);
         info.is_depth.Assign(1);
         return v.ir.ImageSampleDrefExplicitLod(
             handle, v.ir.CompositeConstruct(v.F(reg_a + 1), v.F(reg_b), ReadArray(v, v.X(reg_a))),
             v.F(reg_b + 1), zero, {}, info);
     case 10: // 3D
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::Color3D);
         return v.ir.ImageSampleImplicitLod(handle, Composite(v, reg_a, reg_a + 1, reg_b), {}, {},
                                            {}, info);
     case 11: // 3D.LZ
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::Color3D);
         return v.ir.ImageSampleExplicitLod(handle, Composite(v, reg_a, reg_a + 1, reg_b), zero, {},
                                            info);
     case 12: // CUBE
-        CheckAlignment(reg_a, 2);
         info.type.Assign(TextureType::ColorCube);
         return v.ir.ImageSampleImplicitLod(handle, Composite(v, reg_a, reg_a + 1, reg_b), {}, {},
                                            {}, info);
     case 13: // CUBE.LL
-        CheckAlignment(reg_a, 2);
-        CheckAlignment(reg_b, 2);
         info.type.Assign(TextureType::ColorCube);
         return v.ir.ImageSampleExplicitLod(handle, Composite(v, reg_a, reg_a + 1, reg_b),
                                            v.F(reg_b + 1), {}, info);
@@ -187,12 +167,10 @@ IR::Reg RegStoreComponent32(u64 insn, unsigned index) {
     case 0:
         return texs.dest_reg_a;
     case 1:
-        CheckAlignment(texs.dest_reg_a, 2);
         return texs.dest_reg_a + 1;
     case 2:
         return texs.dest_reg_b;
     case 3:
-        CheckAlignment(texs.dest_reg_b, 2);
         return texs.dest_reg_b + 1;
     }
     throw LogicError("Invalid store index {}", index);

src/shader_recompiler/frontend/maxwell/translate/impl/texture_gather_swizzled.cpp

@@ -34,12 +34,6 @@ union Encoding {
     BitField<36, 13, u64> cbuf_offset;
 };
 
-void CheckAlignment(IR::Reg reg, size_t alignment) {
-    if (!IR::IsAligned(reg, alignment)) {
-        throw NotImplementedException("Unaligned source register {}", reg);
-    }
-}
-
 IR::Value MakeOffset(TranslatorVisitor& v, IR::Reg reg) {
     const IR::U32 value{v.X(reg)};
     return v.ir.CompositeConstruct(v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(6), true),
@@ -60,18 +54,15 @@ IR::Value Sample(TranslatorVisitor& v, u64 insn) {
     info.is_depth.Assign(tld4s.dc != 0 ? 1 : 0);
     IR::Value coords;
     if (tld4s.aoffi != 0) {
-        CheckAlignment(reg_a, 2);
         coords = v.ir.CompositeConstruct(v.F(reg_a), v.F(reg_a + 1));
         IR::Value offset = MakeOffset(v, reg_b);
         if (tld4s.dc != 0) {
-            CheckAlignment(reg_b, 2);
             IR::F32 dref = v.F(reg_b + 1);
             return v.ir.ImageGatherDref(handle, coords, offset, {}, dref, info);
         }
         return v.ir.ImageGather(handle, coords, offset, {}, info);
     }
     if (tld4s.dc != 0) {
-        CheckAlignment(reg_a, 2);
         coords = v.ir.CompositeConstruct(v.F(reg_a), v.F(reg_a + 1));
         IR::F32 dref = v.F(reg_b);
         return v.ir.ImageGatherDref(handle, coords, {}, {}, dref, info);
@@ -86,12 +77,10 @@ IR::Reg RegStoreComponent32(u64 insn, size_t index) {
     case 0:
         return tlds4.dest_reg_a;
     case 1:
-        CheckAlignment(tlds4.dest_reg_a, 2);
         return tlds4.dest_reg_a + 1;
     case 2:
         return tlds4.dest_reg_b;
     case 3:
-        CheckAlignment(tlds4.dest_reg_b, 2);
         return tlds4.dest_reg_b + 1;
     }
     throw LogicError("Invalid store index {}", index);

src/shader_recompiler/frontend/maxwell/translate/impl/texture_load_swizzled.cpp

@@ -55,12 +55,6 @@ union Encoding {
     BitField<53, 4, u64> encoding;
 };
 
-void CheckAlignment(IR::Reg reg, size_t alignment) {
-    if (!IR::IsAligned(reg, alignment)) {
-        throw NotImplementedException("Unaligned source register {}", reg);
-    }
-}
-
 IR::Value MakeOffset(TranslatorVisitor& v, IR::Reg reg) {
     const IR::U32 value{v.X(reg)};
     return v.ir.CompositeConstruct(v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(4), true),
@@ -92,38 +86,31 @@ IR::Value Sample(TranslatorVisitor& v, u64 insn) {
         coords = v.ir.CompositeConstruct(v.X(reg_a), v.X(reg_b));
         break;
     case 4:
-        CheckAlignment(reg_a, 2);
         texture_type = Shader::TextureType::Color2D;
         coords = v.ir.CompositeConstruct(v.X(reg_a), v.X(reg_a + 1));
         offsets = MakeOffset(v, reg_b);
         break;
     case 5:
-        CheckAlignment(reg_a, 2);
         texture_type = Shader::TextureType::Color2D;
         coords = v.ir.CompositeConstruct(v.X(reg_a), v.X(reg_a + 1));
         lod = v.X(reg_b);
         break;
     case 6:
-        CheckAlignment(reg_a, 2);
         texture_type = Shader::TextureType::Color2D;
         coords = v.ir.CompositeConstruct(v.X(reg_a), v.X(reg_a + 1));
         multisample = v.X(reg_b);
         break;
     case 7:
-        CheckAlignment(reg_a, 2);
         texture_type = Shader::TextureType::Color3D;
         coords = v.ir.CompositeConstruct(v.X(reg_a), v.X(reg_a + 1), v.X(reg_b));
         break;
     case 8: {
-        CheckAlignment(reg_b, 2);
         const IR::U32 array{v.ir.BitFieldExtract(v.X(reg_a), v.ir.Imm32(0), v.ir.Imm32(16))};
         texture_type = Shader::TextureType::ColorArray2D;
         coords = v.ir.CompositeConstruct(v.X(reg_b), v.X(reg_b + 1), array);
         break;
     }
     case 12:
-        CheckAlignment(reg_a, 2);
-        CheckAlignment(reg_b, 2);
         texture_type = Shader::TextureType::Color2D;
         coords = v.ir.CompositeConstruct(v.X(reg_a), v.X(reg_a + 1));
         lod = v.X(reg_b);
@@ -166,12 +153,10 @@ IR::Reg RegStoreComponent32(u64 insn, unsigned index) {
     case 0:
         return tlds.dest_reg_a;
     case 1:
-        CheckAlignment(tlds.dest_reg_a, 2);
         return tlds.dest_reg_a + 1;
     case 2:
         return tlds.dest_reg_b;
     case 3:
-        CheckAlignment(tlds.dest_reg_b, 2);
         return tlds.dest_reg_b + 1;
     }
     throw LogicError("Invalid store index {}", index);

src/shader_recompiler/frontend/maxwell/translate_program.cpp

@@ -169,11 +169,36 @@ std::map<IR::Attribute, IR::Attribute> GenerateLegacyToGenericMappings(
     return mapping;
 }
 
+struct PassthroughVertices {
+    u32 count;
+    u32 first;
+    u32 stride;
+};
+
+PassthroughVertices GetPassthroughVertices(InputTopology input_topology) {
+    switch (input_topology) {
+    case InputTopology::Points:
+        return {1, 0, 1};
+    case InputTopology::Lines:
+        return {2, 0, 1};
+    case InputTopology::LinesAdjacency:
+        return {2, 1, 1};
+    case InputTopology::Triangles:
+        return {3, 0, 1};
+    case InputTopology::TrianglesAdjacency:
+        return {3, 0, 2};
+    }
+    return {3, 0, 1};
+}
+
 void EmitGeometryPassthrough(IR::IREmitter& ir, const IR::Program& program,
                              const Shader::VaryingState& passthrough_mask,
                              bool passthrough_position,
-                             std::optional<IR::Attribute> passthrough_layer_attr) {
-    for (u32 i = 0; i < program.output_vertices; i++) {
+                             std::optional<IR::Attribute> passthrough_layer_attr,
+                             InputTopology input_topology) {
+    const PassthroughVertices vertices{GetPassthroughVertices(input_topology)};
+    for (u32 vertex = 0; vertex < vertices.count; vertex++) {
+        const u32 i = vertices.first + vertex * vertices.stride;
         // Assign generics from input
         for (u32 j = 0; j < 32; j++) {
             if (!passthrough_mask.Generic(j)) {
@@ -208,25 +233,16 @@ void EmitGeometryPassthrough(IR::IREmitter& ir, const IR::Program& program,
     ir.EndPrimitive(ir.Imm32(0));
 }
 
-u32 GetOutputTopologyVertices(OutputTopology output_topology) {
-    switch (output_topology) {
-    case OutputTopology::PointList:
-        return 1;
-    case OutputTopology::LineStrip:
-        return 2;
-    default:
-        return 3;
-    }
-}
-
-void LowerGeometryPassthrough(const IR::Program& program, const HostTranslateInfo& host_info) {
+void LowerGeometryPassthrough(const IR::Program& program, const HostTranslateInfo& host_info,
+                              InputTopology input_topology) {
     for (IR::Block* const block : program.blocks) {
         for (IR::Inst& inst : block->Instructions()) {
             if (inst.GetOpcode() == IR::Opcode::Epilogue) {
                 IR::IREmitter ir{*block, IR::Block::InstructionList::s_iterator_to(inst)};
                 EmitGeometryPassthrough(
                     ir, program, program.info.passthrough,
-                    program.info.passthrough.AnyComponent(IR::Attribute::PositionX), {});
+                    program.info.passthrough.AnyComponent(IR::Attribute::PositionX), {},
+                    input_topology);
             }
         }
     }
@@ -235,7 +251,8 @@ void LowerGeometryPassthrough(const IR::Program& program, const HostTranslateInf
 } // Anonymous namespace
 
 IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Block>& block_pool,
-                             Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info) {
+                             Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info,
+                             InputTopology input_topology) {
     HostTranslateInfo normalized_host_info{host_info};
     normalized_host_info.ApplyDescriptorLimitPolicy();
 
@@ -264,8 +281,9 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
             }
 
             if (!normalized_host_info.support_geometry_shader_passthrough) {
-                program.output_vertices = GetOutputTopologyVertices(program.output_topology);
-                LowerGeometryPassthrough(program, normalized_host_info);
+                program.output_vertices = GetPassthroughVertices(input_topology).count;
+                LowerGeometryPassthrough(program, normalized_host_info, input_topology);
+                program.is_geometry_passthrough = false;
             }
         }
         break;
@@ -414,11 +432,12 @@ IR::Program GenerateGeometryPassthrough(ObjectPool<IR::Inst>& inst_pool,
                                         ObjectPool<IR::Block>& block_pool,
                                         const HostTranslateInfo& host_info,
                                         IR::Program& source_program,
-                                        Shader::OutputTopology output_topology) {
+                                        Shader::OutputTopology output_topology,
+                                        InputTopology input_topology) {
     IR::Program program;
     program.stage = Stage::Geometry;
     program.output_topology = output_topology;
-    program.output_vertices = GetOutputTopologyVertices(output_topology);
+    program.output_vertices = GetPassthroughVertices(input_topology).count;
 
     program.is_geometry_passthrough = false;
     program.info.loads.mask = source_program.info.stores.mask;
@@ -433,7 +452,7 @@ IR::Program GenerateGeometryPassthrough(ObjectPool<IR::Inst>& inst_pool,
 
     IR::IREmitter ir{*current_block};
     EmitGeometryPassthrough(ir, program, program.info.stores, true,
-                            source_program.info.emulated_layer);
+                            source_program.info.emulated_layer, input_topology);
 
     IR::Block* return_block{block_pool.Create(inst_pool)};
     IR::IREmitter{*return_block}.Epilogue();

src/shader_recompiler/frontend/maxwell/translate_program.h

@@ -18,7 +18,8 @@ namespace Shader::Maxwell {
 
 [[nodiscard]] IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool,
                                            ObjectPool<IR::Block>& block_pool, Environment& env,
-                                           Flow::CFG& cfg, const HostTranslateInfo& host_info);
+                                           Flow::CFG& cfg, const HostTranslateInfo& host_info,
+                                           InputTopology input_topology);
 
 [[nodiscard]] IR::Program MergeDualVertexPrograms(IR::Program& vertex_a, IR::Program& vertex_b,
                                                   Environment& env_vertex_b);
@@ -32,6 +33,7 @@ void ConvertLegacyToGeneric(IR::Program& program, const RuntimeInfo& runtime_inf
                                                       ObjectPool<IR::Block>& block_pool,
                                                       const HostTranslateInfo& host_info,
                                                       IR::Program& source_program,
-                                                      Shader::OutputTopology output_topology);
+                                                      Shader::OutputTopology output_topology,
+                                                      InputTopology input_topology);
 
 } // namespace Shader::Maxwell

src/video_core/buffer_cache/buffer_cache.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <algorithm>
+#include <bit>
 #include <memory>
 #include <numeric>
 
@@ -809,46 +810,46 @@ void BufferCache<P>::BindHostVertexBuffers() {
 
     if (use_optimized_vertex_buffers) {
         auto& flags = maxwell3d->dirty.flags;
-        u32 enabled_mask = enabled_vertex_buffers_mask;
-        HostBindings<Buffer> bindings{};
-        u32 last_index = (std::numeric_limits<u32>::max)();
-        const auto flush_bindings = [&]() {
-            if (bindings.buffers.empty()) {
-                return;
-            }
-            bindings.max_index = bindings.min_index + static_cast<u32>(bindings.buffers.size());
-            runtime.BindVertexBuffers(bindings);
-            bindings = HostBindings<Buffer>{};
-            last_index = (std::numeric_limits<u32>::max)();
-        };
-        while (enabled_mask != 0) {
-            const u32 index = std::countr_zero(enabled_mask);
-            enabled_mask &= (enabled_mask - 1);
+        const u32 enabled_mask = enabled_vertex_buffers_mask;
+        bool any_dirty = false;
+        u32 pending_mask = enabled_mask;
+        while (pending_mask != 0) {
+            const u32 index = std::countr_zero(pending_mask);
+            pending_mask &= (pending_mask - 1);
             const Binding& binding = VertexBufferSlot(index);
             Buffer& buffer = slot_buffers[binding.buffer_id];
             TouchBuffer(buffer, binding.buffer_id);
             SynchronizeBuffer(buffer, binding.device_addr, binding.size);
-            if (!flags[Dirty::VertexBuffer0 + index]) {
-                flush_bindings();
-                continue;
-            }
+            any_dirty |= flags[Dirty::VertexBuffer0 + index];
+        }
+        if (enabled_mask == 0 || !any_dirty) {
+            return;
+        }
+        const u32 min_index = static_cast<u32>(std::countr_zero(enabled_mask));
+        const u32 max_index = 32u - static_cast<u32>(std::countl_zero(enabled_mask));
+        HostBindings<Buffer> bindings{};
+        bindings.min_index = min_index;
+        bindings.max_index = max_index;
+        for (u32 index = min_index; index < max_index; ++index) {
             flags[Dirty::VertexBuffer0 + index] = false;
             const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
+            if ((enabled_mask & (1u << index)) == 0) {
+                bindings.buffers.push_back(&slot_buffers[NULL_BUFFER_ID]);
+                bindings.offsets.push_back(0);
+                bindings.sizes.push_back(0);
+                bindings.strides.push_back(stride);
+                continue;
+            }
+            const Binding& binding = VertexBufferSlot(index);
+            Buffer& buffer = slot_buffers[binding.buffer_id];
             const u32 offset = buffer.Offset(binding.device_addr);
             buffer.MarkUsage(offset, binding.size);
-            if (!bindings.buffers.empty() && index != last_index + 1) {
-                flush_bindings();
-            }
-            if (bindings.buffers.empty()) {
-                bindings.min_index = index;
-            }
             bindings.buffers.push_back(&buffer);
             bindings.offsets.push_back(offset);
             bindings.sizes.push_back(binding.size);
             bindings.strides.push_back(stride);
-            last_index = index;
         }
-        flush_bindings();
+        runtime.BindVertexBuffers(bindings);
     } else {
         HostBindings<typename P::Buffer> host_bindings;
         bool any_valid{false};
@@ -1448,8 +1449,8 @@ BufferId BufferCache<P>::FindBuffer(DAddr device_addr, u32 size) {
     const BufferId buffer_id = page_table[page];
     if (buffer_id) {
         Buffer& buffer = slot_buffers[buffer_id];
+        WaitForGpuFenceIfNeeded(buffer);
         if (buffer.IsInBounds(device_addr, size)) {
-            SynchronizeBufferIfNeeded(buffer);
             return buffer_id;
         }
     }
@@ -1457,28 +1458,16 @@ BufferId BufferCache<P>::FindBuffer(DAddr device_addr, u32 size) {
 }
 
 template <class P>
-void BufferCache<P>::SynchronizeBufferIfNeeded(Buffer& buffer) {
-    const bool gpu_fence_accurate = Settings::IsGPUFenceBehaviorAccurate();
-    const bool gpu_fence_strict = Settings::IsGPUFenceBehaviorStrict();
-    const bool should_sync = gpu_fence_accurate || gpu_fence_strict;
-    if (should_sync) {
-        if (gpu_fence_accurate) {
-            if constexpr (!IS_OPENGL) {
-                const bool should_wait = buffer.getWriteTick() > runtime.KnownGpuTick() + 3;
-                if (should_wait) {
-                    runtime.Wait(buffer.getWriteTick());
-                }
-            }
-        } else if (gpu_fence_strict) {
-            bool should_download = true;
-            if constexpr (!IS_OPENGL) {
-                should_download = buffer.getWriteTick() > runtime.KnownGpuTick();
-                if (should_download) {
-                    runtime.Wait(buffer.getWriteTick());
-                }
-            }
-            if (should_download) {
-                DownloadBufferMemory(buffer);
+void BufferCache<P>::WaitForGpuFenceIfNeeded(Buffer& buffer) {
+    if constexpr (!IS_OPENGL) {
+        const bool gpu_fence_accurate = Settings::IsGPUFenceBehaviorAccurate();
+        const bool gpu_fence_strict = Settings::IsGPUFenceBehaviorStrict();
+        if (gpu_fence_accurate || gpu_fence_strict) {
+            const u64 gpu_tick_delay = gpu_fence_strict ? 0 : 3;
+            const u64 buffer_tick = buffer.getWriteTick();
+            const u64 gpu_tick = runtime.KnownGpuTick();
+            if (buffer_tick > gpu_tick + gpu_tick_delay) {
+                runtime.Wait(buffer_tick);
             }
         }
     }
@@ -1700,6 +1689,9 @@ void BufferCache<P>::ImmediateUploadMemory([[maybe_unused]] Buffer& buffer,
                 if (immediate_buffer.empty()) {
                     immediate_buffer = ImmediateBuffer(largest_copy);
                 }
+                if (Settings::values.enable_gpu_buffer_readback.GetValue()) {
+                    DownloadBufferMemory(buffer, device_addr, copy.size);
+                }
                 device_memory.ReadBlockUnsafe(device_addr, immediate_buffer.data(), copy.size);
                 upload_span = immediate_buffer.subspan(0, copy.size);
             }
@@ -1718,6 +1710,9 @@ void BufferCache<P>::MappedUploadMemory([[maybe_unused]] Buffer& buffer,
         for (BufferCopy& copy : copies) {
             u8* const src_pointer = staging_pointer.data() + copy.src_offset;
             const DAddr device_addr = buffer.CpuAddr() + copy.dst_offset;
+            if (Settings::values.enable_gpu_buffer_readback.GetValue()) {
+                DownloadBufferMemory(buffer, device_addr, copy.size);
+            }
             device_memory.ReadBlockUnsafe(device_addr, src_pointer, copy.size);
             // Apply the staging offset
             copy.src_offset += upload_staging.offset;

src/video_core/buffer_cache/buffer_cache_base.h

@@ -435,8 +435,6 @@ private:
 
     bool SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 size);
 
-    void SynchronizeBufferIfNeeded(Buffer& buffer);
-
     void UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy,
                       std::span<BufferCopy> copies);
 

src/video_core/host_shaders/CMakeLists.txt

@@ -17,11 +17,13 @@ set(SHADER_FILES
     ${CMAKE_CURRENT_SOURCE_DIR}/astc_decoder.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_color_float.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_2d.comp
+    ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_2d_buffer.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_color_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_stencil_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_bcn.comp
+    ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_buffer.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d24s8.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d32f.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_d32f_to_abgr8.frag
@@ -32,6 +34,8 @@ set(SHADER_FILES
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_msaa_to_non_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.frag
+    ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa_depth.frag
+    ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa_depth_stencil.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_s8d24_to_abgr8.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/full_screen_triangle.vert
     ${CMAKE_CURRENT_SOURCE_DIR}/fxaa.frag

src/video_core/host_shaders/block_linear_unswizzle_2d_buffer.comp

@@ -0,0 +1,104 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 430
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_BUFFER 1
+
+layout(push_constant) uniform PushConstants {
+    uvec3 dim;
+    uint bytes_per_block_log2;
+
+    uvec3 origin;
+    uint layer_stride;
+
+    uint block_size;
+    uint x_shift;
+    uint block_height;
+    uint block_height_mask;
+} pc;
+
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_BUFFER, std430) writeonly buffer OutputBuffer {
+    uint out_u32[];
+};
+
+layout(local_size_x = 16, local_size_y = 8, local_size_z = 1) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT;
+
+const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1u, GOB_SIZE_Y - 1u);
+
+uint SwizzleTable(uint pos) {
+    const uint t[8] = uint[](
+        0x12100200, 0x13110301, 0x16140604, 0x17150705,
+        0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d
+    );
+    const uint i = pos >> 4;
+    const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff;
+    return (h << 4) | (pos & 0xf);
+}
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return SwizzleTable(pos.y * 64u + pos.x);
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (pc.bytes_per_block_log2) {
+    case 2u:
+        return uvec4(u32data[offset / 4u], 0u, 0u, 0u);
+    case 3u:
+        return uvec4(u64data[offset / 8u], 0u, 0u);
+    case 4u:
+        return u128data[offset / 16u];
+    }
+    return uvec4(0u);
+}
+
+void main() {
+    uvec3 coord = gl_GlobalInvocationID;
+    if (coord.x >= pc.dim.x || coord.y >= pc.dim.y || coord.z >= pc.dim.z) {
+        return;
+    }
+
+    uvec3 pos = coord + pc.origin;
+    pos.x <<= pc.bytes_per_block_log2;
+
+    uint swizzle = SwizzleOffset(pos.xy);
+    uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0u;
+    offset += pos.z * pc.layer_stride;
+    offset += (block_y >> pc.block_height) * pc.block_size;
+    offset += (block_y & pc.block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << pc.x_shift;
+    offset += swizzle;
+
+    uvec4 texel = ReadTexel(offset);
+
+    uint words = 1u << (pc.bytes_per_block_log2 - 2u);
+    uint linear_index = coord.x + coord.y * pc.dim.x + coord.z * pc.dim.x * pc.dim.y;
+    uint out_idx = linear_index * words;
+
+    out_u32[out_idx] = texel.x;
+    if (words > 1u) {
+        out_u32[out_idx + 1u] = texel.y;
+    }
+    if (words > 2u) {
+        out_u32[out_idx + 2u] = texel.z;
+        out_u32[out_idx + 3u] = texel.w;
+    }
+}

src/video_core/host_shaders/block_linear_unswizzle_3d_buffer.comp

@@ -0,0 +1,105 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 430
+
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_BUFFER 1
+
+layout(push_constant) uniform PushConstants {
+    uvec3 dim;
+    uint bytes_per_block_log2;
+
+    uvec3 origin;
+    uint slice_size;
+
+    uint block_size;
+    uint x_shift;
+    uint block_height;
+    uint block_height_mask;
+
+    uint block_depth;
+    uint block_depth_mask;
+} pc;
+
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_BUFFER, std430) writeonly buffer OutputBuffer {
+    uint out_u32[];
+};
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 4) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT;
+
+const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1u, GOB_SIZE_Y - 1u);
+
+uint SwizzleTable(uint pos) {
+    const uint t[8] = uint[](
+        0x12100200, 0x13110301, 0x16140604, 0x17150705,
+        0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d
+    );
+    const uint i = pos >> 4;
+    const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff;
+    return (h << 4) | (pos & 0xf);
+}
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return SwizzleTable(pos.y * 64u + pos.x);
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (pc.bytes_per_block_log2) {
+    case 2u:
+        return uvec4(u32data[offset / 4u], 0u, 0u, 0u);
+    case 3u:
+        return uvec4(u64data[offset / 8u], 0u, 0u);
+    case 4u:
+        return u128data[offset / 16u];
+    }
+    return uvec4(0u);
+}
+
+void main() {
+    uvec3 coord = gl_GlobalInvocationID;
+    if (coord.x >= pc.dim.x || coord.y >= pc.dim.y || coord.z >= pc.dim.z) {
+        return;
+    }
+
+    uvec3 pos = coord + pc.origin;
+    pos.x <<= pc.bytes_per_block_log2;
+
+    uint swizzle = SwizzleOffset(pos.xy);
+    uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0u;
+    offset += (pos.z >> pc.block_depth) * pc.slice_size;
+    offset += (pos.z & pc.block_depth_mask) << (GOB_SIZE_SHIFT + pc.block_height);
+    offset += (block_y >> pc.block_height) * pc.block_size;
+    offset += (block_y & pc.block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << pc.x_shift;
+    offset += swizzle;
+
+    uvec4 texel = ReadTexel(offset);
+
+    uint words = 1u << (pc.bytes_per_block_log2 - 2u);
+    uint linear_index = coord.x + coord.y * pc.dim.x + coord.z * pc.dim.x * pc.dim.y;
+    uint out_idx = linear_index * words;
+
+    out_u32[out_idx] = texel.x;
+    if (words > 1u) {
+        out_u32[out_idx + 1u] = texel.y;
+    }
+    if (words > 2u) {
+        out_u32[out_idx + 2u] = texel.z;
+        out_u32[out_idx + 3u] = texel.w;
+    }
+}

src/video_core/host_shaders/convert_non_msaa_to_msaa_depth.frag

@@ -0,0 +1,19 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 450 core
+
+layout(binding = 0) uniform sampler2D img_in;
+
+layout(push_constant) uniform PushConstants {
+    ivec2 dst_offset;
+    ivec2 src_offset;
+    ivec2 scale;
+};
+
+void main() {
+    const ivec2 msaa_coord = ivec2(gl_FragCoord.xy) - dst_offset;
+    const ivec2 sample_offset = ivec2(gl_SampleID % scale.x, gl_SampleID / scale.x);
+    const ivec2 coord = msaa_coord * scale + sample_offset + src_offset;
+    gl_FragDepth = texelFetch(img_in, coord, 0).r;
+}

src/video_core/host_shaders/convert_non_msaa_to_msaa_depth_stencil.frag

@@ -0,0 +1,22 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 450 core
+#extension GL_ARB_shader_stencil_export : require
+
+layout(binding = 0) uniform sampler2D depth_tex;
+layout(binding = 1) uniform usampler2D stencil_tex;
+
+layout(push_constant) uniform PushConstants {
+    ivec2 dst_offset;
+    ivec2 src_offset;
+    ivec2 scale;
+};
+
+void main() {
+    const ivec2 msaa_coord = ivec2(gl_FragCoord.xy) - dst_offset;
+    const ivec2 sample_offset = ivec2(gl_SampleID % scale.x, gl_SampleID / scale.x);
+    const ivec2 coord = msaa_coord * scale + sample_offset + src_offset;
+    gl_FragDepth = texelFetch(depth_tex, coord, 0).r;
+    gl_FragStencilRefARB = int(texelFetch(stencil_tex, coord, 0).r);
+}

src/video_core/host_shaders/sgsr1_shader.vert

@@ -7,6 +7,7 @@ layout(push_constant) uniform constants {
     vec2 scale;
     vec2 size;
     vec2 resize_factor;
+    vec2 crop_offset;
     float edge_sharpness;
 };
 layout(location = 0) out highp vec2 texcoord;
@@ -15,5 +16,5 @@ void main() {
     float x = float((gl_VertexIndex & 1) << 2);
     float y = float((gl_VertexIndex & 2) << 1);
     gl_Position = vec4(x - 1.0f, y - 1.0f, 0.0, 1.0f) * vec4(sign(resize_factor), 1.f, 1.f);
-    texcoord = vec2(x, y) * abs(resize_factor) * 0.5;
+    texcoord = crop_offset + vec2(x, y) * abs(resize_factor) * 0.5;
 }

src/video_core/host_shaders/sgsr1_shader_mobile.frag

@@ -14,6 +14,7 @@ layout(push_constant) uniform constants {
     vec2 scale;
     vec2 size;
     vec2 resize_factor;
+    vec2 crop_offset;
     float edge_sharpness;
 };
 layout(set = 0, binding = 0) uniform sampler2D sampler0;

src/video_core/host_shaders/sgsr1_shader_mobile_edge_direction.frag

@@ -13,6 +13,7 @@
 layout( push_constant ) uniform constants {
     vec4 ViewportInfo[1];
     vec2 ResizeFactor;
+    vec2 CropOffset;
     float EdgeSharpness;
 };
 layout(set = 0, binding = 0) uniform sampler2D ps0;

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -72,6 +72,25 @@ Shader::OutputTopology MaxwellToOutputTopology(Maxwell::PrimitiveTopology topolo
     }
 }
 
+Shader::InputTopology MaxwellToInputTopology(Maxwell::PrimitiveTopology topology) {
+    switch (topology) {
+    case Maxwell::PrimitiveTopology::Points:
+        return Shader::InputTopology::Points;
+    case Maxwell::PrimitiveTopology::Lines:
+    case Maxwell::PrimitiveTopology::LineLoop:
+    case Maxwell::PrimitiveTopology::LineStrip:
+        return Shader::InputTopology::Lines;
+    case Maxwell::PrimitiveTopology::LinesAdjacency:
+    case Maxwell::PrimitiveTopology::LineStripAdjacency:
+        return Shader::InputTopology::LinesAdjacency;
+    case Maxwell::PrimitiveTopology::TrianglesAdjacency:
+    case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
+        return Shader::InputTopology::TrianglesAdjacency;
+    default:
+        return Shader::InputTopology::Triangles;
+    }
+}
+
 Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineKey& key,
                                     const Shader::IR::Program& program,
                                     const Shader::IR::Program* previous_program,
@@ -127,33 +146,7 @@ Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineKey& key,
     default:
         break;
     }
-    switch (key.gs_input_topology) {
-    case Maxwell::PrimitiveTopology::Points:
-        info.input_topology = Shader::InputTopology::Points;
-        break;
-    case Maxwell::PrimitiveTopology::Lines:
-    case Maxwell::PrimitiveTopology::LineLoop:
-    case Maxwell::PrimitiveTopology::LineStrip:
-        info.input_topology = Shader::InputTopology::Lines;
-        break;
-    case Maxwell::PrimitiveTopology::Triangles:
-    case Maxwell::PrimitiveTopology::TriangleStrip:
-    case Maxwell::PrimitiveTopology::TriangleFan:
-    case Maxwell::PrimitiveTopology::Quads:
-    case Maxwell::PrimitiveTopology::QuadStrip:
-    case Maxwell::PrimitiveTopology::Polygon:
-    case Maxwell::PrimitiveTopology::Patches:
-        info.input_topology = Shader::InputTopology::Triangles;
-        break;
-    case Maxwell::PrimitiveTopology::LinesAdjacency:
-    case Maxwell::PrimitiveTopology::LineStripAdjacency:
-        info.input_topology = Shader::InputTopology::LinesAdjacency;
-        break;
-    case Maxwell::PrimitiveTopology::TrianglesAdjacency:
-    case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
-        info.input_topology = Shader::InputTopology::TrianglesAdjacency;
-        break;
-    }
+    info.input_topology = MaxwellToInputTopology(key.gs_input_topology);
     info.glasm_use_storage_buffers = glasm_use_storage_buffers;
     return info;
 }
@@ -483,8 +476,10 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
             && index == u32(Maxwell::ShaderType::Geometry);
         if (key.unique_hashes[index] == 0 && is_emulated_stage) {
             auto topology = MaxwellToOutputTopology(key.gs_input_topology);
-            programs[index] = GenerateGeometryPassthrough(pools.inst, pools.block, host_info,
-                                                          *layer_source_program, topology);
+            programs[index] =
+                GenerateGeometryPassthrough(pools.inst, pools.block, host_info,
+                                            *layer_source_program, topology,
+                                            MaxwellToInputTopology(key.gs_input_topology));
             continue;
         }
         if (key.unique_hashes[index] == 0) {
@@ -502,13 +497,15 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
 
         if (!uses_vertex_a || index != 1) {
             // Normal path
-            programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info);
+            programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info,
+                                              MaxwellToInputTopology(key.gs_input_topology));
 
             total_storage_buffers += Shader::NumDescriptors(programs[index].info.storage_buffers_descriptors);
         } else {
             // VertexB path when VertexA is present.
             auto& program_va{programs[0]};
-            auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
+            auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info,
+                                             MaxwellToInputTopology(key.gs_input_topology))};
             total_storage_buffers += Shader::NumDescriptors(program_vb.info.storage_buffers_descriptors);
             programs[index] = MergeDualVertexPrograms(program_va, program_vb, env);
         }
@@ -597,7 +594,8 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
         env.Dump(hash, key.unique_hash);
     }
 
-    auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
+    auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info,
+                                  Shader::InputTopology::Points)};
     const u32 num_storage_buffers{Shader::NumDescriptors(program.info.storage_buffers_descriptors)};
     Shader::RuntimeInfo info;
     info.glasm_use_storage_buffers = num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks();

src/video_core/renderer_vulkan/blit_image.cpp

@@ -22,6 +22,8 @@
 #include "video_core/host_shaders/convert_float_to_depth_frag_spv.h"
 #include "video_core/host_shaders/convert_msaa_to_non_msaa_frag_spv.h"
 #include "video_core/host_shaders/convert_non_msaa_to_msaa_frag_spv.h"
+#include "video_core/host_shaders/convert_non_msaa_to_msaa_depth_frag_spv.h"
+#include "video_core/host_shaders/convert_non_msaa_to_msaa_depth_stencil_frag_spv.h"
 #include "video_core/host_shaders/convert_s8d24_to_abgr8_frag_spv.h"
 #include "video_core/host_shaders/full_screen_triangle_vert_spv.h"
 #include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h"
@@ -519,7 +521,8 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
 }
 
 [[nodiscard]] vk::ImageView MakeMSAACopyView(const vk::Device& device, VkImage image,
-                                             VkFormat format, u32 base_level) {
+                                             VkFormat format, u32 base_level,
+                                             VkImageAspectFlags aspect_mask) {
     return device.CreateImageView(VkImageViewCreateInfo{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
         .pNext = nullptr,
@@ -534,7 +537,7 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
             .a = VK_COMPONENT_SWIZZLE_IDENTITY,
         },
         .subresourceRange{
-            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .aspectMask = aspect_mask,
             .baseMipLevel = base_level,
             .levelCount = 1,
             .baseArrayLayer = 0,
@@ -586,6 +589,10 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
       msaa_copy_pipeline_layout(device.GetLogical().CreatePipelineLayout(PipelineLayoutCreateInfo(
           one_texture_set_layout.address(),
           PUSH_CONSTANT_RANGE<VK_SHADER_STAGE_FRAGMENT_BIT, sizeof(MSAACopyPushConstants)>))),
+      msaa_copy_depth_stencil_pipeline_layout(
+          device.GetLogical().CreatePipelineLayout(PipelineLayoutCreateInfo(
+              two_textures_set_layout.address(),
+              PUSH_CONSTANT_RANGE<VK_SHADER_STAGE_FRAGMENT_BIT, sizeof(MSAACopyPushConstants)>))),
       full_screen_vert(BuildShader(device, FULL_SCREEN_TRIANGLE_VERT_SPV)),
       blit_color_to_color_frag(BuildShader(device, BLIT_COLOR_FLOAT_FRAG_SPV)),
       blit_color_msaa_frag(BuildShader(device, BLIT_COLOR_MSAA_FRAG_SPV)),
@@ -610,6 +617,12 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
       convert_s8d24_to_abgr8_frag(BuildShader(device, CONVERT_S8D24_TO_ABGR8_FRAG_SPV)),
       convert_msaa_to_non_msaa_frag(BuildShader(device, CONVERT_MSAA_TO_NON_MSAA_FRAG_SPV)),
       convert_non_msaa_to_msaa_frag(BuildShader(device, CONVERT_NON_MSAA_TO_MSAA_FRAG_SPV)),
+      convert_non_msaa_to_msaa_depth_frag(
+          BuildShader(device, CONVERT_NON_MSAA_TO_MSAA_DEPTH_FRAG_SPV)),
+      convert_non_msaa_to_msaa_depth_stencil_frag(
+          device.IsExtShaderStencilExportSupported()
+              ? BuildShader(device, CONVERT_NON_MSAA_TO_MSAA_DEPTH_STENCIL_FRAG_SPV)
+              : vk::ShaderModule{}),
       linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)),
       nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {}
 
@@ -920,10 +933,12 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
         ASSERT(copy.dst_subresource.num_layers == 1);
         vk::ImageView src_view =
             MakeMSAACopyView(device.GetLogical(), src_image, src_vk_format,
-                             static_cast<u32>(copy.src_subresource.base_level));
+                             static_cast<u32>(copy.src_subresource.base_level),
+                             VK_IMAGE_ASPECT_COLOR_BIT);
         vk::ImageView dst_view =
             MakeMSAACopyView(device.GetLogical(), dst_image, dst_vk_format,
-                             static_cast<u32>(copy.dst_subresource.base_level));
+                             static_cast<u32>(copy.dst_subresource.base_level),
+                             VK_IMAGE_ASPECT_COLOR_BIT);
         const VkOffset2D dst_offset{copy.dst_offset.x, copy.dst_offset.y};
         const VkExtent2D dst_extent{copy.extent.width, copy.extent.height};
         const VkRect2D render_area{
@@ -1191,7 +1206,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKe
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *blit_color_pipelines.back();
 }
 
@@ -1223,7 +1238,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePip
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *blit_depth_stencil_pipelines.back();
 }
 
@@ -1276,7 +1291,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *clear_color_pipelines.back();
 }
 
@@ -1332,7 +1347,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *clear_stencil_pipelines.back();
 }
 
@@ -1375,7 +1390,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPip
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *blit_msaa_color_pipelines.back();
 }
 
@@ -1413,10 +1428,202 @@ VkPipeline BlitImageHelper::FindOrEmplaceResolveDepthStencilPipeline(VkRenderPas
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *pipelines.back();
 }
 
+void BlitImageHelper::CopyMSAADepth(RenderPassCache& render_pass_cache, VkImage dst_image,
+                                    VideoCore::Surface::PixelFormat dst_format, VkImage src_image,
+                                    VideoCore::Surface::PixelFormat src_format, u32 num_samples,
+                                    std::span<const VideoCommon::ImageCopy> copies,
+                                    bool copy_stencil) {
+    while (!msaa_copy_resources.empty() && scheduler.IsFree(msaa_copy_resources.front().tick)) {
+        msaa_copy_resources.pop_front();
+    }
+    const auto [samples_x, samples_y] = VideoCommon::SamplesLog2(static_cast<int>(num_samples));
+    const s32 scale_x = 1 << samples_x;
+    const s32 scale_y = 1 << samples_y;
+    const VkSampleCountFlagBits samples = SampleCountFlag(num_samples);
+    RenderPassKey renderpass_key{};
+    renderpass_key.color_formats.fill(VideoCore::Surface::PixelFormat::Invalid);
+    renderpass_key.depth_format = dst_format;
+    renderpass_key.samples = samples;
+    const VkRenderPass renderpass = render_pass_cache.Get(renderpass_key);
+    const MSAACopyPipelineKey key{
+        .renderpass = renderpass,
+        .samples = samples,
+        .msaa_to_non_msaa = false,
+    };
+    const VkPipeline pipeline = FindOrEmplaceMSAACopyDepthPipeline(key, copy_stencil);
+    const VkPipelineLayout layout = copy_stencil ? *msaa_copy_depth_stencil_pipeline_layout
+                                                 : *msaa_copy_pipeline_layout;
+    const VkSampler sampler = *nearest_sampler;
+    const VkFormat src_vk_format =
+        MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, src_format).format;
+    const VkFormat dst_vk_format =
+        MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, dst_format).format;
+    const VkImageAspectFlags attachment_aspect =
+        VideoCore::Surface::GetFormatType(dst_format) ==
+                VideoCore::Surface::SurfaceType::DepthStencil
+            ? VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT
+            : VK_IMAGE_ASPECT_DEPTH_BIT;
+    for (const VideoCommon::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);
+        vk::ImageView src_view =
+            MakeMSAACopyView(device.GetLogical(), src_image, src_vk_format,
+                             static_cast<u32>(copy.src_subresource.base_level),
+                             VK_IMAGE_ASPECT_DEPTH_BIT);
+        vk::ImageView src_stencil_view =
+            copy_stencil ? MakeMSAACopyView(device.GetLogical(), src_image, src_vk_format,
+                                            static_cast<u32>(copy.src_subresource.base_level),
+                                            VK_IMAGE_ASPECT_STENCIL_BIT)
+                         : vk::ImageView{};
+        vk::ImageView dst_view =
+            MakeMSAACopyView(device.GetLogical(), dst_image, dst_vk_format,
+                             static_cast<u32>(copy.dst_subresource.base_level),
+                             attachment_aspect);
+        const VkOffset2D dst_offset{copy.dst_offset.x, copy.dst_offset.y};
+        const VkExtent2D dst_extent{copy.extent.width, copy.extent.height};
+        const VkRect2D render_area{
+            .offset = dst_offset,
+            .extent = dst_extent,
+        };
+        vk::Framebuffer framebuffer = device.GetLogical().CreateFramebuffer(VkFramebufferCreateInfo{
+            .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .renderPass = renderpass,
+            .attachmentCount = 1,
+            .pAttachments = dst_view.address(),
+            .width = static_cast<u32>(dst_offset.x) + dst_extent.width,
+            .height = static_cast<u32>(dst_offset.y) + dst_extent.height,
+            .layers = 1,
+        });
+        const MSAACopyPushConstants push_constants{
+            .dst_offset = {dst_offset.x, dst_offset.y},
+            .src_offset = {copy.src_offset.x, copy.src_offset.y},
+            .scale = {scale_x, scale_y},
+        };
+        scheduler.RequestOutsideRenderPassOperationContext();
+        const VkImageView src_stencil_handle = copy_stencil ? *src_stencil_view : VK_NULL_HANDLE;
+        scheduler.Record([this, pipeline, layout, sampler, renderpass,
+                          framebuffer_handle = *framebuffer, src_view_handle = *src_view,
+                          src_stencil_handle, src = src_image, dst = dst_image, render_area,
+                          attachment_aspect, push_constants](vk::CommandBuffer cmdbuf) {
+            const VkImageSubresourceRange src_range{
+                .aspectMask = attachment_aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            };
+            const std::array pre_barriers{
+                VkImageMemoryBarrier{
+                    .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+                    .pNext = nullptr,
+                    .srcAccessMask = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
+                                     VK_ACCESS_TRANSFER_WRITE_BIT,
+                    .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+                    .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .image = src,
+                    .subresourceRange = src_range,
+                },
+                VkImageMemoryBarrier{
+                    .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+                    .pNext = nullptr,
+                    .srcAccessMask = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
+                                     VK_ACCESS_TRANSFER_WRITE_BIT,
+                    .dstAccessMask = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
+                                     VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+                    .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .image = dst,
+                    .subresourceRange = src_range,
+                },
+            };
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                       VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+                                       VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                   VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
+                                       VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT,
+                                   0, nullptr, nullptr, pre_barriers);
+            const VkRenderPassBeginInfo renderpass_bi{
+                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
+                .pNext = nullptr,
+                .renderPass = renderpass,
+                .framebuffer = framebuffer_handle,
+                .renderArea = render_area,
+                .clearValueCount = 0,
+                .pClearValues = nullptr,
+            };
+            cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
+            const VkDescriptorSet descriptor_set =
+                src_stencil_handle != VK_NULL_HANDLE
+                    ? two_textures_descriptor_allocator.Commit()
+                    : one_texture_descriptor_allocator.Commit();
+            if (src_stencil_handle != VK_NULL_HANDLE) {
+                UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_view_handle,
+                                               src_stencil_handle);
+            } else {
+                UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view_handle);
+            }
+            cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
+                                      nullptr);
+            const VkViewport viewport{
+                .x = static_cast<float>(render_area.offset.x),
+                .y = static_cast<float>(render_area.offset.y),
+                .width = static_cast<float>(render_area.extent.width),
+                .height = static_cast<float>(render_area.extent.height),
+                .minDepth = 0.0f,
+                .maxDepth = 1.0f,
+            };
+            cmdbuf.SetViewport(0, viewport);
+            cmdbuf.SetScissor(0, render_area);
+            cmdbuf.PushConstants(layout, VK_SHADER_STAGE_FRAGMENT_BIT, push_constants);
+            cmdbuf.Draw(3, 1, 0, 0);
+            cmdbuf.EndRenderPass();
+            const VkImageMemoryBarrier post_barrier{
+                .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+                .pNext = nullptr,
+                .srcAccessMask = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+                .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT |
+                                 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT,
+                .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+                .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                .image = dst,
+                .subresourceRange = src_range,
+            };
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
+                                   vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, post_barrier);
+        });
+        msaa_copy_resources.push_back(MSAACopyResources{
+            .tick = scheduler.CurrentTick(),
+            .src_view = std::move(src_view),
+            .dst_view = std::move(dst_view),
+            .framebuffer = std::move(framebuffer),
+        });
+        if (copy_stencil) {
+            msaa_copy_resources.push_back(MSAACopyResources{
+                .tick = scheduler.CurrentTick(),
+                .src_view = std::move(src_stencil_view),
+                .dst_view = vk::ImageView{},
+                .framebuffer = vk::Framebuffer{},
+            });
+        }
+    }
+}
+
 VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipelineKey& key) {
     const auto it = std::ranges::find(msaa_copy_keys, key);
     if (it != msaa_copy_keys.end()) {
@@ -1458,10 +1665,83 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *msaa_copy_pipelines.back();
 }
 
+VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyDepthPipeline(const MSAACopyPipelineKey& key,
+                                                               bool copy_stencil) {
+    auto& keys = copy_stencil ? msaa_copy_depth_stencil_keys : msaa_copy_depth_keys;
+    auto& pipelines = copy_stencil ? msaa_copy_depth_stencil_pipelines : msaa_copy_depth_pipelines;
+    const auto it = std::ranges::find(keys, key);
+    if (it != keys.end()) {
+        return *pipelines[std::distance(keys.begin(), it)];
+    }
+    keys.push_back(key);
+    const std::array stages =
+        MakeStages(*clear_color_vert, copy_stencil ? *convert_non_msaa_to_msaa_depth_stencil_frag
+                                                   : *convert_non_msaa_to_msaa_depth_frag);
+    const VkPipelineMultisampleStateCreateInfo multisample_ci{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .rasterizationSamples = key.samples,
+        .sampleShadingEnable = VK_TRUE,
+        .minSampleShading = 1.0f,
+        .pSampleMask = nullptr,
+        .alphaToCoverageEnable = VK_FALSE,
+        .alphaToOneEnable = VK_FALSE,
+    };
+    static constexpr VkStencilOpState REPLACE_STENCIL_OP{
+        .failOp = VK_STENCIL_OP_REPLACE,
+        .passOp = VK_STENCIL_OP_REPLACE,
+        .depthFailOp = VK_STENCIL_OP_REPLACE,
+        .compareOp = VK_COMPARE_OP_ALWAYS,
+        .compareMask = 0xFF,
+        .writeMask = 0xFF,
+        .reference = 0,
+    };
+    const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .depthTestEnable = VK_TRUE,
+        .depthWriteEnable = VK_TRUE,
+        .depthCompareOp = VK_COMPARE_OP_ALWAYS,
+        .depthBoundsTestEnable = VK_FALSE,
+        .stencilTestEnable = copy_stencil ? VK_TRUE : VK_FALSE,
+        .front = copy_stencil ? REPLACE_STENCIL_OP : VkStencilOpState{},
+        .back = copy_stencil ? REPLACE_STENCIL_OP : VkStencilOpState{},
+        .minDepthBounds = 0.0f,
+        .maxDepthBounds = 0.0f,
+    };
+    const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci =
+        GetPipelineInputAssemblyStateCreateInfo(device);
+    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 = &input_assembly_ci,
+        .pTessellationState = nullptr,
+        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+        .pMultisampleState = &multisample_ci,
+        .pDepthStencilState = &depth_stencil_ci,
+        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO,
+        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+        .layout = copy_stencil ? *msaa_copy_depth_stencil_pipeline_layout
+                               : *msaa_copy_pipeline_layout,
+        .renderPass = key.renderpass,
+        .subpass = 0,
+        .basePipelineHandle = VK_NULL_HANDLE,
+        .basePipelineIndex = 0,
+    }, device.StaticPipelineCache()));
+    return *pipelines.back();
+}
+
 void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
     ConvertPipeline(pipeline, renderpass, false);
 }
@@ -1499,7 +1779,7 @@ void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass ren
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 void BlitImageHelper::ConvertPipelineColorTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
@@ -1542,7 +1822,7 @@ void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass rende
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/blit_image.h

@@ -116,6 +116,11 @@ public:
                   VideoCore::Surface::PixelFormat src_format, u32 num_samples,
                   std::span<const VideoCommon::ImageCopy> copies, bool msaa_to_non_msaa);
 
+    void CopyMSAADepth(RenderPassCache& render_pass_cache, VkImage dst_image,
+                       VideoCore::Surface::PixelFormat dst_format, VkImage src_image,
+                       VideoCore::Surface::PixelFormat src_format, u32 num_samples,
+                       std::span<const VideoCommon::ImageCopy> copies, bool copy_stencil);
+
 private:
     void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
                  const ImageView& src_image_view);
@@ -131,6 +136,9 @@ private:
     [[nodiscard]] VkPipeline FindOrEmplaceClearStencilPipeline(
         const BlitDepthStencilPipelineKey& key);
     [[nodiscard]] VkPipeline FindOrEmplaceMSAACopyPipeline(const MSAACopyPipelineKey& key);
+
+    [[nodiscard]] VkPipeline FindOrEmplaceMSAACopyDepthPipeline(const MSAACopyPipelineKey& key,
+                                                                bool copy_stencil);
     [[nodiscard]] VkPipeline FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPipelineKey& key);
     [[nodiscard]] VkPipeline FindOrEmplaceResolveDepthStencilPipeline(VkRenderPass renderpass,
                                                                       bool resolve_stencil);
@@ -162,6 +170,7 @@ private:
     vk::PipelineLayout two_textures_pipeline_layout;
     vk::PipelineLayout clear_color_pipeline_layout;
     vk::PipelineLayout msaa_copy_pipeline_layout;
+    vk::PipelineLayout msaa_copy_depth_stencil_pipeline_layout;
     vk::ShaderModule full_screen_vert;
     vk::ShaderModule blit_color_to_color_frag;
     vk::ShaderModule blit_color_msaa_frag;
@@ -180,6 +189,8 @@ private:
     vk::ShaderModule convert_s8d24_to_abgr8_frag;
     vk::ShaderModule convert_msaa_to_non_msaa_frag;
     vk::ShaderModule convert_non_msaa_to_msaa_frag;
+    vk::ShaderModule convert_non_msaa_to_msaa_depth_frag;
+    vk::ShaderModule convert_non_msaa_to_msaa_depth_stencil_frag;
     vk::Sampler linear_sampler;
     vk::Sampler nearest_sampler;
 
@@ -193,6 +204,10 @@ private:
     std::vector<vk::Pipeline> clear_stencil_pipelines;
     std::vector<MSAACopyPipelineKey> msaa_copy_keys;
     std::vector<vk::Pipeline> msaa_copy_pipelines;
+    std::vector<MSAACopyPipelineKey> msaa_copy_depth_keys;
+    std::vector<vk::Pipeline> msaa_copy_depth_pipelines;
+    std::vector<MSAACopyPipelineKey> msaa_copy_depth_stencil_keys;
+    std::vector<vk::Pipeline> msaa_copy_depth_stencil_pipelines;
     std::vector<BlitMSAAPipelineKey> blit_msaa_color_keys;
     std::vector<vk::Pipeline> blit_msaa_color_pipelines;
     std::vector<VkRenderPass> resolve_depth_keys;

src/video_core/renderer_vulkan/present/sgsr.cpp

@@ -1,6 +1,8 @@
 // SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
+#include <algorithm>
+
 #include "common/common_types.h"
 #include "common/div_ceil.h"
 #include "common/settings.h"
@@ -17,7 +19,7 @@
 
 namespace Vulkan {
 
-using PushConstants = std::array<u32, 4 + 2 + 1>;
+using PushConstants = std::array<u32, 4 + 2 + 2 + 1>;
 
 SGSR::SGSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count, VkExtent2D extent, bool edge_dir)
     : m_memory_allocator{memory_allocator}
@@ -100,26 +102,28 @@ VkImageView SGSR::Draw(const Device& device, Scheduler& scheduler, size_t image_
 
     const f32 input_image_width = f32(input_image_extent.width);
     const f32 input_image_height = f32(input_image_extent.height);
-    const f32 viewport_width = (crop_rect.right - crop_rect.left) * input_image_width;
-    const f32 viewport_height = (crop_rect.bottom - crop_rect.top) * input_image_height;
-    // expected [0, 2]
-    const f32 sharpening = f32(Settings::values.fsr_sharpening_slider.GetValue()) / 100.0f;
+    const f32 crop_width = (crop_rect.right - crop_rect.left) * input_image_width;
+    const f32 crop_height = (crop_rect.bottom - crop_rect.top) * input_image_height;
+    static constexpr f32 EDGE_SHARPNESS_MAX = 2.0f;
+    const f32 edge_sharpness =
+        EDGE_SHARPNESS_MAX - f32(Settings::values.fsr_sharpening_slider.GetValue()) / 200.0f;
 
-    // p = (tex * viewport) / input = [0,n] (normalized texcoords)
-    // p * input = [0,1024], [0,768]
     // layout( push_constant ) uniform constants {
     //     highp vec4 ViewportInfo[1];
     //     highp vec2 ResizeFactor;
+    //     highp vec2 CropOffset;
     //     highp float EdgeSharpness;
     // };
     PushConstants viewport_con{};
-    viewport_con[0] = std::bit_cast<u32>(std::abs(1.f / viewport_width));
-    viewport_con[1] = std::bit_cast<u32>(std::abs(1.f / viewport_height));
-    viewport_con[2] = std::bit_cast<u32>(std::abs(viewport_width));
-    viewport_con[3] = std::bit_cast<u32>(std::abs(viewport_height));
-    viewport_con[4] = std::bit_cast<u32>(viewport_width / input_image_width);
-    viewport_con[5] = std::bit_cast<u32>(viewport_height / input_image_height);
-    viewport_con[6] = std::bit_cast<u32>(sharpening);
+    viewport_con[0] = std::bit_cast<u32>(1.f / input_image_width);
+    viewport_con[1] = std::bit_cast<u32>(1.f / input_image_height);
+    viewport_con[2] = std::bit_cast<u32>(input_image_width);
+    viewport_con[3] = std::bit_cast<u32>(input_image_height);
+    viewport_con[4] = std::bit_cast<u32>(crop_width / input_image_width);
+    viewport_con[5] = std::bit_cast<u32>(crop_height / input_image_height);
+    viewport_con[6] = std::bit_cast<u32>((std::min)(crop_rect.left, crop_rect.right));
+    viewport_con[7] = std::bit_cast<u32>((std::min)(crop_rect.top, crop_rect.bottom));
+    viewport_con[8] = std::bit_cast<u32>(edge_sharpness);
 
     UploadImages(device, scheduler);
     UpdateDescriptorSets(device, source_image_view, image_index);

src/video_core/renderer_vulkan/present/util.cpp

@@ -491,7 +491,7 @@ static vk::Pipeline CreateWrappedPipelineImpl(
         .subpass = 0,
         .basePipelineHandle = 0,
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 vk::Pipeline CreateWrappedPipeline(const Device& device, vk::RenderPass& renderpass,

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -22,7 +22,9 @@
 #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h"
 #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
 #include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_2d_buffer_comp_spv.h"
 #include "video_core/host_shaders/block_linear_unswizzle_3d_bcn_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_3d_buffer_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/surface.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
@@ -268,7 +270,7 @@ ComputePass::ComputePass(const Device& device_, Scheduler& scheduler, Descriptor
         .layout = *layout,
         .basePipelineHandle = {},
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 ComputePass::~ComputePass() = default;
@@ -872,4 +874,479 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
     });
 }
 
+namespace {
+constexpr u32 BL2D_BINDING_INPUT_BUFFER = 0;
+constexpr u32 BL2D_BINDING_OUTPUT_BUFFER = 1;
+
+struct alignas(16) BlockLinearUnswizzle2DPushConstants {
+    std::array<u32, 3> dim;
+    u32 bytes_per_block_log2;
+    std::array<u32, 3> origin;
+    u32 layer_stride;
+    u32 block_size;
+    u32 x_shift;
+    u32 block_height;
+    u32 block_height_mask;
+};
+static_assert(sizeof(BlockLinearUnswizzle2DPushConstants) <= 128);
+
+constexpr std::array<VkDescriptorSetLayoutBinding, 2> BL2D_BINDINGS{{
+    {
+        .binding = BL2D_BINDING_INPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+    {
+        .binding = BL2D_BINDING_OUTPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+}};
+
+constexpr std::array<VkDescriptorUpdateTemplateEntry, 2> BL2D_TEMPLATE{{
+    {
+        .dstBinding = BL2D_BINDING_INPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL2D_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+    {
+        .dstBinding = BL2D_BINDING_OUTPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL2D_BINDING_OUTPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+}};
+
+constexpr DescriptorBankInfo BL2D_BANK_INFO{
+    .uniform_buffers = 0,
+    .storage_buffers = 2,
+    .texture_buffers = 0,
+    .image_buffers = 0,
+    .textures = 0,
+    .images = 0,
+    .score = 2,
+};
+} // Anonymous namespace
+
+BlockLinearUnswizzle2DPass::BlockLinearUnswizzle2DPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    StagingBufferPool& staging_buffer_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, BL2D_BINDINGS, BL2D_TEMPLATE,
+                  BL2D_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearUnswizzle2DPushConstants)>,
+                  BLOCK_LINEAR_UNSWIZZLE_2D_BUFFER_COMP_SPV),
+      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+BlockLinearUnswizzle2DPass::~BlockLinearUnswizzle2DPass() = default;
+
+bool BlockLinearUnswizzle2DPass::IsSupported(const VideoCommon::ImageInfo& info) {
+    if (info.type != VideoCommon::ImageType::e2D) {
+        return false;
+    }
+    if (info.resources.levels != 1 || info.resources.layers != 1) {
+        return false;
+    }
+    if (info.num_samples > 1) {
+        return false;
+    }
+    if (VideoCore::Surface::IsPixelFormatASTC(info.format) ||
+        VideoCore::Surface::IsPixelFormatBCn(info.format)) {
+        return false;
+    }
+    const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(info.format);
+    if (bytes_per_block != 4 && bytes_per_block != 8 && bytes_per_block != 16) {
+        return false;
+    }
+    return VideoCore::Surface::DefaultBlockWidth(info.format) == 1 &&
+           VideoCore::Surface::DefaultBlockHeight(info.format) == 1;
+}
+
+void BlockLinearUnswizzle2DPass::Unswizzle(
+    Image& image, const StagingBufferRef& swizzled,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    if (swizzles.empty()) {
+        return;
+    }
+    const VideoCommon::SwizzleParameters& sw = swizzles.front();
+    const auto params = VideoCommon::Accelerated::MakeBlockLinearSwizzle2DParams(sw, image.info);
+
+    const u32 width = sw.num_tiles.width;
+    const u32 height = sw.num_tiles.height;
+    const u32 depth = image.info.resources.layers;
+    const u32 bytes_per_block = 1u << params.bytes_per_block_log2;
+    const VkDeviceSize output_size =
+        static_cast<VkDeviceSize>(width) * height * depth * bytes_per_block;
+
+    const StagingBufferRef output =
+        staging_buffer_pool.Request(static_cast<size_t>(output_size), MemoryUsage::DeviceLocal);
+
+    BlockLinearUnswizzle2DPushConstants pc{};
+    pc.dim = {width, height, depth};
+    pc.bytes_per_block_log2 = params.bytes_per_block_log2;
+    pc.origin = params.origin;
+    pc.layer_stride = params.layer_stride;
+    pc.block_size = params.block_size;
+    pc.x_shift = params.x_shift;
+    pc.block_height = params.block_height;
+    pc.block_height_mask = params.block_height_mask;
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+
+    compute_pass_descriptor_queue.Acquire(scheduler, 2);
+    compute_pass_descriptor_queue.AddBuffer(swizzled.buffer, sw.buffer_offset + swizzled.offset,
+                                            image.guest_size_bytes - sw.buffer_offset);
+    compute_pass_descriptor_queue.AddBuffer(output.buffer, output.offset, output_size);
+
+    const void* descriptor_data = compute_pass_descriptor_queue.UpdateData();
+    const VkDescriptorSet set = descriptor_allocator.Commit();
+
+    const u32 gx = Common::DivCeil(width, 16u);
+    const u32 gy = Common::DivCeil(height, 8u);
+    const bool is_initialized = image.ExchangeInitialization();
+
+    const VkBuffer out_buffer = output.buffer;
+    const VkDeviceSize out_offset = output.offset;
+    const VkImage dst_image = image.Handle();
+    const VkImageAspectFlags aspect = image.AspectMask();
+
+    scheduler.Record([this, set, descriptor_data, pc, gx, gy, depth, output_size, out_buffer,
+                      out_offset, dst_image, aspect, width, height,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        if (dst_image == VK_NULL_HANDLE || out_buffer == VK_NULL_HANDLE) {
+            return;
+        }
+        device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
+        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+        cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(pc), &pc);
+        cmdbuf.Dispatch(gx, gy, depth);
+
+        const VkBufferMemoryBarrier buffer_barrier{
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .buffer = out_buffer,
+            .offset = out_offset,
+            .size = output_size,
+        };
+        const VkImageMemoryBarrier pre_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_READ_BIT
+                                                                      : VK_ACCESS_NONE),
+            .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                                   (is_initialized ? vk::PIPELINE_STAGE_GRAPHICS_COMPUTE
+                                                   : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
+                               VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, buffer_barrier, pre_copy);
+
+        const VkBufferImageCopy copy{
+            .bufferOffset = out_offset,
+            .bufferRowLength = 0,
+            .bufferImageHeight = 0,
+            .imageSubresource{
+                .aspectMask = aspect,
+                .mipLevel = 0,
+                .baseArrayLayer = 0,
+                .layerCount = depth,
+            },
+            .imageOffset = {0, 0, 0},
+            .imageExtent = {width, height, 1},
+        };
+        cmdbuf.CopyBufferToImage(out_buffer, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
+
+        const VkImageMemoryBarrier post_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
+                               0, {}, {}, post_copy);
+    });
+}
+
+namespace {
+constexpr u32 BL3DB_BINDING_INPUT_BUFFER = 0;
+constexpr u32 BL3DB_BINDING_OUTPUT_BUFFER = 1;
+
+struct alignas(16) BlockLinearUnswizzle3DBufferPushConstants {
+    std::array<u32, 3> dim;
+    u32 bytes_per_block_log2;
+    std::array<u32, 3> origin;
+    u32 slice_size;
+    u32 block_size;
+    u32 x_shift;
+    u32 block_height;
+    u32 block_height_mask;
+    u32 block_depth;
+    u32 block_depth_mask;
+};
+static_assert(sizeof(BlockLinearUnswizzle3DBufferPushConstants) <= 128);
+
+constexpr std::array<VkDescriptorSetLayoutBinding, 2> BL3DB_BINDINGS{{
+    {
+        .binding = BL3DB_BINDING_INPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+    {
+        .binding = BL3DB_BINDING_OUTPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+}};
+
+constexpr std::array<VkDescriptorUpdateTemplateEntry, 2> BL3DB_TEMPLATE{{
+    {
+        .dstBinding = BL3DB_BINDING_INPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL3DB_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+    {
+        .dstBinding = BL3DB_BINDING_OUTPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL3DB_BINDING_OUTPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+}};
+
+constexpr DescriptorBankInfo BL3DB_BANK_INFO{
+    .uniform_buffers = 0,
+    .storage_buffers = 2,
+    .texture_buffers = 0,
+    .image_buffers = 0,
+    .textures = 0,
+    .images = 0,
+    .score = 2,
+};
+} // Anonymous namespace
+
+BlockLinearUnswizzle3DBufferPass::BlockLinearUnswizzle3DBufferPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    StagingBufferPool& staging_buffer_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, BL3DB_BINDINGS, BL3DB_TEMPLATE,
+                  BL3DB_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearUnswizzle3DBufferPushConstants)>,
+                  BLOCK_LINEAR_UNSWIZZLE_3D_BUFFER_COMP_SPV),
+      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+BlockLinearUnswizzle3DBufferPass::~BlockLinearUnswizzle3DBufferPass() = default;
+
+bool BlockLinearUnswizzle3DBufferPass::IsSupported(const Device& device,
+                                                   const VideoCommon::ImageInfo& info) {
+    if (info.type != VideoCommon::ImageType::e3D) {
+        return false;
+    }
+    if (info.resources.levels != 1 || info.resources.layers != 1) {
+        return false;
+    }
+    if (info.num_samples > 1) {
+        return false;
+    }
+    if (info.size.depth <= 1) {
+        return false;
+    }
+    if (VideoCore::Surface::IsPixelFormatASTC(info.format)) {
+        return false;
+    }
+    if (VideoCore::Surface::IsPixelFormatBCn(info.format) && !device.IsOptimalBcnSupported()) {
+        return false;
+    }
+    const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(info.format);
+    return bytes_per_block == 4 || bytes_per_block == 8 || bytes_per_block == 16;
+}
+
+void BlockLinearUnswizzle3DBufferPass::Unswizzle(
+    Image& image, const StagingBufferRef& swizzled,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    if (swizzles.empty()) {
+        return;
+    }
+
+    const VideoCommon::SwizzleParameters& sw = swizzles.front();
+    const auto params = VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams(sw, image.info);
+
+    const u32 blocks_x = sw.num_tiles.width;
+    const u32 blocks_y = sw.num_tiles.height;
+    const u32 blocks_z = sw.num_tiles.depth;
+    const u32 bytes_per_block = 1u << params.bytes_per_block_log2;
+    const VkDeviceSize output_size =
+        static_cast<VkDeviceSize>(blocks_x) * blocks_y * blocks_z * bytes_per_block;
+
+    const StagingBufferRef output =
+        staging_buffer_pool.Request(static_cast<size_t>(output_size), MemoryUsage::DeviceLocal);
+
+    BlockLinearUnswizzle3DBufferPushConstants pc{};
+    pc.dim = {blocks_x, blocks_y, blocks_z};
+    pc.bytes_per_block_log2 = params.bytes_per_block_log2;
+    pc.origin = params.origin;
+    pc.slice_size = params.slice_size;
+    pc.block_size = params.block_size;
+    pc.x_shift = params.x_shift;
+    pc.block_height = params.block_height;
+    pc.block_height_mask = params.block_height_mask;
+    pc.block_depth = params.block_depth;
+    pc.block_depth_mask = params.block_depth_mask;
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+
+    compute_pass_descriptor_queue.Acquire(scheduler, 2);
+    compute_pass_descriptor_queue.AddBuffer(swizzled.buffer, sw.buffer_offset + swizzled.offset,
+                                            image.guest_size_bytes - sw.buffer_offset);
+    compute_pass_descriptor_queue.AddBuffer(output.buffer, output.offset, output_size);
+
+    const void* descriptor_data = compute_pass_descriptor_queue.UpdateData();
+    const VkDescriptorSet set = descriptor_allocator.Commit();
+
+    const u32 gx = Common::DivCeil(blocks_x, 8u);
+    const u32 gy = Common::DivCeil(blocks_y, 8u);
+    const u32 gz = Common::DivCeil(blocks_z, 4u);
+    const bool is_initialized = image.ExchangeInitialization();
+
+    const VkBuffer out_buffer = output.buffer;
+    const VkDeviceSize out_offset = output.offset;
+    const VkImage dst_image = image.Handle();
+    const VkImageAspectFlags aspect = image.AspectMask();
+    const VkExtent3D extent{
+        .width = image.info.size.width,
+        .height = image.info.size.height,
+        .depth = image.info.size.depth,
+    };
+
+    scheduler.Record([this, set, descriptor_data, pc, gx, gy, gz, output_size, out_buffer,
+                      out_offset, dst_image, aspect, extent,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        if (dst_image == VK_NULL_HANDLE || out_buffer == VK_NULL_HANDLE) {
+            return;
+        }
+        device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
+        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+        cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(pc), &pc);
+        cmdbuf.Dispatch(gx, gy, gz);
+
+        const VkBufferMemoryBarrier buffer_barrier{
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .buffer = out_buffer,
+            .offset = out_offset,
+            .size = output_size,
+        };
+        const VkImageMemoryBarrier pre_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_READ_BIT
+                                                                      : VK_ACCESS_NONE),
+            .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                                   (is_initialized ? vk::PIPELINE_STAGE_GRAPHICS_COMPUTE
+                                                   : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
+                               VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, buffer_barrier, pre_copy);
+
+        const VkBufferImageCopy copy{
+            .bufferOffset = out_offset,
+            .bufferRowLength = 0,
+            .bufferImageHeight = 0,
+            .imageSubresource{
+                .aspectMask = aspect,
+                .mipLevel = 0,
+                .baseArrayLayer = 0,
+                .layerCount = 1,
+            },
+            .imageOffset = {0, 0, 0},
+            .imageExtent = extent,
+        };
+        cmdbuf.CopyBufferToImage(out_buffer, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
+
+        const VkImageMemoryBarrier post_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
+                               0, {}, {}, post_copy);
+    });
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -164,4 +164,42 @@ private:
     ComputePassDescriptorQueue& compute_pass_descriptor_queue;
 };
 
+class BlockLinearUnswizzle2DPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzle2DPass(const Device& device_, Scheduler& scheduler_,
+                                        DescriptorPool& descriptor_pool_,
+                                        StagingBufferPool& staging_buffer_pool_,
+                                        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzle2DPass();
+
+    [[nodiscard]] static bool IsSupported(const VideoCommon::ImageInfo& info);
+
+    void Unswizzle(Image& image, const StagingBufferRef& swizzled,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
+class BlockLinearUnswizzle3DBufferPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzle3DBufferPass(
+        const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+        StagingBufferPool& staging_buffer_pool_,
+        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzle3DBufferPass();
+
+    [[nodiscard]] static bool IsSupported(const Device& device, const VideoCommon::ImageInfo& info);
+
+    void Unswizzle(Image& image, const StagingBufferRef& swizzled,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -140,6 +140,8 @@ RenderPassKey MakeRenderPassKey(const FixedPipelineState& state, const Device& d
     });
     key.resolve_color =
         key.samples != VK_SAMPLE_COUNT_1_BIT && has_color && device.IsTiler();
+    key.resolve_depth_stencil = key.samples != VK_SAMPLE_COUNT_1_BIT && device.IsTiler() &&
+                                SupportsDepthStencilResolve(device, key.depth_format);
     return key;
 }
 

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -63,6 +63,8 @@ using VideoCommon::GenericEnvironment;
 using VideoCommon::GraphicsEnvironment;
 
 constexpr u32 CACHE_VERSION = 18;
+constexpr size_t VULKAN_CACHE_FLUSH_PIPELINES = 128;
+constexpr size_t VULKAN_CACHE_FLUSH_MIN_SECONDS = 30;
 constexpr std::array<char, 8> VULKAN_CACHE_MAGIC_NUMBER{'y', 'u', 'z', 'u', 'v', 'k', 'c', 'h'};
 
 template <typename Container>
@@ -150,6 +152,25 @@ Shader::AttributeType AttributeType(const FixedPipelineState& state, size_t inde
     return Shader::AttributeType::Disabled;
 }
 
+Shader::InputTopology MaxwellToInputTopology(Maxwell::PrimitiveTopology topology) {
+    switch (topology) {
+    case Maxwell::PrimitiveTopology::Points:
+        return Shader::InputTopology::Points;
+    case Maxwell::PrimitiveTopology::Lines:
+    case Maxwell::PrimitiveTopology::LineLoop:
+    case Maxwell::PrimitiveTopology::LineStrip:
+        return Shader::InputTopology::Lines;
+    case Maxwell::PrimitiveTopology::LinesAdjacency:
+    case Maxwell::PrimitiveTopology::LineStripAdjacency:
+        return Shader::InputTopology::LinesAdjacency;
+    case Maxwell::PrimitiveTopology::TrianglesAdjacency:
+    case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
+        return Shader::InputTopology::TrianglesAdjacency;
+    default:
+        return Shader::InputTopology::Triangles;
+    }
+}
+
 Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> programs,
                                     const GraphicsPipelineCacheKey& key,
                                     const Shader::IR::Program& program,
@@ -268,33 +289,7 @@ Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> program
     default:
         break;
     }
-    switch (key.state.topology) {
-    case Maxwell::PrimitiveTopology::Points:
-        info.input_topology = Shader::InputTopology::Points;
-        break;
-    case Maxwell::PrimitiveTopology::Lines:
-    case Maxwell::PrimitiveTopology::LineLoop:
-    case Maxwell::PrimitiveTopology::LineStrip:
-        info.input_topology = Shader::InputTopology::Lines;
-        break;
-    case Maxwell::PrimitiveTopology::Triangles:
-    case Maxwell::PrimitiveTopology::TriangleStrip:
-    case Maxwell::PrimitiveTopology::TriangleFan:
-    case Maxwell::PrimitiveTopology::Quads:
-    case Maxwell::PrimitiveTopology::QuadStrip:
-    case Maxwell::PrimitiveTopology::Polygon:
-    case Maxwell::PrimitiveTopology::Patches:
-        info.input_topology = Shader::InputTopology::Triangles;
-        break;
-    case Maxwell::PrimitiveTopology::LinesAdjacency:
-    case Maxwell::PrimitiveTopology::LineStripAdjacency:
-        info.input_topology = Shader::InputTopology::LinesAdjacency;
-        break;
-    case Maxwell::PrimitiveTopology::TrianglesAdjacency:
-    case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
-        info.input_topology = Shader::InputTopology::TrianglesAdjacency;
-        break;
-    }
+    info.input_topology = MaxwellToInputTopology(key.state.topology);
     info.force_early_z = key.state.early_z != 0;
     info.y_negate = key.state.y_negate != 0;
     return info;
@@ -699,6 +694,10 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
     if (use_vulkan_pipeline_cache) {
         SerializeVulkanPipelineCache(vulkan_pipeline_cache_filename, vulkan_pipeline_cache,
                                      CACHE_VERSION);
+        size_t size = 0;
+        vulkan_pipeline_cache.Read(&size, nullptr);
+        last_cache_size.store(size, std::memory_order_relaxed);
+        last_flush = std::chrono::steady_clock::now();
     }
 
     if (state.statistics) {
@@ -706,6 +705,35 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
     }
 }
 
+void PipelineCache::QueueVulkanPipelineCacheFlush() {
+    if (!use_vulkan_pipeline_cache || vulkan_pipeline_cache_filename.empty()) {
+        return;
+    }
+    if (++pipelines_since_flush < VULKAN_CACHE_FLUSH_PIPELINES) {
+        return;
+    }
+    const auto now = std::chrono::steady_clock::now();
+    const auto megabytes = last_cache_size.load(std::memory_order_relaxed) / (1024 * 1024);
+    const std::chrono::seconds interval{
+        std::max<size_t>(VULKAN_CACHE_FLUSH_MIN_SECONDS, megabytes)};
+    if (last_flush.time_since_epoch().count() != 0 && now - last_flush < interval) {
+        return;
+    }
+    if (flush_in_flight.exchange(true, std::memory_order_acq_rel)) {
+        return;
+    }
+    pipelines_since_flush = 0;
+    last_flush = now;
+    serialization_thread.QueueWork([this] {
+        SerializeVulkanPipelineCache(vulkan_pipeline_cache_filename, vulkan_pipeline_cache,
+                                     CACHE_VERSION);
+        size_t size = 0;
+        vulkan_pipeline_cache.Read(&size, nullptr);
+        last_cache_size.store(size, std::memory_order_relaxed);
+        flush_in_flight.store(false, std::memory_order_release);
+    });
+}
+
 GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() {
     const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)};
     auto& pipeline{pair->second};
@@ -744,7 +772,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
     bool build_in_parallel) try {
     auto hash = key.Hash();
-    LOG_INFO(Render_Vulkan, "{:#016x}", hash);
+    LOG_DEBUG(Render_Vulkan, "{:#016x}", hash);
     size_t env_index{0};
     std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
     const bool uses_vertex_a{key.unique_hashes[0] != 0};
@@ -758,8 +786,10 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
                                        index == static_cast<u32>(Maxwell::ShaderType::Geometry);
         if (key.unique_hashes[index] == 0 && is_emulated_stage) {
             auto topology = MaxwellToOutputTopology(key.state.topology);
-            programs[index] = GenerateGeometryPassthrough(pools.inst, pools.block, host_info,
-                                                          *layer_source_program, topology);
+            programs[index] =
+                GenerateGeometryPassthrough(pools.inst, pools.block, host_info,
+                                            *layer_source_program, topology,
+                                            MaxwellToInputTopology(key.state.topology));
             continue;
         }
         if (key.unique_hashes[index] == 0) {
@@ -772,11 +802,13 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
         Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0);
         if (!uses_vertex_a || index != 1) {
             // Normal path
-            programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info);
+            programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info,
+                                              MaxwellToInputTopology(key.state.topology));
         } else {
             // VertexB path when VertexA is present.
             auto& program_va{programs[0]};
-            auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
+            auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info,
+                                             MaxwellToInputTopology(key.state.topology))};
             programs[index] = MergeDualVertexPrograms(program_va, program_vb, env);
         }
 
@@ -880,6 +912,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
         }
         SerializePipeline(key, env_ptrs, pipeline_cache_filename, CACHE_VERSION);
     });
+    QueueVulkanPipelineCacheFlush();
     return pipeline;
 }
 
@@ -899,6 +932,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
         SerializePipeline(key, std::array<const GenericEnvironment*, 1>{&env_},
                           pipeline_cache_filename, CACHE_VERSION);
     });
+    QueueVulkanPipelineCacheFlush();
     return pipeline;
 }
 
@@ -911,7 +945,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
         return nullptr;
     }
 
-    LOG_INFO(Render_Vulkan, "{:#016x}", hash);
+    LOG_DEBUG(Render_Vulkan, "{:#016x}", hash);
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
 
@@ -920,7 +954,8 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
         env.Dump(hash, key.unique_hash);
     }
 
-    auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
+    auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info,
+                                  Shader::InputTopology::Points)};
     const VkDriverIdKHR driver_id = device.GetDriverID();
     const bool needs_shared_mem_clamp =
         driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY ||

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -7,6 +7,8 @@
 #pragma once
 
 #include <array>
+#include <atomic>
+#include <chrono>
 #include <cstddef>
 #include <filesystem>
 #include <memory>
@@ -144,6 +146,8 @@ private:
     vk::PipelineCache LoadVulkanPipelineCache(const std::filesystem::path& filename,
                                               u32 expected_cache_version);
 
+    void QueueVulkanPipelineCacheFlush();
+
     const Device& device;
     Scheduler& scheduler;
     DescriptorPool& descriptor_pool;
@@ -171,6 +175,10 @@ private:
 
     std::filesystem::path vulkan_pipeline_cache_filename;
     vk::PipelineCache vulkan_pipeline_cache;
+    size_t pipelines_since_flush{};
+    std::chrono::steady_clock::time_point last_flush{};
+    std::atomic<size_t> last_cache_size{};
+    std::atomic_bool flush_in_flight{};
 
     Common::ThreadWorker workers;
     Common::ThreadWorker serialization_thread;

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -725,6 +725,9 @@ public:
         if (!device.IsExtTransformFeedbackSupported()) {
             return;
         }
+        if (!scheduler.IsRenderPassActive()) {
+            return;
+        }
         FlushBeginTFB();
         has_started = true;
     }
@@ -741,6 +744,9 @@ public:
         if (has_flushed_end_pending) {
             if (scheduler.IsRenderPassActive()) {
                 FlushEndTFB();
+            } else {
+                has_flushed_end_pending = false;
+                has_started = false;
             }
         }
         runtime.View3DRegs([this](Maxwell3D& maxwell3d) {

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -1286,33 +1286,42 @@ void RasterizerVulkan::UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs) {
                         regs.zeta.format == Tegra::DepthFormat::X8Z24_UNORM ||
                         regs.zeta.format == Tegra::DepthFormat::S8Z24_UNORM ||
                         regs.zeta.format == Tegra::DepthFormat::V8Z24_UNORM;
+    const bool forces_unorm_representation = device.IsExtDepthBiasControlSupported();
 
     if (is_d24 && !device.SupportsD24DepthBuffer()) {
-        static constexpr const size_t length = sizeof(NEEDS_D24) / sizeof(NEEDS_D24[0]);
+        static constexpr double GUEST_TO_HOST_UNORM_BITS =
+            static_cast<double>(1ULL << (32 - 24));
 
-        static constexpr const u64* start = NEEDS_D24;
-        static constexpr const u64* end = NEEDS_D24 + length;
+        if (forces_unorm_representation) {
+            units = static_cast<float>(static_cast<double>(units) * GUEST_TO_HOST_UNORM_BITS);
+        } else {
+            static constexpr const size_t length = sizeof(NEEDS_D24) / sizeof(NEEDS_D24[0]);
 
-        const u64* it = std::find(start, end, program_id);
+            static constexpr const u64* start = NEEDS_D24;
+            static constexpr const u64* end = NEEDS_D24 + length;
 
-        if (it != end) {
-            // the base formulas can be obtained from here:
-            //   https://docs.microsoft.com/en-us/windows/win32/direct3d11/d3d10-graphics-programming-guide-output-merger-stage-depth-bias
-            const double rescale_factor =
-                static_cast<double>(1ULL << (32 - 24)) / (static_cast<double>(0x1.ep+127));
-            units = static_cast<float>(static_cast<double>(units) * rescale_factor);
+            const u64* it = std::find(start, end, program_id);
+
+            if (it != end) {
+                // the base formulas can be obtained from here:
+                //   https://docs.microsoft.com/en-us/windows/win32/direct3d11/d3d10-graphics-programming-guide-output-merger-stage-depth-bias
+                const double rescale_factor =
+                    GUEST_TO_HOST_UNORM_BITS / (static_cast<double>(0x1.ep+127));
+                units = static_cast<float>(static_cast<double>(units) * rescale_factor);
+            }
         }
     }
 
     scheduler.Record([constant = units, clamp = regs.depth_bias_clamp,
-                      factor = regs.slope_scale_depth_bias, this](vk::CommandBuffer cmdbuf) {
-        if (device.IsExtDepthBiasControlSupported()) {
-            static VkDepthBiasRepresentationInfoEXT bias_info{
+                      factor = regs.slope_scale_depth_bias,
+                      forces_unorm_representation, this](vk::CommandBuffer cmdbuf) {
+        if (forces_unorm_representation) {
+            const VkDepthBiasRepresentationInfoEXT bias_info{
                 .sType = VK_STRUCTURE_TYPE_DEPTH_BIAS_REPRESENTATION_INFO_EXT,
                 .pNext = nullptr,
                 .depthBiasRepresentation =
                     VK_DEPTH_BIAS_REPRESENTATION_LEAST_REPRESENTABLE_VALUE_FORCE_UNORM_EXT,
-                .depthBiasExact = VK_FALSE,
+                .depthBiasExact = static_cast<VkBool32>(device.HasExactDepthBiasControl()),
             };
 
             cmdbuf.SetDepthBias(constant, clamp, factor, &bias_info);
@@ -1336,8 +1345,14 @@ void RasterizerVulkan::UpdateDepthBounds(Tegra::Engines::Maxwell3D::Regs& regs)
     if (!state_tracker.TouchDepthBounds()) {
         return;
     }
-    scheduler.Record([min = regs.depth_bounds[0], max = regs.depth_bounds[1]](
-                         vk::CommandBuffer cmdbuf) { cmdbuf.SetDepthBounds(min, max); });
+    float min = regs.depth_bounds[0];
+    float max = regs.depth_bounds[1];
+    if (!device.IsExtDepthRangeUnrestrictedSupported()) {
+        min = std::clamp(min, 0.0f, 1.0f);
+        max = std::clamp(max, 0.0f, 1.0f);
+    }
+    scheduler.Record(
+        [min, max](vk::CommandBuffer cmdbuf) { cmdbuf.SetDepthBounds(min, max); });
 }
 
 void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs) {

src/video_core/renderer_vulkan/vk_render_pass_cache.cpp

@@ -65,8 +65,61 @@ using VideoCore::Surface::SurfaceType;
                 .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
             };
         }
+
+        struct ResolveAspects {
+            bool depth;
+            bool stencil;
+        };
+
+        struct ResolveModes {
+            VkResolveModeFlagBits depth;
+            VkResolveModeFlagBits stencil;
+        };
+
+        constexpr ResolveAspects GetResolveAspects(PixelFormat format) {
+            const SurfaceType surface_type = GetSurfaceType(format);
+            return ResolveAspects{
+                .depth = surface_type == SurfaceType::Depth ||
+                         surface_type == SurfaceType::DepthStencil,
+                .stencil = surface_type == SurfaceType::Stencil ||
+                           surface_type == SurfaceType::DepthStencil,
+            };
+        }
+
+        ResolveModes PickResolveModes(const Device& device, PixelFormat format) {
+            constexpr VkResolveModeFlagBits mode = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT;
+
+            const ResolveAspects aspects = GetResolveAspects(format);
+            ResolveModes modes{
+                .depth = VK_RESOLVE_MODE_NONE,
+                .stencil = VK_RESOLVE_MODE_NONE,
+            };
+            if (aspects.depth && (device.GetDepthResolveModes() & mode) != 0) {
+                modes.depth = mode;
+            }
+            if (aspects.stencil && (device.GetStencilResolveModes() & mode) != 0) {
+                modes.stencil = mode;
+            }
+            return modes;
+        }
     } // Anonymous namespace
 
+bool SupportsDepthStencilResolve(const Device& device, PixelFormat depth_format) {
+    if (depth_format == PixelFormat::Invalid || !device.IsKhrDepthStencilResolveSupported()) {
+        return false;
+    }
+    const ResolveAspects aspects = GetResolveAspects(depth_format);
+    if (!aspects.depth && !aspects.stencil) {
+        return false;
+    }
+    const ResolveModes modes = PickResolveModes(device, depth_format);
+    if ((aspects.depth && modes.depth == VK_RESOLVE_MODE_NONE) ||
+        (aspects.stencil && modes.stencil == VK_RESOLVE_MODE_NONE)) {
+        return false;
+    }
+    return modes.depth == modes.stencil || device.SupportsIndependentResolveNone();
+}
+
 RenderPassCache::RenderPassCache(const Device& device_) : device{&device_} {}
 
 VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
@@ -75,7 +128,9 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
     if (!is_new) {
         return *pair->second;
     }
-    boost::container::static_vector<VkAttachmentDescription, 9> descriptions;
+    static constexpr size_t MAX_ATTACHMENTS =
+        2 * std::tuple_size_v<decltype(RenderPassKey::color_formats)> + 2;
+    boost::container::static_vector<VkAttachmentDescription, MAX_ATTACHMENTS> descriptions;
     std::array<VkAttachmentReference, 8> references{};
     u32 num_attachments{};
     u32 num_colors{};
@@ -133,6 +188,21 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
             }
         }
     }
+    const bool do_resolve_depth_stencil = key.resolve_depth_stencil && has_depth &&
+                                          key.samples != VK_SAMPLE_COUNT_1_BIT &&
+                                          SupportsDepthStencilResolve(*device, key.depth_format);
+    VkAttachmentReference depth_resolve_reference{};
+    if (do_resolve_depth_stencil) {
+        depth_resolve_reference = VkAttachmentReference{
+            .attachment = static_cast<u32>(descriptions.size()),
+            .layout = VK_IMAGE_LAYOUT_GENERAL,
+        };
+        VkAttachmentDescription resolve_desc =
+            AttachmentDescription(*device, key.depth_format, VK_SAMPLE_COUNT_1_BIT,
+                                  VK_ATTACHMENT_LOAD_OP_DONT_CARE, VK_ATTACHMENT_STORE_OP_STORE);
+        resolve_desc.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
+        descriptions.push_back(resolve_desc);
+    }
     const VkSubpassDescription subpass{
         .flags = 0,
         .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
@@ -145,18 +215,117 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
         .preserveAttachmentCount = 0,
         .pPreserveAttachments = nullptr,
     };
-    const VkSubpassDependency dependency{
-            .srcSubpass = 0,  // Current subpass
-            .dstSubpass = 0,  // Same subpass (self-dependency)
-            .srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
-            .dstStageMask = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
-            .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                             VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
-            .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT
+    const VkSubpassDependency counter_resume_dependency{
+        .srcSubpass = 0,
+        .dstSubpass = 0,
+        .srcStageMask = VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT,
+        .dstStageMask = VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT,
+        .srcAccessMask = VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT,
+        .dstAccessMask = VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_READ_BIT_EXT,
+        .dependencyFlags = 0,
     };
+    const bool can_resume_transform_feedback = device->IsExtTransformFeedbackSupported();
+
+    if (device->IsKhrCreateRenderPass2Supported()) {
+        boost::container::static_vector<VkAttachmentDescription2, MAX_ATTACHMENTS> descriptions2;
+        for (const VkAttachmentDescription& description : descriptions) {
+            descriptions2.push_back(VkAttachmentDescription2{
+                .sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2,
+                .pNext = nullptr,
+                .flags = description.flags,
+                .format = description.format,
+                .samples = description.samples,
+                .loadOp = description.loadOp,
+                .storeOp = description.storeOp,
+                .stencilLoadOp = description.stencilLoadOp,
+                .stencilStoreOp = description.stencilStoreOp,
+                .initialLayout = description.initialLayout,
+                .finalLayout = description.finalLayout,
+            });
+        }
+        const auto promote = [](const VkAttachmentReference& reference) {
+            return VkAttachmentReference2{
+                .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2,
+                .pNext = nullptr,
+                .attachment = reference.attachment,
+                .layout = reference.layout,
+                .aspectMask = 0,
+            };
+        };
+        std::array<VkAttachmentReference2, 8> references2{};
+        std::array<VkAttachmentReference2, 8> resolve_references2{};
+        for (size_t index = 0; index < references.size(); ++index) {
+            references2[index] = promote(references[index]);
+            resolve_references2[index] = promote(resolve_references[index]);
+        }
+        const VkAttachmentReference2 depth_reference2 = promote(depth_reference);
+        const VkAttachmentReference2 depth_resolve_reference2 = promote(depth_resolve_reference);
+        const ResolveModes resolve_modes = PickResolveModes(*device, key.depth_format);
+        const VkSubpassDescriptionDepthStencilResolve depth_stencil_resolve{
+            .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE,
+            .pNext = nullptr,
+            .depthResolveMode = resolve_modes.depth,
+            .stencilResolveMode = resolve_modes.stencil,
+            .pDepthStencilResolveAttachment = &depth_resolve_reference2,
+        };
+        const VkSubpassDescription2 subpass2{
+            .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2,
+            .pNext = do_resolve_depth_stencil ? &depth_stencil_resolve : nullptr,
+            .flags = 0,
+            .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
+            .viewMask = 0,
+            .inputAttachmentCount = 0,
+            .pInputAttachments = nullptr,
+            .colorAttachmentCount = num_attachments,
+            .pColorAttachments = references2.data(),
+            .pResolveAttachments = do_resolve_color ? resolve_references2.data() : nullptr,
+            .pDepthStencilAttachment = has_depth ? &depth_reference2 : nullptr,
+            .preserveAttachmentCount = 0,
+            .pPreserveAttachments = nullptr,
+        };
+        const VkMemoryBarrier2 counter_resume_barrier{
+            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2,
+            .pNext = nullptr,
+            .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT,
+            .srcAccessMask = VK_ACCESS_2_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT,
+            .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT,
+            .dstAccessMask = VK_ACCESS_2_TRANSFORM_FEEDBACK_COUNTER_READ_BIT_EXT,
+        };
+        VkSubpassDependency2 counter_resume_dependency2{
+            .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2,
+            .pNext = nullptr,
+            .srcSubpass = counter_resume_dependency.srcSubpass,
+            .dstSubpass = counter_resume_dependency.dstSubpass,
+            .srcStageMask = counter_resume_dependency.srcStageMask,
+            .dstStageMask = counter_resume_dependency.dstStageMask,
+            .srcAccessMask = counter_resume_dependency.srcAccessMask,
+            .dstAccessMask = counter_resume_dependency.dstAccessMask,
+            .dependencyFlags = counter_resume_dependency.dependencyFlags,
+            .viewOffset = 0,
+        };
+        if (device->HasSynchronization2()) {
+            counter_resume_dependency2.pNext = &counter_resume_barrier;
+            counter_resume_dependency2.srcStageMask = 0;
+            counter_resume_dependency2.dstStageMask = 0;
+            counter_resume_dependency2.srcAccessMask = 0;
+            counter_resume_dependency2.dstAccessMask = 0;
+        }
+        pair->second = device->GetLogical().CreateRenderPass2({
+            .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2,
+            .pNext = nullptr,
+            .flags = 0,
+            .attachmentCount = static_cast<u32>(descriptions2.size()),
+            .pAttachments = descriptions2.empty() ? nullptr : descriptions2.data(),
+            .subpassCount = 1,
+            .pSubpasses = &subpass2,
+            .dependencyCount = can_resume_transform_feedback ? 1u : 0u,
+            .pDependencies = can_resume_transform_feedback ? &counter_resume_dependency2 : nullptr,
+            .correlatedViewMaskCount = 0,
+            .pCorrelatedViewMasks = nullptr,
+        });
+        return *pair->second;
+    }
+
     pair->second = device->GetLogical().CreateRenderPass({
         .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
         .pNext = nullptr,
@@ -165,8 +334,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
         .pAttachments = descriptions.empty() ? nullptr : descriptions.data(),
         .subpassCount = 1,
         .pSubpasses = &subpass,
-        .dependencyCount = 1,
-        .pDependencies = &dependency,
+        .dependencyCount = can_resume_transform_feedback ? 1u : 0u,
+        .pDependencies = can_resume_transform_feedback ? &counter_resume_dependency : nullptr,
     });
     return *pair->second;
 }

src/video_core/renderer_vulkan/vk_render_pass_cache.h

@@ -9,6 +9,7 @@
 #include <mutex>
 #include <ankerl/unordered_dense.h>
 
+#include "common/container_hash.h"
 #include "video_core/surface.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
@@ -21,6 +22,7 @@ struct RenderPassKey {
     VideoCore::Surface::PixelFormat depth_format;
     VkSampleCountFlagBits samples;
     bool resolve_color;
+    bool resolve_depth_stencil;
     u32 color_clear_mask;
     bool depth_stencil_clear;
     u32 color_discard_mask;
@@ -31,17 +33,27 @@ struct RenderPassKey {
 namespace std {
 template <>
 struct hash<Vulkan::RenderPassKey> {
+    static_assert(std::tuple_size_v<decltype(Vulkan::RenderPassKey::color_formats)> <= 8);
+    static_assert(static_cast<u32>(VideoCore::Surface::PixelFormat::Invalid) <= 0xFF);
+    static_assert(static_cast<u32>(VideoCore::Surface::PixelFormat::Max) <= 0xFF);
+    static_assert(VK_SAMPLE_COUNT_64_BIT <= 0xFF);
+
     [[nodiscard]] size_t operator()(const Vulkan::RenderPassKey& key) const noexcept {
-        size_t value = static_cast<size_t>(key.depth_format) << 48;
-        value ^= static_cast<size_t>(key.samples) << 52;
-        value ^= static_cast<size_t>(key.resolve_color) << 63;
-        value ^= static_cast<size_t>(key.color_clear_mask) << 54;
-        value ^= static_cast<size_t>(key.depth_stencil_clear) << 62;
-        value ^= static_cast<size_t>(key.color_discard_mask) << 24;
-        for (size_t i = 0; i < key.color_formats.size(); ++i) {
-            value ^= static_cast<size_t>(key.color_formats[i]) << (i * 6);
+        u64 formats = 0;
+        for (size_t index = 0; index < key.color_formats.size(); ++index) {
+            formats |= static_cast<u64>(key.color_formats[index]) << (index * 8);
         }
-        return value;
+        const u64 state = static_cast<u64>(key.depth_format) |
+                          (static_cast<u64>(key.samples) << 8) |
+                          (static_cast<u64>(key.color_clear_mask) << 16) |
+                          (static_cast<u64>(key.color_discard_mask) << 24) |
+                          (static_cast<u64>(key.resolve_color) << 32) |
+                          (static_cast<u64>(key.depth_stencil_clear) << 33) |
+                          (static_cast<u64>(key.resolve_depth_stencil) << 34);
+        size_t seed = 0;
+        Common::HashCombine(seed, formats);
+        Common::HashCombine(seed, state);
+        return seed;
     }
 };
 } // namespace std
@@ -50,6 +62,9 @@ namespace Vulkan {
 
 class Device;
 
+[[nodiscard]] bool SupportsDepthStencilResolve(const Device& device,
+                                               VideoCore::Surface::PixelFormat depth_format);
+
 class RenderPassCache {
 public:
     explicit RenderPassCache(const Device& device_);

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -410,8 +410,17 @@ void Scheduler::EndRenderPass()
         Record([num_images = num_renderpass_images,
                        images = renderpass_images,
                        ranges = renderpass_image_ranges,
+                       consumer_stages = device.AttachmentConsumerStages(),
                        has_transform_feedback = device.IsExtTransformFeedbackSupported()](
                           vk::CommandBuffer cmdbuf) {
+            static constexpr VkAccessFlags SHADER_ACCESS =
+                VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
+            static constexpr VkAccessFlags COLOR_ACCESS =
+                VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+            static constexpr VkAccessFlags DEPTH_STENCIL_ACCESS =
+                VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
+                VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+
             std::array<VkImageMemoryBarrier, 9> barriers;
             for (size_t i = 0; i < num_images; ++i) {
                 const VkImageSubresourceRange& range = ranges[i];
@@ -421,24 +430,25 @@ void Scheduler::EndRenderPass()
                                                  | VK_IMAGE_ASPECT_STENCIL_BIT)) !=0;
 
                 VkAccessFlags src_access = 0;
+                VkAccessFlags dst_access = SHADER_ACCESS;
 
-                if (is_color)
+                if (is_color) {
                     src_access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
-                else if (is_depth_stencil)
+                    dst_access |= COLOR_ACCESS;
+                } else if (is_depth_stencil) {
                     src_access |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
-                else
+                    dst_access |= DEPTH_STENCIL_ACCESS;
+                } else {
                     src_access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
                                   | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+                    dst_access |= COLOR_ACCESS | DEPTH_STENCIL_ACCESS;
+                }
 
                 barriers[i] = VkImageMemoryBarrier{
                         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                         .pNext = nullptr,
                         .srcAccessMask = src_access,
-                        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT
-                                         | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT
-                                         | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
-                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
-                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+                        .dstAccessMask = dst_access,
                         .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                         .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -449,7 +459,7 @@ void Scheduler::EndRenderPass()
             }
             cmdbuf.EndRenderPass();
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                                   VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
+                                   VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, consumer_stages,
                                    0, nullptr, nullptr, vk::Span(barriers.data(), num_images));
             if (has_transform_feedback) {
                 static constexpr VkMemoryBarrier XFB_OUTPUT_BARRIER{

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -628,18 +628,12 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
             .subresourceRange = subresource_range,
     };
 
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                           VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                           VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
+    cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
                            read_barrier);
     cmdbuf.CopyBufferToImage(src_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copies);
     // TODO: Move this to another API
-    cmdbuf.PipelineBarrier(
-            VK_PIPELINE_STAGE_TRANSFER_BIT,
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
-            0, nullptr, nullptr, write_barrier);
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0,
+                           nullptr, nullptr, write_barrier);
 }
 
 [[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region,
@@ -933,6 +927,10 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
         bl3d_unswizzle_pass.emplace(device, scheduler, descriptor_pool,
                                    staging_buffer_pool, compute_pass_descriptor_queue);
     }
+    bl2d_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
+                                compute_pass_descriptor_queue);
+    bl3db_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
+                                 compute_pass_descriptor_queue);
 }
 
 void TextureCacheRuntime::Finish() {
@@ -995,13 +993,25 @@ VkBuffer TextureCacheRuntime::GetTemporaryBuffer(size_t needed_size) {
 }
 
 VkImageView TextureCacheRuntime::GetOrCreateResolveShadow(VkImage msaa_image, VkFormat format,
-                                                          VkExtent2D extent, u32 layers) {
+                                                          VkExtent2D extent, u32 layers,
+                                                          VkImageAspectFlags aspect_mask) {
     ResolveShadow& shadow = resolve_shadows[msaa_image];
     if (shadow.image && shadow.format == format && shadow.extent.width == extent.width &&
-        shadow.extent.height == extent.height && shadow.layers == layers) {
+        shadow.extent.height == extent.height && shadow.layers == layers &&
+        shadow.aspect_mask == aspect_mask) {
         shadow.up_to_date = true;
         return *shadow.view;
     }
+    VkImageUsageFlags shadow_usage =
+        VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
+    if ((aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) {
+        shadow_usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
+    } else {
+        shadow_usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
+    }
+    if (shadow.image) {
+        pending_resolve_shadows.emplace_back(scheduler.CurrentTick(), std::move(shadow));
+    }
     shadow.image = memory_allocator.CreateImage(VkImageCreateInfo{
         .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
         .pNext = nullptr,
@@ -1013,8 +1023,7 @@ VkImageView TextureCacheRuntime::GetOrCreateResolveShadow(VkImage msaa_image, Vk
         .arrayLayers = layers,
         .samples = VK_SAMPLE_COUNT_1_BIT,
         .tiling = VK_IMAGE_TILING_OPTIMAL,
-        .usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_SAMPLED_BIT |
-                 VK_IMAGE_USAGE_TRANSFER_SRC_BIT,
+        .usage = shadow_usage,
         .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,
@@ -1029,7 +1038,7 @@ VkImageView TextureCacheRuntime::GetOrCreateResolveShadow(VkImage msaa_image, Vk
         .format = format,
         .components{},
         .subresourceRange{
-            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .aspectMask = aspect_mask,
             .baseMipLevel = 0,
             .levelCount = 1,
             .baseArrayLayer = 0,
@@ -1039,6 +1048,7 @@ VkImageView TextureCacheRuntime::GetOrCreateResolveShadow(VkImage msaa_image, Vk
     shadow.format = format;
     shadow.extent = extent;
     shadow.layers = layers;
+    shadow.aspect_mask = aspect_mask;
     shadow.up_to_date = true;
     return *shadow.view;
 }
@@ -1060,7 +1070,14 @@ void TextureCacheRuntime::InvalidateResolveShadow(VkImage msaa_image) {
 }
 
 void TextureCacheRuntime::EraseResolveShadow(VkImage msaa_image) {
-    resolve_shadows.erase(msaa_image);
+    const auto it = resolve_shadows.find(msaa_image);
+    if (it == resolve_shadows.end()) {
+        return;
+    }
+    if (it->second.image) {
+        pending_resolve_shadows.emplace_back(scheduler.CurrentTick(), std::move(it->second));
+    }
+    resolve_shadows.erase(it);
 }
 
 void TextureCacheRuntime::BarrierFeedbackLoop() {
@@ -1195,7 +1212,8 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src,
 
         cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT,
                        0, READ_BARRIER, {}, middle_out_barrier);
-        cmdbuf.CopyBufferToImage(copy_buffer, dst_image, VK_IMAGE_LAYOUT_GENERAL, vk_out_copies);
+        cmdbuf.CopyBufferToImage(copy_buffer, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+                                 vk_out_copies);
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
                        0, {}, {}, post_barriers);
     });
@@ -1357,133 +1375,83 @@ void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, Im
     }
 
     switch (dst_view.format) {
-    case PixelFormat::D24_UNORM_S8_UINT:
-        if (src_view.format == PixelFormat::A8B8G8R8_UNORM
-        || src_view.format == PixelFormat::B8G8R8A8_UNORM
-        || src_view.format == PixelFormat::A8B8G8R8_SRGB
-        || src_view.format == PixelFormat::B8G8R8A8_SRGB) {
-            return blit_image_helper.ConvertABGR8ToD24S8(dst, src_view);
+    case PixelFormat::R16_UNORM:
+        if (src_view.format == PixelFormat::D16_UNORM) {
+            return blit_image_helper.ConvertD16ToR16(dst, src_view);
+        }
+        break;
+    case PixelFormat::A8B8G8R8_SRGB:
+    case PixelFormat::B8G8R8A8_SRGB:
+    case PixelFormat::B8G8R8A8_UNORM:
+        if (src_view.format == PixelFormat::D32_FLOAT) {
+            return blit_image_helper.ConvertD32FToABGR8(dst, src_view);
         }
         break;
     case PixelFormat::A8B8G8R8_UNORM:
-    case PixelFormat::A8B8G8R8_SNORM:
-    case PixelFormat::A8B8G8R8_SINT:
-    case PixelFormat::A8B8G8R8_UINT:
-    case PixelFormat::R5G6B5_UNORM:
-    case PixelFormat::B5G6R5_UNORM:
-    case PixelFormat::A1R5G5B5_UNORM:
-    case PixelFormat::A2B10G10R10_UNORM:
-    case PixelFormat::A2B10G10R10_UINT:
-    case PixelFormat::A2R10G10B10_UNORM:
-    case PixelFormat::A1B5G5R5_UNORM:
-    case PixelFormat::A5B5G5R1_UNORM:
-    case PixelFormat::R8_UNORM:
-    case PixelFormat::R8_SNORM:
-    case PixelFormat::R8_SINT:
-    case PixelFormat::R8_UINT:
-    case PixelFormat::R16G16B16A16_FLOAT:
-    case PixelFormat::R16G16B16A16_UNORM:
-    case PixelFormat::R16G16B16A16_SNORM:
-    case PixelFormat::R16G16B16A16_SINT:
-    case PixelFormat::R16G16B16A16_UINT:
-    case PixelFormat::B10G11R11_FLOAT:
-    case PixelFormat::R32G32B32A32_UINT:
-    case PixelFormat::BC1_RGBA_UNORM:
-    case PixelFormat::BC2_UNORM:
-    case PixelFormat::BC3_UNORM:
-    case PixelFormat::BC4_UNORM:
-    case PixelFormat::BC4_SNORM:
-    case PixelFormat::BC5_UNORM:
-    case PixelFormat::BC5_SNORM:
-    case PixelFormat::BC7_UNORM:
-    case PixelFormat::BC6H_UFLOAT:
-    case PixelFormat::BC6H_SFLOAT:
-    case PixelFormat::ASTC_2D_4X4_UNORM:
-    case PixelFormat::B8G8R8A8_UNORM:
-    case PixelFormat::R32G32B32A32_FLOAT:
-    case PixelFormat::R32G32B32A32_SINT:
-    case PixelFormat::R32G32_FLOAT:
-    case PixelFormat::R32G32_SINT:
-    case PixelFormat::R32_FLOAT:
-        if (src_view.format == PixelFormat::D32_FLOAT &&
-            (dst_view.format == PixelFormat::B5G6R5_UNORM ||
-             Settings::values.fix_bloom_effects.GetValue())) {
-            const Region2D region{
-                .start = {0, 0},
-                .end = {static_cast<s32>(dst->RenderArea().width),
-                        static_cast<s32>(dst->RenderArea().height)},
-            };
-            return blit_image_helper.BlitColor(dst, src_view, region, region,
-                                            Tegra::Engines::Fermi2D::Filter::Point,
-                                            Tegra::Engines::Fermi2D::Operation::SrcCopy);
+        if (src_view.format == PixelFormat::S8_UINT_D24_UNORM) {
+            return blit_image_helper.ConvertD24S8ToABGR8(dst, src_view);
+        }
+        if (src_view.format == PixelFormat::D24_UNORM_S8_UINT) {
+            return blit_image_helper.ConvertS8D24ToABGR8(dst, src_view);
+        }
+        if (src_view.format == PixelFormat::D32_FLOAT) {
+            return blit_image_helper.ConvertD32FToABGR8(dst, src_view);
+        }
+        break;
+    case PixelFormat::R32_FLOAT:
+        if (src_view.format == PixelFormat::D32_FLOAT) {
+            return blit_image_helper.ConvertD32ToR32(dst, src_view);
         }
         break;
-    case PixelFormat::R16_FLOAT:
-    case PixelFormat::R16_UNORM:
-    case PixelFormat::R16_SNORM:
-    case PixelFormat::R16_UINT:
-    case PixelFormat::R16_SINT:
-    case PixelFormat::R16G16_UNORM:
-    case PixelFormat::R16G16_FLOAT:
-    case PixelFormat::R16G16_UINT:
-    case PixelFormat::R16G16_SINT:
-    case PixelFormat::R16G16_SNORM:
-    case PixelFormat::R32G32B32_FLOAT:
-    case PixelFormat::A8B8G8R8_SRGB:
-    case PixelFormat::R8G8_UNORM:
-    case PixelFormat::R8G8_SNORM:
-    case PixelFormat::R8G8_SINT:
-    case PixelFormat::R8G8_UINT:
-    case PixelFormat::R32G32_UINT:
-    case PixelFormat::R16G16B16X16_FLOAT:
-    case PixelFormat::R32_UINT:
-    case PixelFormat::R32_SINT:
-    case PixelFormat::ASTC_2D_8X8_UNORM:
-    case PixelFormat::ASTC_2D_8X5_UNORM:
-    case PixelFormat::ASTC_2D_5X4_UNORM:
-    case PixelFormat::B8G8R8A8_SRGB:
-    case PixelFormat::BC1_RGBA_SRGB:
-    case PixelFormat::BC2_SRGB:
-    case PixelFormat::BC3_SRGB:
-    case PixelFormat::BC7_SRGB:
-    case PixelFormat::A4B4G4R4_UNORM:
-    case PixelFormat::G4R4_UNORM:
-    case PixelFormat::ASTC_2D_4X4_SRGB:
-    case PixelFormat::ASTC_2D_8X8_SRGB:
-    case PixelFormat::ASTC_2D_8X5_SRGB:
-    case PixelFormat::ASTC_2D_5X4_SRGB:
-    case PixelFormat::ASTC_2D_5X5_UNORM:
-    case PixelFormat::ASTC_2D_5X5_SRGB:
-    case PixelFormat::ASTC_2D_10X8_UNORM:
-    case PixelFormat::ASTC_2D_10X8_SRGB:
-    case PixelFormat::ASTC_2D_6X6_UNORM:
-    case PixelFormat::ASTC_2D_6X6_SRGB:
-    case PixelFormat::ASTC_2D_10X6_UNORM:
-    case PixelFormat::ASTC_2D_10X6_SRGB:
-    case PixelFormat::ASTC_2D_10X5_UNORM:
-    case PixelFormat::ASTC_2D_10X5_SRGB:
-    case PixelFormat::ASTC_2D_10X10_UNORM:
-    case PixelFormat::ASTC_2D_10X10_SRGB:
-    case PixelFormat::ASTC_2D_12X10_UNORM:
-    case PixelFormat::ASTC_2D_12X10_SRGB:
-    case PixelFormat::ASTC_2D_12X12_UNORM:
-    case PixelFormat::ASTC_2D_12X12_SRGB:
-    case PixelFormat::ASTC_2D_8X6_UNORM:
-    case PixelFormat::ASTC_2D_8X6_SRGB:
-    case PixelFormat::ASTC_2D_6X5_UNORM:
-    case PixelFormat::ASTC_2D_6X5_SRGB:
-    case PixelFormat::E5B9G9R9_FLOAT:
-    case PixelFormat::D32_FLOAT:
     case PixelFormat::D16_UNORM:
-    case PixelFormat::X8_D24_UNORM:
-    case PixelFormat::S8_UINT:
+        if (src_view.format == PixelFormat::R16_UNORM) {
+            return blit_image_helper.ConvertR16ToD16(dst, src_view);
+        }
+        break;
     case PixelFormat::S8_UINT_D24_UNORM:
-    case PixelFormat::D32_FLOAT_S8_UINT:
-    case PixelFormat::Invalid:
+        if (src_view.format == PixelFormat::A8B8G8R8_UNORM ||
+            src_view.format == PixelFormat::B8G8R8A8_UNORM) {
+            return blit_image_helper.ConvertABGR8ToD24S8(dst, src_view);
+        }
+        break;
+    case PixelFormat::D32_FLOAT:
+        if (src_view.format == PixelFormat::A8B8G8R8_UNORM ||
+            src_view.format == PixelFormat::B8G8R8A8_UNORM ||
+            src_view.format == PixelFormat::A8B8G8R8_SRGB ||
+            src_view.format == PixelFormat::B8G8R8A8_SRGB) {
+            return blit_image_helper.ConvertABGR8ToD32F(dst, src_view);
+        }
+        if (src_view.format == PixelFormat::R32_FLOAT) {
+            return blit_image_helper.ConvertR32ToD32(dst, src_view);
+        }
+        break;
+    case PixelFormat::D24_UNORM_S8_UINT:
+        if (src_view.format == PixelFormat::A8B8G8R8_UNORM ||
+            src_view.format == PixelFormat::B8G8R8A8_UNORM ||
+            src_view.format == PixelFormat::A8B8G8R8_SRGB ||
+            src_view.format == PixelFormat::B8G8R8A8_SRGB) {
+            return blit_image_helper.ConvertABGR8ToD24S8(dst, src_view);
+        }
+        break;
     default:
-        LOG_DEBUG(Render_Vulkan, "Unimplemented texture conversion from {} to {} format type", src_view.format, dst_view.format);
         break;
     }
+
+    if (src_view.format == PixelFormat::D32_FLOAT &&
+        VideoCore::Surface::GetFormatType(dst_view.format) == SurfaceType::ColorTexture &&
+        (dst_view.format == PixelFormat::B5G6R5_UNORM ||
+         Settings::values.fix_bloom_effects.GetValue())) {
+        const Region2D region{
+            .start = {0, 0},
+            .end = {static_cast<s32>(dst->RenderArea().width),
+                    static_cast<s32>(dst->RenderArea().height)},
+        };
+        return blit_image_helper.BlitColor(dst, src_view, region, region,
+                                        Tegra::Engines::Fermi2D::Filter::Point,
+                                        Tegra::Engines::Fermi2D::Operation::SrcCopy);
+    }
+
+    LOG_DEBUG(Render_Vulkan, "Unimplemented texture conversion from {} to {} format type", src_view.format, dst_view.format);
 }
 
 VkFormat TextureCacheRuntime::GetSupportedFormat(VkFormat requested_format,
@@ -1640,31 +1608,41 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
                                         std::span<const VideoCommon::ImageCopy> copies) {
     const bool msaa_to_non_msaa = src.info.num_samples > 1 && dst.info.num_samples == 1;
     const u32 num_samples = msaa_to_non_msaa ? src.info.num_samples : dst.info.num_samples;
-    if (dst.AspectMask() != VK_IMAGE_ASPECT_COLOR_BIT ||
-        VideoCore::Surface::IsPixelFormatInteger(dst.info.format)) {
-        UNIMPLEMENTED_MSG("Copying images with different samples is not supported.");
-        return;
-    }
     if (ENABLE_MSAA_RESOLVE_CONSUME && msaa_to_non_msaa && copies.size() == 1 &&
         src.info.format == dst.info.format) {
         const VideoCommon::ImageCopy& copy = copies.front();
         const ResolveShadow* const shadow = GetValidResolveShadow(src.Handle());
-        if (shadow != nullptr && copy.src_offset.x == 0 && copy.src_offset.y == 0 &&
+        if (shadow != nullptr && shadow->aspect_mask == dst.AspectMask() &&
+            copy.src_offset.x == 0 && copy.src_offset.y == 0 &&
             copy.src_subresource.base_level == 0 &&
             static_cast<u32>(copy.extent.width) <= shadow->extent.width &&
-            static_cast<u32>(copy.extent.height) <= shadow->extent.height) {
+            static_cast<u32>(copy.extent.height) <= shadow->extent.height &&
+            static_cast<u32>(copy.src_subresource.base_layer + copy.src_subresource.num_layers) <=
+                shadow->layers) {
+            const VkImageAspectFlags aspect_mask = shadow->aspect_mask;
+            VkPipelineStageFlags attachment_stage = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
+            VkAccessFlags attachment_write = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+            VkAccessFlags attachment_read_write =
+                VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+            if ((aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) {
+                attachment_stage = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                   VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
+                attachment_write = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+                attachment_read_write = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
+                                        VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+            }
             const VkImage shadow_image = *shadow->image;
             const VkImage dst_image = dst.Handle();
             const VkImageCopy region{
                 .srcSubresource{
-                    .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                    .aspectMask = aspect_mask,
                     .mipLevel = 0,
                     .baseArrayLayer = static_cast<u32>(copy.src_subresource.base_layer),
                     .layerCount = static_cast<u32>(copy.src_subresource.num_layers),
                 },
                 .srcOffset = {0, 0, 0},
                 .dstSubresource{
-                    .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                    .aspectMask = aspect_mask,
                     .mipLevel = static_cast<u32>(copy.dst_subresource.base_level),
                     .baseArrayLayer = static_cast<u32>(copy.dst_subresource.base_layer),
                     .layerCount = static_cast<u32>(copy.dst_subresource.num_layers),
@@ -1673,26 +1651,27 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
                 .extent = {copy.extent.width, copy.extent.height, 1},
             };
             scheduler.RequestOutsideRenderPassOperationContext();
-            scheduler.Record([shadow_image, dst_image, region](vk::CommandBuffer cmdbuf) {
+            scheduler.Record([shadow_image, dst_image, region, aspect_mask, attachment_stage,
+                              attachment_write,
+                              attachment_read_write](vk::CommandBuffer cmdbuf) {
                 const std::array pre_barriers{
                     VkImageMemoryBarrier{
                         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                         .pNext = nullptr,
-                        .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                        .srcAccessMask = attachment_write,
                         .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
                         .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                         .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
                         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .image = shadow_image,
-                        .subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
+                        .subresourceRange{aspect_mask, 0, VK_REMAINING_MIP_LEVELS, 0,
                                           VK_REMAINING_ARRAY_LAYERS},
                     },
                     VkImageMemoryBarrier{
                         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                         .pNext = nullptr,
-                        .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT |
-                                         VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                        .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | attachment_write |
                                          VK_ACCESS_TRANSFER_WRITE_BIT,
                         .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
                         .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
@@ -1700,7 +1679,7 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
                         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .image = dst_image,
-                        .subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
+                        .subresourceRange{aspect_mask, 0, VK_REMAINING_MIP_LEVELS, 0,
                                           VK_REMAINING_ARRAY_LAYERS},
                     },
                 };
@@ -1715,28 +1694,25 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
                         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .image = shadow_image,
-                        .subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
+                        .subresourceRange{aspect_mask, 0, VK_REMAINING_MIP_LEVELS, 0,
                                           VK_REMAINING_ARRAY_LAYERS},
                     },
                     VkImageMemoryBarrier{
                         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                         .pNext = nullptr,
                         .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-                        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT |
-                                         VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
-                                         VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | attachment_read_write |
                                          VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
                         .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
                         .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .image = dst_image,
-                        .subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
+                        .subresourceRange{aspect_mask, 0, VK_REMAINING_MIP_LEVELS, 0,
                                           VK_REMAINING_ARRAY_LAYERS},
                     },
                 };
-                cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                                           VK_PIPELINE_STAGE_TRANSFER_BIT,
+                cmdbuf.PipelineBarrier(attachment_stage | VK_PIPELINE_STAGE_TRANSFER_BIT,
                                        VK_PIPELINE_STAGE_TRANSFER_BIT, 0, nullptr, nullptr,
                                        pre_barriers);
                 cmdbuf.CopyImage(shadow_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image,
@@ -1748,6 +1724,11 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
             return;
         }
     }
+    if (dst.AspectMask() != VK_IMAGE_ASPECT_COLOR_BIT ||
+        VideoCore::Surface::IsPixelFormatInteger(dst.info.format)) {
+        UNIMPLEMENTED_MSG("Copying images with different samples is not supported.");
+        return;
+    }
     blit_image_helper.CopyMSAA(render_pass_cache, dst.Handle(), dst.info.format, src.Handle(),
                                src.info.format, num_samples, copies, msaa_to_non_msaa);
 }
@@ -1772,6 +1753,9 @@ void TextureCacheRuntime::TickFrame() {
     std::erase_if(pending_msaa_images, [this](const auto& pending) {
         return scheduler.IsFree(pending.first);
     });
+    std::erase_if(pending_resolve_shadows, [this](const auto& pending) {
+        return scheduler.IsFree(pending.first);
+    });
 }
 
 Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_,
@@ -1796,6 +1780,13 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
         }
         flags |= VideoCommon::ImageFlagBits::Converted;
         flags |= VideoCommon::ImageFlagBits::CostlyLoad;
+    } else if (runtime->bl2d_unswizzle_pass && BlockLinearUnswizzle2DPass::IsSupported(info)) {
+        flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
+        flags |= VideoCommon::ImageFlagBits::CostlyLoad;
+    } else if (runtime->bl3db_unswizzle_pass &&
+               BlockLinearUnswizzle3DBufferPass::IsSupported(runtime->device, info)) {
+        flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
+        flags |= VideoCommon::ImageFlagBits::CostlyLoad;
     }
     if (IsPixelFormatBCn(info.format) && !runtime->device.IsOptimalBcnSupported()) {
         flags |= VideoCommon::ImageFlagBits::Converted;
@@ -1882,11 +1873,21 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
         ScaleDown(true);
     }
 
+    const bool msaa_upload_is_depth = (aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) != 0;
     const bool wants_msaa_upload = info.num_samples > 1
-        && (aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0
+        && ((aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0 || msaa_upload_is_depth)
         && !VideoCore::Surface::IsPixelFormatInteger(info.format);
 
     if (wants_msaa_upload) {
+        const bool msaa_upload_copies_stencil =
+            msaa_upload_is_depth && (aspect_mask & VK_IMAGE_ASPECT_STENCIL_BIT) != 0 &&
+            runtime->device.IsExtShaderStencilExportSupported();
+        const VkImageAspectFlags upload_aspect_mask =
+            msaa_upload_is_depth
+                ? (msaa_upload_copies_stencil
+                       ? VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT
+                       : VK_IMAGE_ASPECT_DEPTH_BIT)
+                : aspect_mask;
         ImageInfo temp_info = info;
         temp_info.num_samples = 1;
 
@@ -1898,10 +1899,10 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
         vk::Image temp_image = runtime->memory_allocator.CreateImage(image_ci);
 
         scheduler->RequestOutsideRenderPassOperationContext();
-        auto vk_copies = TransformBufferImageCopies(copies, offset, aspect_mask);
+        auto vk_copies = TransformBufferImageCopies(copies, offset, upload_aspect_mask);
         const VkBuffer src_buffer = buffer;
         const VkImage temp_vk_image = *temp_image;
-        const VkImageAspectFlags vk_aspect_mask = aspect_mask;
+        const VkImageAspectFlags vk_aspect_mask = upload_aspect_mask;
 
         scheduler->Record([src_buffer, temp_vk_image, vk_aspect_mask,
                            vk_copies](vk::CommandBuffer cmdbuf) {
@@ -1923,9 +1924,16 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
             image_copies.push_back(image_copy);
         }
 
-        runtime->blit_image_helper.CopyMSAA(runtime->render_pass_cache, Handle(), info.format,
-                                            temp_vk_image, info.format, info.num_samples,
-                                            image_copies, false);
+        if (msaa_upload_is_depth) {
+            runtime->blit_image_helper.CopyMSAADepth(runtime->render_pass_cache, Handle(),
+                                                     info.format, temp_vk_image, info.format,
+                                                     info.num_samples, image_copies,
+                                                     msaa_upload_copies_stencil);
+        } else {
+            runtime->blit_image_helper.CopyMSAA(runtime->render_pass_cache, Handle(), info.format,
+                                                temp_vk_image, info.format, info.num_samples,
+                                                image_copies, false);
+        }
         initialized = true;
         runtime->pending_msaa_images.emplace_back(scheduler->CurrentTick(), std::move(temp_image));
 
@@ -2690,7 +2698,7 @@ Framebuffer::~Framebuffer() = default;
 void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
                                     std::span<ImageView*, NUM_RT> color_buffers,
                                     ImageView* depth_buffer, bool is_rescaled_) {
-    boost::container::small_vector<VkImageView, NUM_RT + 1> attachments;
+    boost::container::small_vector<VkImageView, NUM_RT * 2 + 2> attachments;
     RenderPassKey renderpass_key{};
     s32 num_layers = 1;
 
@@ -2719,6 +2727,8 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
         ++num_images;
     }
     const size_t num_colors = attachments.size();
+    VkImage depth_image = VK_NULL_HANDLE;
+    VkImageAspectFlags depth_aspect_mask = 0;
     if (depth_buffer) {
         width = (std::min)(width, is_rescaled ? resolution.ScaleUp(depth_buffer->size.width)
                                             : depth_buffer->size.width);
@@ -2734,6 +2744,8 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
         ++num_images;
         has_depth = (subresource_range.aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT) != 0;
         has_stencil = (subresource_range.aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT) != 0;
+        depth_image = depth_buffer->ImageHandle();
+        depth_aspect_mask = subresource_range.aspectMask;
     } else {
         renderpass_key.depth_format = PixelFormat::Invalid;
     }
@@ -2742,6 +2754,12 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
         samples != VK_SAMPLE_COUNT_1_BIT && num_colors > 0 && runtime.device.IsTiler();
     renderpass_key.resolve_color = do_resolve_color;
 
+    const bool do_resolve_depth_stencil =
+        samples != VK_SAMPLE_COUNT_1_BIT && depth_image != VK_NULL_HANDLE &&
+        runtime.device.IsTiler() &&
+        SupportsDepthStencilResolve(runtime.device, renderpass_key.depth_format);
+    renderpass_key.resolve_depth_stencil = do_resolve_depth_stencil;
+
     discard_msaa_color =
         ENABLE_MSAA_RESOLVE_CONSUME && ENABLE_MSAA_COLOR_DISCARD && do_resolve_color;
 
@@ -2762,8 +2780,8 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
                 MaxwellToVK::SurfaceFormat(runtime.device, FormatType::Optimal, true, format).format;
             if (ENABLE_MSAA_RESOLVE_CONSUME) {
                 const VkImage msaa_image = images[rt_map[index]];
-                attachments.push_back(runtime.GetOrCreateResolveShadow(msaa_image, vk_format,
-                                                                       render_area, layers));
+                attachments.push_back(runtime.GetOrCreateResolveShadow(
+                    msaa_image, vk_format, render_area, layers, VK_IMAGE_ASPECT_COLOR_BIT));
                 continue;
             }
             VkImageCreateInfo resolve_ci{
@@ -2808,6 +2826,16 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
         }
     }
 
+    if (do_resolve_depth_stencil) {
+        const u32 layers = static_cast<u32>((std::max)(num_layers, 1));
+        const VkFormat vk_format =
+            MaxwellToVK::SurfaceFormat(runtime.device, FormatType::Optimal, true,
+                                       renderpass_key.depth_format)
+                .format;
+        attachments.push_back(runtime.GetOrCreateResolveShadow(depth_image, vk_format, render_area,
+                                                               layers, depth_aspect_mask));
+    }
+
     num_color_buffers = static_cast<u32>(num_colors);
     framebuffer = runtime.device.GetLogical().CreateFramebuffer({
         .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
@@ -2827,11 +2855,25 @@ VkRenderPass Framebuffer::RenderPassVariant(u32 color_clear_mask, bool depth_ste
     if (color_clear_mask == 0 && !depth_stencil_clear && color_discard_mask == 0) {
         return renderpass;
     }
+    static_assert(NUM_RT <= 8);
+    const u32 variant_key = color_clear_mask | (color_discard_mask << 8) |
+                            (static_cast<u32>(depth_stencil_clear) << 16);
+    for (u32 index = 0; index < num_memoized_variants; ++index) {
+        if (variant_keys[index] == variant_key) {
+            return variant_render_passes[index];
+        }
+    }
     RenderPassKey key = render_pass_key;
     key.color_clear_mask = color_clear_mask;
     key.depth_stencil_clear = depth_stencil_clear;
     key.color_discard_mask = color_discard_mask;
-    return render_pass_cache->Get(key);
+    const VkRenderPass variant = render_pass_cache->Get(key);
+    if (num_memoized_variants < variant_keys.size()) {
+        variant_keys[num_memoized_variants] = variant_key;
+        variant_render_passes[num_memoized_variants] = variant;
+        ++num_memoized_variants;
+    }
+    return variant;
 }
 
 void TextureCacheRuntime::AccelerateImageUpload(
@@ -2843,6 +2885,15 @@ void TextureCacheRuntime::AccelerateImageUpload(
         return astc_decoder_pass->Assemble(image, map, swizzles);
     }
 
+    if (bl2d_unswizzle_pass && BlockLinearUnswizzle2DPass::IsSupported(image.info)) {
+        return bl2d_unswizzle_pass->Unswizzle(image, map, swizzles);
+    }
+
+    if (bl3db_unswizzle_pass &&
+        BlockLinearUnswizzle3DBufferPass::IsSupported(device, image.info)) {
+        return bl3db_unswizzle_pass->Unswizzle(image, map, swizzles);
+    }
+
     if (!Settings::values.gpu_unswizzle_enabled.GetValue() || !bl3d_unswizzle_pass) {
         if (IsPixelFormatBCn(image.info.format) && image.info.type == ImageType::e3D) {
             ASSERT(false && "GPU unswizzle is disabled for BCn 3D texture");

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -117,11 +117,13 @@ public:
         VkFormat format = VK_FORMAT_UNDEFINED;
         VkExtent2D extent{};
         u32 layers = 0;
+        VkImageAspectFlags aspect_mask = VK_IMAGE_ASPECT_COLOR_BIT;
         bool up_to_date = false;
     };
 
     [[nodiscard]] VkImageView GetOrCreateResolveShadow(VkImage msaa_image, VkFormat format,
-                                                       VkExtent2D extent, u32 layers);
+                                                       VkExtent2D extent, u32 layers,
+                                                       VkImageAspectFlags aspect_mask);
 
     [[nodiscard]] const ResolveShadow* GetValidResolveShadow(VkImage msaa_image) const;
 
@@ -149,6 +151,8 @@ public:
     std::optional<ASTCDecoderPass> astc_decoder_pass;
 
     std::optional<BlockLinearUnswizzle3DPass> bl3d_unswizzle_pass;
+    std::optional<BlockLinearUnswizzle2DPass> bl2d_unswizzle_pass;
+    std::optional<BlockLinearUnswizzle3DBufferPass> bl3db_unswizzle_pass;
     const Settings::ResolutionScalingInfo& resolution;
     std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;
 
@@ -156,6 +160,7 @@ public:
     std::array<vk::Buffer, indexing_slots> buffers{};
     std::vector<std::pair<u64, vk::Image>> pending_msaa_images;
     ankerl::unordered_dense::map<VkImage, ResolveShadow> resolve_shadows;
+    std::vector<std::pair<u64, ResolveShadow>> pending_resolve_shadows;
 };
 
 class Framebuffer {
@@ -246,6 +251,8 @@ public:
     }
 
 private:
+    static constexpr size_t NUM_MEMOIZED_RENDER_PASS_VARIANTS = 8;
+
     vk::Framebuffer framebuffer;
     VkRenderPass renderpass{};
     VkExtent2D render_area{};
@@ -263,6 +270,9 @@ private:
     RenderPassKey render_pass_key{};
     RenderPassCache* render_pass_cache{nullptr};
     bool discard_msaa_color{};
+    mutable std::array<u32, NUM_MEMOIZED_RENDER_PASS_VARIANTS> variant_keys{};
+    mutable std::array<VkRenderPass, NUM_MEMOIZED_RENDER_PASS_VARIANTS> variant_render_passes{};
+    mutable u32 num_memoized_variants{};
 };
 
 class Image : public VideoCommon::ImageBase {

src/video_core/textures/decoders.cpp

@@ -1,9 +1,10 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
 #include <array>
 #include <cmath>
 #include <cstring>
@@ -30,12 +31,62 @@ constexpr u32 pdep(u32 value) {
     return result;
 }
 
+constexpr u32 SWIZZLE_RUN_BYTES = 16;
+constexpr u32 SWIZZLE_RUN_SHIFT = 4;
+constexpr u32 SWIZZLE_RUN_MASK = SWIZZLE_RUN_BYTES - 1;
+constexpr u32 SWIZZLE_RUN_INDEX_MASK = GOB_SIZE_X / SWIZZLE_RUN_BYTES - 1;
+
+static_assert((SWIZZLE_X_BITS & SWIZZLE_RUN_MASK) == SWIZZLE_RUN_MASK);
+
+constexpr std::array<u32, GOB_SIZE_X / SWIZZLE_RUN_BYTES> SWIZZLE_X_RUN_TABLE = [] {
+    std::array<u32, GOB_SIZE_X / SWIZZLE_RUN_BYTES> table{};
+    for (u32 index = 0; index < static_cast<u32>(table.size()); ++index) {
+        table[index] = pdep<SWIZZLE_X_BITS>(index << SWIZZLE_RUN_SHIFT);
+    }
+    return table;
+}();
+
 template <u32 mask, u32 incr_amount>
 void incrpdep(u32& value) {
     static constexpr u32 swizzled_incr = pdep<mask>(incr_amount);
     value = ((value | ~mask) + swizzled_incr) & mask;
 }
 
+template <bool TO_LINEAR>
+void SwizzleRow(std::span<u8> output, std::span<const u8> input, u32 offset_zy, u32 swizzled_y,
+                u32 x_shift, u32 x, u32 num_bytes, u32 linear) {
+    const auto copy = [&](u32 swizzled_x, u32 count) {
+        const u32 swizzled =
+            offset_zy + ((x >> GOB_SIZE_X_SHIFT) << x_shift) + (swizzled_x | swizzled_y);
+        u8* const dst = &output[TO_LINEAR ? swizzled : linear];
+        const u8* const src = &input[TO_LINEAR ? linear : swizzled];
+        std::memcpy(dst, src, count);
+        x += count;
+        linear += count;
+    };
+
+    u32 swizzled_run = SWIZZLE_X_RUN_TABLE[(x >> SWIZZLE_RUN_SHIFT) & SWIZZLE_RUN_INDEX_MASK];
+    u32 remaining = num_bytes;
+
+    const u32 head =
+        (std::min)(SWIZZLE_RUN_BYTES - (x & SWIZZLE_RUN_MASK), remaining) & SWIZZLE_RUN_MASK;
+    if (head != 0) {
+        copy(swizzled_run | (x & SWIZZLE_RUN_MASK), head);
+        remaining -= head;
+        incrpdep<SWIZZLE_X_BITS, SWIZZLE_RUN_BYTES>(swizzled_run);
+    }
+
+    while (remaining >= SWIZZLE_RUN_BYTES) {
+        copy(swizzled_run, SWIZZLE_RUN_BYTES);
+        remaining -= SWIZZLE_RUN_BYTES;
+        incrpdep<SWIZZLE_X_BITS, SWIZZLE_RUN_BYTES>(swizzled_run);
+    }
+
+    if (remaining != 0) {
+        copy(swizzled_run, remaining);
+    }
+}
+
 template <bool TO_LINEAR, u32 BYTES_PER_PIXEL>
 void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, u32 depth,
                  u32 block_height, u32 block_depth, u32 stride) {
@@ -70,23 +121,9 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32
             const u32 offset_y = (block_y >> block_height) * block_size +
                                  ((block_y & block_height_mask) << GOB_SIZE_SHIFT);
 
-            u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL);
-            for (u32 column = 0; column < width;
-                 ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) {
-                const u32 x = (column + origin_x) * BYTES_PER_PIXEL;
-                const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift;
-
-                const u32 base_swizzled_offset = offset_z + offset_y + offset_x;
-                const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y);
-
-                const u32 unswizzled_offset =
-                    slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
-
-                u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset];
-                const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset];
-
-                std::memcpy(dst, src, BYTES_PER_PIXEL);
-            }
+            SwizzleRow<TO_LINEAR>(output, input, offset_z + offset_y, swizzled_y, x_shift,
+                                  origin_x * BYTES_PER_PIXEL, width * BYTES_PER_PIXEL,
+                                  slice * pitch * height + line * pitch);
         }
     }
 }
@@ -129,23 +166,9 @@ void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 wid
             const u32 offset_y = (block_y >> block_height) * block_size +
                                  ((block_y & block_height_mask) << GOB_SIZE_SHIFT);
 
-            u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL);
-            for (u32 column = 0; column < extent_x;
-                 ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) {
-                const u32 x = (column + origin_x) * BYTES_PER_PIXEL;
-                const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift;
-
-                const u32 base_swizzled_offset = offset_z + offset_y + offset_x;
-                const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y);
-
-                const u32 unswizzled_offset =
-                    slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
-
-                u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset];
-                const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset];
-
-                std::memcpy(dst, src, BYTES_PER_PIXEL);
-            }
+            SwizzleRow<TO_LINEAR>(output, input, offset_z + offset_y, swizzled_y, x_shift,
+                                  origin_x * BYTES_PER_PIXEL, extent_x * BYTES_PER_PIXEL,
+                                  slice * pitch * height + line * pitch);
         }
         unprocessed_lines -= lines_in_y;
         if (unprocessed_lines == 0) {

src/video_core/vulkan_common/vulkan_device.cpp

@@ -7,6 +7,8 @@
 #include <algorithm>
 #include <bitset>
 #include <chrono>
+#include <filesystem>
+#include <fstream>
 #include <optional>
 #include <thread>
 #include <ankerl/unordered_dense.h>
@@ -16,6 +18,8 @@
 #include <fmt/format.h>
 
 #include "common/assert.h"
+#include "common/fs/fs.h"
+#include "common/fs/path_util.h"
 #include "common/literals.h"
 #include <ranges>
 #include "common/settings.h"
@@ -393,6 +397,17 @@ std::vector<const char*> ExtensionListForVulkan(
     return output;
 }
 
+constexpr std::array<char, 8> STATIC_CACHE_MAGIC_NUMBER{'e', 'd', 'e', 'n', 's', 't', 'p', 'c'};
+constexpr u32 STATIC_CACHE_VERSION = 1;
+
+std::filesystem::path StaticPipelineCacheFilename() {
+    const auto shader_dir = Common::FS::GetEdenPath(Common::FS::EdenPath::ShaderDir);
+    if (!Common::FS::CreateDir(shader_dir)) {
+        return {};
+    }
+    return shader_dir / "vulkan_static_pipelines.bin";
+}
+
 } // Anonymous namespace
 
 void Device::RemoveExtension(bool& extension, const std::string& extension_name) {
@@ -780,15 +795,100 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
 
     vk::Check(vmaCreateAllocator(&allocator_info, &allocator));
 
+    owns_static_pipeline_cache = surface != VkSurfaceKHR{};
+    LoadStaticPipelineCache();
+
     // Initialize GPU logging if enabled
     InitializeGPULogging();
 }
 
 Device::~Device() {
+    SaveStaticPipelineCache();
     ShutdownGPULogging();
     vmaDestroyAllocator(allocator);
 }
 
+void Device::LoadStaticPipelineCache() {
+    const auto create = [this](size_t size, const void* data) {
+        static_pipeline_cache = logical.CreatePipelineCache({
+            .sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .initialDataSize = size,
+            .pInitialData = data,
+        });
+    };
+    if (!owns_static_pipeline_cache) {
+        create(0, nullptr);
+        return;
+    }
+    const auto filename = StaticPipelineCacheFilename();
+    if (filename.empty()) {
+        create(0, nullptr);
+        return;
+    }
+    std::vector<char> data;
+    try {
+        std::ifstream file(filename, std::ios::binary | std::ios::ate);
+        if (!file.is_open()) {
+            create(0, nullptr);
+            return;
+        }
+        file.exceptions(std::ifstream::failbit | std::ifstream::badbit);
+        const size_t total = static_cast<size_t>(file.tellg());
+        file.seekg(0, std::ios::beg);
+        std::array<char, 8> magic{};
+        u32 version{};
+        if (total < magic.size() + sizeof(version)) {
+            create(0, nullptr);
+            return;
+        }
+        file.read(magic.data(), magic.size())
+            .read(reinterpret_cast<char*>(&version), sizeof(version));
+        if (magic != STATIC_CACHE_MAGIC_NUMBER || version != STATIC_CACHE_VERSION) {
+            create(0, nullptr);
+            return;
+        }
+        data.resize(total - magic.size() - sizeof(version));
+        file.read(data.data(), static_cast<std::streamsize>(data.size()));
+    } catch (const std::ios_base::failure& e) {
+        create(0, nullptr);
+        return;
+    }
+    create(data.size(), data.empty() ? nullptr : data.data());
+}
+
+void Device::SaveStaticPipelineCache() const {
+    if (!owns_static_pipeline_cache || !static_pipeline_cache) {
+        return;
+    }
+    const auto filename = StaticPipelineCacheFilename();
+    if (filename.empty()) {
+        return;
+    }
+    size_t size = 0;
+    std::vector<char> data;
+    static_pipeline_cache.Read(&size, nullptr);
+    if (size == 0) {
+        return;
+    }
+    data.resize(size);
+    static_pipeline_cache.Read(&size, data.data());
+    try {
+        std::ofstream file(filename, std::ios::binary | std::ios::trunc);
+        file.exceptions(std::ofstream::failbit);
+        if (!file.is_open()) {
+            return;
+        }
+        file.write(STATIC_CACHE_MAGIC_NUMBER.data(), STATIC_CACHE_MAGIC_NUMBER.size())
+            .write(reinterpret_cast<const char*>(&STATIC_CACHE_VERSION),
+                   sizeof(STATIC_CACHE_VERSION))
+            .write(data.data(), static_cast<std::streamsize>(size));
+    } catch (const std::ios_base::failure& e) {
+        Common::FS::RemoveFile(filename);
+    }
+}
+
 VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
                                     FormatType format_type) const {
     if (IsFormatSupported(wanted_format, wanted_usage, format_type)) {
@@ -975,6 +1075,12 @@ bool Device::GetSuitability(bool requires_swapchain) {
     FOR_EACH_VK_FEATURE_EXT(FEATURE_EXTENSION);
     FOR_EACH_VK_EXTENSION(EXTENSION);
 
+    extensions.depth_stencil_resolve =
+        extensions.depth_stencil_resolve &&
+        (instance_version >= VK_API_VERSION_1_2 || extensions.create_renderpass2);
+    RemoveExtensionIfUnsuitable(extensions.depth_stencil_resolve,
+                                VK_KHR_DEPTH_STENCIL_RESOLVE_EXTENSION_NAME);
+
     if (supported_extensions.contains(VK_KHR_ROBUSTNESS_2_EXTENSION_NAME)) {
         loaded_extensions.erase(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME);
         loaded_extensions.insert(VK_KHR_ROBUSTNESS_2_EXTENSION_NAME);
@@ -1122,6 +1228,11 @@ bool Device::GetSuitability(bool requires_swapchain) {
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR;
         SetNext(next, properties.push_descriptor);
     }
+    if (extensions.depth_stencil_resolve || instance_version >= VK_API_VERSION_1_2) {
+        properties.depth_stencil_resolve.sType =
+            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES;
+        SetNext(next, properties.depth_stencil_resolve);
+    }
     if (extensions.descriptor_buffer) {
         properties.descriptor_buffer.sType =
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_BUFFER_PROPERTIES_EXT;

src/video_core/vulkan_common/vulkan_device.h

@@ -90,6 +90,8 @@ VK_DEFINE_HANDLE(VmaAllocator)
     EXTENSION(EXT, SHADER_VIEWPORT_INDEX_LAYER, shader_viewport_index_layer)                       \
     EXTENSION(EXT, TOOLING_INFO, tooling_info)                                                     \
     EXTENSION(EXT, VERTEX_ATTRIBUTE_DIVISOR, vertex_attribute_divisor)                             \
+    EXTENSION(KHR, CREATE_RENDERPASS_2, create_renderpass2)                                        \
+    EXTENSION(KHR, DEPTH_STENCIL_RESOLVE, depth_stencil_resolve)                                   \
     EXTENSION(KHR, DRAW_INDIRECT_COUNT, draw_indirect_count)                                       \
     EXTENSION(KHR, DRIVER_PROPERTIES, driver_properties)                                           \
     EXTENSION(KHR, PUSH_DESCRIPTOR, push_descriptor)                                               \
@@ -268,6 +270,10 @@ public:
         return physical;
     }
 
+    VkPipelineCache StaticPipelineCache() const noexcept {
+        return *static_pipeline_cache;
+    }
+
     /// Returns the main graphics queue.
     vk::Queue GetGraphicsQueue() const {
         return graphics_queue;
@@ -607,6 +613,37 @@ FN_MAX_LIMIT_LIST
         return extensions.shader_stencil_export;
     }
 
+    /// Returns true if the device supports VK_KHR_create_renderpass2.
+    bool IsKhrCreateRenderPass2Supported() const {
+        return extensions.create_renderpass2 || instance_version >= VK_API_VERSION_1_2;
+    }
+
+    /// Returns true if the device supports VK_KHR_depth_stencil_resolve.
+    bool IsKhrDepthStencilResolveSupported() const {
+        return (extensions.depth_stencil_resolve || instance_version >= VK_API_VERSION_1_2) &&
+               IsKhrCreateRenderPass2Supported();
+    }
+
+    /// Returns the supported resolve modes for the depth aspect.
+    VkResolveModeFlags GetDepthResolveModes() const {
+        return properties.depth_stencil_resolve.supportedDepthResolveModes;
+    }
+
+    /// Returns the supported resolve modes for the stencil aspect.
+    VkResolveModeFlags GetStencilResolveModes() const {
+        return properties.depth_stencil_resolve.supportedStencilResolveModes;
+    }
+
+    /// Returns true if the depth and stencil aspects may resolve with different modes.
+    bool SupportsIndependentResolve() const {
+        return properties.depth_stencil_resolve.independentResolve == VK_TRUE;
+    }
+
+    /// Returns true if only one of the depth and stencil aspects may be resolved.
+    bool SupportsIndependentResolveNone() const {
+        return properties.depth_stencil_resolve.independentResolveNone == VK_TRUE;
+    }
+
     /// Returns true if depth/stencil operations can be performed efficiently.
     /// Either through shader export or hardware blits.
     bool CanPerformDepthStencilOperations() const {
@@ -957,6 +994,21 @@ FN_MAX_LIMIT_LIST
         return features2.features.multiViewport;
     }
 
+    VkPipelineStageFlags AttachmentConsumerStages() const {
+        VkPipelineStageFlags stages = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
+                                      VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT |
+                                      VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT |
+                                      VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
+                                      VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                      VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+                                      VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
+                                      VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+        if (features2.features.geometryShader) {
+            stages |= VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT;
+        }
+        return stages;
+    }
+
     /// Returns true if the device supports VK_KHR_maintenance1.
     bool IsKhrMaintenance1Supported() const {
         return extensions.maintenance1;
@@ -1088,6 +1140,9 @@ private:
     /// Returns true if the device natively supports blitting depth stencil images.
     bool TestDepthStencilBlits(VkFormat format) const;
 
+    void LoadStaticPipelineCache();
+    void SaveStaticPipelineCache() const;
+
 private:
     VkInstance instance;         ///< Vulkan instance.
     VmaAllocator allocator;      ///< VMA allocator.
@@ -1096,6 +1151,8 @@ private:
     vk::Device logical;          ///< Logical device.
     vk::Queue graphics_queue;    ///< Main graphics queue.
     vk::Queue present_queue;     ///< Main present queue.
+    vk::PipelineCache static_pipeline_cache;
+    bool owns_static_pipeline_cache{};
     u32 instance_version{};      ///< Vulkan instance version.
     u32 graphics_family{};       ///< Main graphics queue family index.
     u32 present_family{};        ///< Main present queue family index.
@@ -1142,6 +1199,7 @@ private:
         VkPhysicalDeviceSubgroupSizeControlProperties subgroup_size_control{};
         VkPhysicalDeviceTransformFeedbackPropertiesEXT transform_feedback{};
         VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{};
+        VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{};
 
         VkPhysicalDeviceProperties properties{};
     };

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -184,6 +184,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkCreatePipelineLayout);
     X(vkCreateQueryPool);
     X(vkCreateRenderPass);
+    X(vkCreateRenderPass2);
     X(vkCreateSampler);
     X(vkCreateSemaphore);
     X(vkCreateShaderModule);
@@ -270,6 +271,11 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     if (!dld.vkQueueSubmit2) {
         Proc(dld.vkQueueSubmit2, dld, "vkQueueSubmit2KHR", device);
     }
+
+    // Render pass creation v2 is core in Vulkan 1.2, otherwise requires VK_KHR_create_renderpass2
+    if (!dld.vkCreateRenderPass2) {
+        Proc(dld.vkCreateRenderPass2, dld, "vkCreateRenderPass2KHR", device);
+    }
 #undef X
 }
 
@@ -725,6 +731,12 @@ RenderPass Device::CreateRenderPass(const VkRenderPassCreateInfo& ci) const {
     return RenderPass(object, handle, *dld);
 }
 
+RenderPass Device::CreateRenderPass2(const VkRenderPassCreateInfo2& ci) const {
+    VkRenderPass object;
+    Check(dld->vkCreateRenderPass2(handle, &ci, nullptr, &object));
+    return RenderPass(object, handle, *dld);
+}
+
 DescriptorSetLayout Device::CreateDescriptorSetLayout(
     const VkDescriptorSetLayoutCreateInfo& ci) const {
     VkDescriptorSetLayout object;

src/video_core/vulkan_common/vulkan_wrapper.h

@@ -300,6 +300,7 @@ struct DeviceDispatch : InstanceDispatch {
     PFN_vkCreatePipelineLayout vkCreatePipelineLayout{};
     PFN_vkCreateQueryPool vkCreateQueryPool{};
     PFN_vkCreateRenderPass vkCreateRenderPass{};
+    PFN_vkCreateRenderPass2 vkCreateRenderPass2{};
     PFN_vkCreateSampler vkCreateSampler{};
     PFN_vkCreateSemaphore vkCreateSemaphore{};
     PFN_vkCreateShaderModule vkCreateShaderModule{};
@@ -1045,6 +1046,8 @@ public:
 
     [[nodiscard]] RenderPass CreateRenderPass(const VkRenderPassCreateInfo& ci) const;
 
+    [[nodiscard]] RenderPass CreateRenderPass2(const VkRenderPassCreateInfo2& ci) const;
+
     [[nodiscard]] DescriptorSetLayout CreateDescriptorSetLayout(
         const VkDescriptorSetLayoutCreateInfo& ci) const;
 
@@ -1526,7 +1529,7 @@ public:
     }
 
     void SetDepthBias(float constant_factor, float clamp, float slope_factor,
-                      VkDepthBiasRepresentationInfoEXT* extra) const noexcept {
+                      const VkDepthBiasRepresentationInfoEXT* extra) const noexcept {
         VkDepthBiasInfoEXT info{
             .sType = VK_STRUCTURE_TYPE_DEPTH_BIAS_INFO_EXT,
             .pNext = extra,

src/yuzu_cmd/emu_window/emu_window_sdl3.cpp

@@ -5,10 +5,6 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <SDL3/SDL.h>
-#include <SDL3/SDL_timer.h>
-#ifdef __EMSCRIPTEN__
-#include <emscripten.h>
-#endif
 
 #include "common/logging.h"
 #include "common/scm_rev.h"
@@ -30,17 +26,9 @@ EmuWindow_SDL3::EmuWindow_SDL3(InputCommon::InputSubsystem* input_subsystem_, Co
         LOG_CRITICAL(Frontend, "Failed to initialize SDL3: {}, Exiting...", SDL_GetError());
         exit(1);
     }
-    titlebar_timer = SDL_AddTimer(2000, [](void *userdata, SDL_TimerID, Uint32) -> Uint32 {
-        auto* this_ = (EmuWindow_SDL3*)userdata;
-        auto const results = this_->system.GetAndResetPerfStats();
-        auto const title = fmt::format("{} | {}-{} | FPS: {:.0f} ({:.0f}%)", Common::g_build_fullname, Common::g_scm_branch, Common::g_scm_desc, results.average_game_fps, results.emulation_speed * 100.0f);
-        SDL_SetWindowTitle(this_->render_window, title.c_str());
-        return 2000;
-    }, this);
 }
 
 EmuWindow_SDL3::~EmuWindow_SDL3() {
-    SDL_RemoveTimer(titlebar_timer);
     system.HIDCore().UnloadInputDevices();
     input_subsystem->Shutdown();
     SDL_Quit();
@@ -144,7 +132,8 @@ void EmuWindow_SDL3::Fullscreen() {
     switch (Settings::values.fullscreen_mode.GetValue()) {
     case Settings::FullscreenMode::Exclusive:
         // Set window size to render size before entering fullscreen in exclusive mode.
-        if (const SDL_DisplayMode* display_mode_ptr = SDL_GetDesktopDisplayMode(SDL_GetDisplayForWindow(render_window))) {
+        if (const SDL_DisplayMode* display_mode_ptr =
+                SDL_GetDesktopDisplayMode(SDL_GetDisplayForWindow(render_window))) {
             display_mode = *display_mode_ptr;
             SDL_SetWindowSize(render_window, display_mode.w, display_mode.h);
             SDL_SetWindowFullscreenMode(render_window, &display_mode);
@@ -176,60 +165,92 @@ void EmuWindow_SDL3::Fullscreen() {
     }
 }
 
-void EmuWindow_SDL3::OnEvent(SDL_Event& event) {
-    // Notice how we skip the "update title" aspect on most events
-    // this is because some WMs do NOT like changing titles while resizing
-    // so let's just... not do that, thanks :)
-    // Afterall we don't really expect the user to pay attention to the titlebar
-    // while they're moving a lot of shit around...
+void EmuWindow_SDL3::WaitEvent() {
+    // Called on main thread
+    SDL_Event event;
+
+    if (!SDL_WaitEvent(&event)) {
+        const char* error = SDL_GetError();
+        if (!error || strcmp(error, "") == 0) {
+            // https://github.com/libsdl-org/SDL/issues/5780
+            // Sometimes SDL will return without actually having hit an error condition;
+            // just ignore it in this case.
+            return;
+        }
+
+        LOG_CRITICAL(Frontend, "SDL_WaitEvent failed: {}", error);
+        exit(1);
+    }
+
     switch (event.type) {
     case SDL_EVENT_WINDOW_RESIZED:
     case SDL_EVENT_WINDOW_PIXEL_SIZE_CHANGED:
     case SDL_EVENT_WINDOW_MAXIMIZED:
     case SDL_EVENT_WINDOW_RESTORED:
-        return OnResize();
+        OnResize();
+        break;
     case SDL_EVENT_WINDOW_MINIMIZED:
         is_shown = false;
-        return OnResize();
+        OnResize();
+        break;
     case SDL_EVENT_WINDOW_EXPOSED:
         is_shown = true;
-        return OnResize();
+        OnResize();
+        break;
     case SDL_EVENT_WINDOW_CLOSE_REQUESTED:
         is_open = false;
-        return;
+        break;
     case SDL_EVENT_KEY_DOWN:
     case SDL_EVENT_KEY_UP:
-        return OnKeyEvent(int(event.key.scancode), event.key.down ? 1 : 0);
+        OnKeyEvent(static_cast<int>(event.key.scancode), event.key.down ? 1 : 0);
+        break;
     case SDL_EVENT_MOUSE_MOTION:
         // ignore if it came from touch
         if (event.button.which != SDL_TOUCH_MOUSEID)
             OnMouseMotion(event.motion.x, event.motion.y);
-        return;
+        break;
     case SDL_EVENT_MOUSE_BUTTON_DOWN:
     case SDL_EVENT_MOUSE_BUTTON_UP:
         // ignore if it came from touch
-        if (event.button.which != SDL_TOUCH_MOUSEID)
-            OnMouseButton(event.button.button, event.button.down ? 1 : 0, s32(event.button.x), s32(event.button.y));
-        return;
+        if (event.button.which != SDL_TOUCH_MOUSEID) {
+            OnMouseButton(event.button.button, event.button.down ? 1 : 0,
+                          static_cast<s32>(event.button.x), static_cast<s32>(event.button.y));
+        }
+        break;
     case SDL_EVENT_FINGER_DOWN:
-        return OnFingerDown(event.tfinger.x, event.tfinger.y, std::size_t(event.tfinger.touchID));
+        OnFingerDown(event.tfinger.x, event.tfinger.y,
+                     static_cast<std::size_t>(event.tfinger.touchID));
+        break;
     case SDL_EVENT_FINGER_MOTION:
-        return OnFingerMotion(event.tfinger.x, event.tfinger.y, std::size_t(event.tfinger.touchID));
+        OnFingerMotion(event.tfinger.x, event.tfinger.y,
+                       static_cast<std::size_t>(event.tfinger.touchID));
+        break;
     case SDL_EVENT_FINGER_UP:
-        return OnFingerUp();
+        OnFingerUp();
+        break;
     case SDL_EVENT_QUIT:
         is_open = false;
-        return;
+        break;
     default:
         break;
     }
+
+    const u32 current_time = SDL_GetTicks();
+    if (current_time > last_time + 2000) {
+        const auto results = system.GetAndResetPerfStats();
+        const auto title = fmt::format("{} | {}-{} | FPS: {:.0f} ({:.0f}%)",
+                                       Common::g_build_fullname,
+                                       Common::g_scm_branch,
+                                       Common::g_scm_desc,
+                                       results.average_game_fps,
+                                       results.emulation_speed * 100.0);
+        SDL_SetWindowTitle(render_window, title.c_str());
+        last_time = current_time;
+    }
 }
 
 // Credits to Samantas5855 and others for this function.
 void EmuWindow_SDL3::SetWindowIcon() {
-#if defined(__EMSCRIPTEN__) || defined(__wasi__)
-    // Icons do not work yet
-#else
     SDL_IOStream* const yuzu_icon_stream = SDL_IOFromConstMem((void*)yuzu_icon, yuzu_icon_size);
     if (yuzu_icon_stream == nullptr) {
         LOG_WARNING(Frontend, "Failed to create Eden icon stream.");
@@ -243,7 +264,6 @@ void EmuWindow_SDL3::SetWindowIcon() {
     // The icon is attached to the window pointer
     SDL_SetWindowIcon(render_window, window_icon);
     SDL_DestroySurface(window_icon);
-#endif
 }
 
 void EmuWindow_SDL3::OnMinimalClientAreaChangeRequest(std::pair<u32, u32> minimal_size) {

src/yuzu_cmd/emu_window/emu_window_sdl3.h

@@ -9,12 +9,11 @@
 #include <tuple>
 #include <utility>
 
-#include <SDL3/SDL.h>
-#include <SDL3/SDL_timer.h>
-
 #include "core/frontend/emu_window.h"
 #include "core/frontend/graphics_context.h"
 
+struct SDL_Window;
+
 namespace Core {
 class System;
 }
@@ -36,7 +35,7 @@ public:
     bool IsShown() const override;
 
     /// Wait for the next event on the main thread.
-    void OnEvent(SDL_Event& event);
+    void WaitEvent();
 
     // Sets the window icon from yuzu.bmp
     void SetWindowIcon();
@@ -81,9 +80,6 @@ protected:
     /// Called when a configuration change affects the minimal size of the window
     void OnMinimalClientAreaChangeRequest(std::pair<u32, u32> minimal_size) override;
 
-    /// Periodic changer of titlebar (independent of event loop)
-    SDL_TimerID titlebar_timer;
-
     /// Is the window still open?
     bool is_open = true;
 
@@ -93,6 +89,9 @@ protected:
     /// Internal SDL3 render window
     SDL_Window* render_window{};
 
+    /// Keeps track of how often to update the title bar during gameplay
+    u32 last_time = 0;
+
     /// Input subsystem to use with this window.
     InputCommon::InputSubsystem* input_subsystem;
 

src/yuzu_cmd/yuzu.cpp

@@ -8,12 +8,6 @@
 #include <memory>
 #include <regex>
 #include <string>
-#include "common/settings_enums.h"
-#ifdef __EMSCRIPTEN__
-#include <emscripten.h>
-#endif
-#define SDL_MAIN_USE_CALLBACKS 1
-#include <SDL3/SDL_main.h>
 
 #include <fmt/ostream.h>
 
@@ -45,7 +39,9 @@
 #ifdef _WIN32
 // windows.h needs to be included before shellapi.h
 #include <windows.h>
+
 #include <shellapi.h>
+
 #include "common/windows/timer_resolution.h"
 #endif
 
@@ -178,14 +174,8 @@ static void OnStatusMessageReceived(const Network::StatusMessageEntry& msg) {
         std::cout << std::endl << "* " << message << std::endl << std::endl;
 }
 
-struct SdlState {
-    Core::System system{};
-    std::unique_ptr<EmuWindow_SDL3> emu_window;
-};
-
-extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
-    SdlState* state = new SdlState();
-
+/// Application entry point
+int main(int argc, char** argv) {
 #ifdef _WIN32
     if (AttachConsole(ATTACH_PARENT_PROCESS)) {
         freopen("CONOUT$", "wb", stdout);
@@ -203,7 +193,7 @@ extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
     auto argv_w = CommandLineToArgvW(GetCommandLineW(), &argc_w);
     if (argv_w == nullptr) {
         LOG_CRITICAL(Frontend, "Failed to get command line arguments");
-        return SDL_APP_FAILURE;
+        return -1;
     }
 #endif
     std::string filepath;
@@ -257,7 +247,7 @@ extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
                 break;
             case 'h':
                 PrintHelp(argv[0]);
-                return SDL_APP_FAILURE;
+                return 0;
             case 'g':
                 filepath = std::string(optarg);
                 break;
@@ -274,7 +264,7 @@ extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
                 if (!std::regex_match(str_arg, re)) {
                     std::cout << "Wrong format for option --multiplayer\n";
                     PrintHelp(argv[0]);
-                    return SDL_APP_FAILURE;
+                    return -1;
                 }
 
                 std::smatch match;
@@ -289,11 +279,11 @@ extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
                 std::regex nickname_re("^[a-zA-Z0-9._\\- ]+$");
                 if (!std::regex_match(nickname, nickname_re)) {
                     LOG_ERROR(Frontend, "Nickname is not valid. Must be 4 to 20 alphanumeric characters");
-                    return SDL_APP_FAILURE;
+                    return -1;
                 }
                 if (address.empty()) {
                     LOG_ERROR(Frontend, "Address to room must not be empty");
-                    return SDL_APP_FAILURE;
+                    return -1;
                 }
                 break;
             }
@@ -306,7 +296,7 @@ extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
                 break;
             case 'v':
                 PrintVersion();
-                return SDL_APP_FAILURE;
+                return -1;
             case 'n':
                 force_null_render = true;
                 break;
@@ -368,73 +358,79 @@ extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
 
     if (filepath.empty()) {
         LOG_CRITICAL(Frontend, "Failed to load ROM: No ROM specified");
-        return SDL_APP_FAILURE;
+        return -1;
     }
 
-    state->system.Initialize();
+    Core::System system{};
+    system.Initialize();
 
     InputCommon::InputSubsystem input_subsystem{};
 
     // Apply the command line arguments
-    state->system.ApplySettings();
+    system.ApplySettings();
 
+    std::unique_ptr<EmuWindow_SDL3> emu_window;
     switch (Settings::values.renderer_backend.GetValue()) {
 #ifdef HAS_OPENGL
     case Settings::RendererBackend::OpenGL_GLSL:
     case Settings::RendererBackend::OpenGL_GLASM:
     case Settings::RendererBackend::OpenGL_SPIRV:
-        state->emu_window = std::make_unique<EmuWindow_SDL3_GL>(&input_subsystem, state->system, fullscreen);
+        emu_window = std::make_unique<EmuWindow_SDL3_GL>(&input_subsystem, system, fullscreen);
         break;
 #endif
     case Settings::RendererBackend::Vulkan:
-        state->emu_window = std::make_unique<EmuWindow_SDL3_VK>(&input_subsystem, state->system, fullscreen);
+        emu_window = std::make_unique<EmuWindow_SDL3_VK>(&input_subsystem, system, fullscreen);
         break;
     case Settings::RendererBackend::Null:
-        state->emu_window = std::make_unique<EmuWindow_SDL3_Null>(&input_subsystem, state->system, fullscreen);
+        emu_window = std::make_unique<EmuWindow_SDL3_Null>(&input_subsystem, system, fullscreen);
         break;
     default:
         LOG_CRITICAL(Frontend, "Invalid renderer backend");
-        return SDL_APP_FAILURE;
+        return -1;
     }
 
 #ifdef _WIN32
     Common::Windows::SetCurrentTimerResolutionToMaximum();
-    state->system.CoreTiming().SetTimerResolutionNs(Common::Windows::GetCurrentTimerResolution());
+    system.CoreTiming().SetTimerResolutionNs(Common::Windows::GetCurrentTimerResolution());
 #endif
 
-    state->system.SetContentProvider(std::make_unique<FileSys::ContentProviderUnion>());
-    state->system.SetFilesystem(std::make_shared<FileSys::RealVfsFilesystem>());
-    state->system.GetFileSystemController().CreateFactories(*state->system.GetFilesystem());
-    state->system.GetUserChannel().clear();
+    system.SetContentProvider(std::make_unique<FileSys::ContentProviderUnion>());
+    system.SetFilesystem(std::make_shared<FileSys::RealVfsFilesystem>());
+    system.GetFileSystemController().CreateFactories(*system.GetFilesystem());
+    system.GetUserChannel().clear();
 
     Service::AM::FrontendAppletParameters load_parameters{
         .applet_id = Service::AM::AppletId::Application,
     };
-    const Core::SystemResultStatus load_result = state->system.Load(*state->emu_window, filepath, load_parameters);
+    const Core::SystemResultStatus load_result{system.Load(*emu_window, filepath, load_parameters)};
+
     switch (load_result) {
-    case Core::SystemResultStatus::Success:
-        break; // Expected case
     case Core::SystemResultStatus::ErrorGetLoader:
         LOG_CRITICAL(Frontend, "Failed to obtain loader for {}!", filepath);
-        return SDL_APP_FAILURE;
+        return -1;
     case Core::SystemResultStatus::ErrorLoader:
         LOG_CRITICAL(Frontend, "Failed to load ROM!");
-        return SDL_APP_FAILURE;
+        return -1;
     case Core::SystemResultStatus::ErrorNotInitialized:
         LOG_CRITICAL(Frontend, "CPUCore not initialized");
-        return SDL_APP_FAILURE;
+        return -1;
     case Core::SystemResultStatus::ErrorVideoCore:
         LOG_CRITICAL(Frontend, "Failed to initialize VideoCore!");
-        return SDL_APP_FAILURE;
+        return -1;
+    case Core::SystemResultStatus::Success:
+        break; // Expected case
     default:
-        const u16 loader_id = u16(Core::SystemResultStatus::ErrorLoader);
-        const u16 error_id = u16(load_result) - loader_id;
-        LOG_CRITICAL(Frontend,
-            "While attempting to load the ROM requested, an error occurred. Please "
-            "refer to the Eden wiki for more information or the Eden discord for "
-            "additional help.\n\nError Code: {:04X}-{:04X}\nError Description: {}",
-            loader_id, error_id, Loader::ResultStatus(error_id));
-        return SDL_APP_FAILURE;
+        if (static_cast<u32>(load_result) >
+            static_cast<u32>(Core::SystemResultStatus::ErrorLoader)) {
+            const u16 loader_id = static_cast<u16>(Core::SystemResultStatus::ErrorLoader);
+            const u16 error_id = static_cast<u16>(load_result) - loader_id;
+            LOG_CRITICAL(Frontend,
+                         "While attempting to load the ROM requested, an error occurred. Please "
+                         "refer to the Eden wiki for more information or the Eden discord for "
+                         "additional help.\n\nError Code: {:04X}-{:04X}\nError Description: {}",
+                         loader_id, error_id, static_cast<Loader::ResultStatus>(error_id));
+        }
+        break;
     }
 
     if (use_multiplayer) {
@@ -443,46 +439,40 @@ extern "C" SDL_AppResult SDL_AppInit(void **appstate, int argc, char **argv) {
             member->BindOnStatusMessageReceived(OnStatusMessageReceived);
             member->BindOnStateChanged(OnStateChanged);
             member->BindOnError(OnNetworkError);
-            LOG_DEBUG(Network, "Start connection to {}:{} with nickname {}", address, port, nickname);
+            LOG_DEBUG(Network, "Start connection to {}:{} with nickname {}", address, port,
+                      nickname);
             member->Join(nickname, address.c_str(), port, 0, Network::NoPreferredIP, password);
         } else {
             LOG_ERROR(Network, "Could not access RoomMember");
-            return SDL_APP_FAILURE;
+            return 0;
         }
     }
 
     // Core is loaded, start the GPU (makes the GPU contexts current to this thread)
-    state->system.GPU().Start();
-    state->system.GetCpuManager().OnGpuReady();
+    system.GPU().Start();
+    system.GetCpuManager().OnGpuReady();
 
     if (Settings::values.use_disk_shader_cache.GetValue()) {
-        state->system.Renderer().ReadRasterizer()->LoadDiskResources(
-            state->system.GetApplicationProcessProgramID(), std::stop_token{},
+        system.Renderer().ReadRasterizer()->LoadDiskResources(
+            system.GetApplicationProcessProgramID(), std::stop_token{},
             [](VideoCore::LoadCallbackStage, size_t value, size_t total) {});
     }
 
-    // don't do anything, SDL3 already exists for us :D
-    state->system.RegisterExitCallback([] {});
-    void(state->system.Run());
-    if (state->system.DebuggerEnabled())
-        state->system.InitializeDebugger();
-    return SDL_APP_SUCCESS;
-}
-extern "C" SDL_AppResult SDL_AppIterate(void *appstate) {
-    SdlState *state = (SdlState *)appstate;
-    return state->emu_window->IsOpen() ? SDL_APP_CONTINUE : SDL_APP_SUCCESS;
-}
-extern "C" SDL_AppResult SDL_AppEvent(void *appstate, SDL_Event *event) {
-    SdlState *state = (SdlState *)appstate;
-    state->emu_window->OnEvent(*event);
-    return SDL_APP_SUCCESS;
-}
-extern "C" void SDL_AppQuit(void *appstate, SDL_AppResult result) {
-    SdlState *state = (SdlState *)appstate;
-    state->system.DetachDebugger();
-    void(state->system.Pause());
-    state->system.ShutdownMainProcess();
-    delete state;
+    system.RegisterExitCallback([&] {
+        // Just exit right away.
+        exit(0);
+    });
+    void(system.Run());
+    if (system.DebuggerEnabled()) {
+        system.InitializeDebugger();
+    }
+    while (emu_window->IsOpen()) {
+        emu_window->WaitEvent();
+    }
+    system.DetachDebugger();
+    void(system.Pause());
+    system.ShutdownMainProcess();
+    return 0;
 }
 
 #define VMA_IMPLEMENTATION