Remove opts per tick frames

Don't compile GLSL/GLASM backends without OpenGL enabled. These codepaths are unused on non-OpenGL platforms.

Fixes other miscellaneous errors that pop up on Clang 23 as well.

Signed-off-by: Lizzie lizzie@eden-emu.dev
Signed-off-by: crueter crueter@eden-emu.dev

- [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions).
- [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md)
- [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability.

-------------------

Co-authored-by: crueter <crueter@eden-emu.dev>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4350
Reviewed-by: crueter <crueter@eden-emu.dev>
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>

.patch/vulkan-memory-allocator/0001-macos-clang.patch

@@ -0,0 +1,112 @@
+diff --git a/include/vk_mem_alloc.h b/include/vk_mem_alloc.h
+index 8df0364..4856064 100644
+--- a/include/vk_mem_alloc.h
++++ b/include/vk_mem_alloc.h
+@@ -3017,7 +3017,7 @@ remove them if not needed.
+ 
+ #if defined(__ANDROID_API__) && (__ANDROID_API__ < 16)
+ #include <cstdlib>
+-static void* vma_aligned_alloc(size_t alignment, size_t size)
++static inline void* vma_aligned_alloc(size_t alignment, size_t size)
+ {
+     // alignment must be >= sizeof(void*)
+     if(alignment < sizeof(void*))
+@@ -3860,7 +3860,7 @@ Returned value is the found element, if present in the collection or place where
+ new element with value (key) should be inserted.
+ */
+ template <typename CmpLess, typename IterT, typename KeyT>
+-static IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT& key, const CmpLess& cmp)
++static inline IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT& key, const CmpLess& cmp)
+ {
+     size_t down = 0;
+     size_t up = size_t(end - beg);
+@@ -3898,7 +3898,7 @@ Warning! O(n^2) complexity. Use only inside VMA_HEAVY_ASSERT.
+ T must be pointer type, e.g. VmaAllocation, VmaPool.
+ */
+ template<typename T>
+-static bool VmaValidatePointerArray(uint32_t count, const T* arr)
++static inline bool VmaValidatePointerArray(uint32_t count, const T* arr)
+ {
+     for (uint32_t i = 0; i < count; ++i)
+     {
+@@ -4188,13 +4188,13 @@ static void VmaFree(const VkAllocationCallbacks* pAllocationCallbacks, void* ptr
+ }
+ 
+ template<typename T>
+-static T* VmaAllocate(const VkAllocationCallbacks* pAllocationCallbacks)
++static inline T* VmaAllocate(const VkAllocationCallbacks* pAllocationCallbacks)
+ {
+     return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T), VMA_ALIGN_OF(T));
+ }
+ 
+ template<typename T>
+-static T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, size_t count)
++static inline T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, size_t count)
+ {
+     return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T) * count, VMA_ALIGN_OF(T));
+ }
+@@ -4204,14 +4204,14 @@ static T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, si
+ #define vma_new_array(allocator, type, count)   new(VmaAllocateArray<type>((allocator), (count)))(type)
+ 
+ template<typename T>
+-static void vma_delete(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr)
++static inline void vma_delete(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr)
+ {
+     ptr->~T();
+     VmaFree(pAllocationCallbacks, ptr);
+ }
+ 
+ template<typename T>
+-static void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr, size_t count)
++static inline void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr, size_t count)
+ {
+     if (ptr != VMA_NULL)
+     {
+@@ -4658,13 +4658,13 @@ void VmaVector<T, AllocatorT>::remove(size_t index)
+ #endif // _VMA_VECTOR_FUNCTIONS
+ 
+ template<typename T, typename allocatorT>
+-static void VmaVectorInsert(VmaVector<T, allocatorT>& vec, size_t index, const T& item)
++static inline void VmaVectorInsert(VmaVector<T, allocatorT>& vec, size_t index, const T& item)
+ {
+     vec.insert(index, item);
+ }
+ 
+ template<typename T, typename allocatorT>
+-static void VmaVectorRemove(VmaVector<T, allocatorT>& vec, size_t index)
++static inline void VmaVectorRemove(VmaVector<T, allocatorT>& vec, size_t index)
+ {
+     vec.remove(index);
+ }
+@@ -10620,19 +10620,19 @@ static void VmaFree(VmaAllocator hAllocator, void* ptr)
+ }
+ 
+ template<typename T>
+-static T* VmaAllocate(VmaAllocator hAllocator)
++static inline T* VmaAllocate(VmaAllocator hAllocator)
+ {
+     return (T*)VmaMalloc(hAllocator, sizeof(T), VMA_ALIGN_OF(T));
+ }
+ 
+ template<typename T>
+-static T* VmaAllocateArray(VmaAllocator hAllocator, size_t count)
++static inline T* VmaAllocateArray(VmaAllocator hAllocator, size_t count)
+ {
+     return (T*)VmaMalloc(hAllocator, sizeof(T) * count, VMA_ALIGN_OF(T));
+ }
+ 
+ template<typename T>
+-static void vma_delete(VmaAllocator hAllocator, T* ptr)
++static inline void vma_delete(VmaAllocator hAllocator, T* ptr)
+ {
+     if(ptr != VMA_NULL)
+     {
+@@ -10642,7 +10642,7 @@ static void vma_delete(VmaAllocator hAllocator, T* ptr)
+ }
+ 
+ template<typename T>
+-static void vma_delete_array(VmaAllocator hAllocator, T* ptr, size_t count)
++static inline void vma_delete_array(VmaAllocator hAllocator, T* ptr, size_t count)
+ {
+     if(ptr != VMA_NULL)
+     {

CMakeLists.txt

@@ -3,6 +3,7 @@
 
 cmake_minimum_required(VERSION 3.31)
 
+set(CMAKE_OSX_DEPLOYMENT_TARGET "15.0" CACHE STRING "macOS deployment target")
 project(yuzu)
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/CMakeModules")
@@ -238,7 +239,7 @@ option(YUZU_USE_BUNDLED_SIRIT "Download bundled sirit" ${BUNDLED_SIRIT_DEFAULT})
 # FreeBSD 15+ has libusb, versions below should disable it
 cmake_dependent_option(ENABLE_LIBUSB "Enable the use of LibUSB" ON "WIN32 OR LINUX OR FREEBSD OR APPLE" OFF)
 
-cmake_dependent_option(ENABLE_OPENGL "Enable OpenGL" ON "NOT (WIN32 AND ARCHITECTURE_arm64) AND NOT APPLE" OFF)
+cmake_dependent_option(ENABLE_OPENGL "Enable OpenGL" ON "NOT (WIN32 AND ARCHITECTURE_arm64) AND NOT APPLE AND NOT ANDROID" OFF)
 mark_as_advanced(FORCE ENABLE_OPENGL)
 
 option(ENABLE_WEB_SERVICE "Enable web services (telemetry, etc.)" ON)
@@ -512,7 +513,7 @@ endfunction()
 # =============================================
 
 if (APPLE)
-    foreach(fw Carbon Metal Cocoa IOKit CoreVideo CoreMedia Security UniformTypeIdentifiers)
+    foreach(fw Carbon Metal Cocoa IOKit CoreVideo CoreMedia Security UniformTypeIdentifiers Foundation)
         find_library(${fw}_LIBRARY ${fw} REQUIRED)
         list(APPEND PLATFORM_LIBRARIES ${${fw}_LIBRARY})
     endforeach()

cpmfile.json

@@ -1,12 +1,4 @@
 {
-    "": {
-        "ci": true,
-        "hash": "9f50d993c39529e022ad456163de91ac5934e16faf8fc348f305355fc0a643c534f87ded707e11bd03bcbc0a1760bd20852b6533a67ac0558a078385181cb184",
-        "name": "SDL3",
-        "package": "SDL3",
-        "repo": "crueter-ci/SDL3",
-        "version": "3.4.14-1788231389-147a8ee32d"
-    },
     "biscuit": {
         "hash": "1229f345b014f7ca544dedb4edb3311e41ba736f9aa9a67f88b5f26f3c983288c6bb6cdedcfb0b8a02c63088a37e6a0d7ba97d9c2a4d721b213916327cffe28a",
         "min_version": "0.9.1",
@@ -68,10 +60,10 @@
     },
     "discord-rpc": {
         "find_args": "MODULE",
-        "hash": "8213c43dcb0f7d479f5861091d111ed12fbdec1e62e6d729d65a4bc181d82f48a35d5fd3cd5c291f2393ac7c9681eabc6b76609755f55376284c8a8d67e148f3",
+        "hash": "8d680b3a16d6f6bf292ad823cf8635595ff986f2a49f758e3009f505b540a9d75194a8cf44cddea75736a8c3e3b5160166e716a0939ce3f18cc463cf648753fe",
         "package": "DiscordRPC",
         "repo": "eden-emulator/discord-rpc",
-        "version": "0d8b2d6a37"
+        "version": "76616d8675"
     },
     "enet": {
         "find_args": "MODULE",
@@ -311,6 +303,9 @@
         "find_args": "CONFIG",
         "hash": "deb5902ef8db0e329fbd5f3f4385eb0e26bdd9f14f3a2334823fb3fe18f36bc5d235d620d6e5f6fe3551ec3ea7038638899db8778c09f6d5c278f5ff95c3344b",
         "package": "VulkanMemoryAllocator",
+        "patches": [
+            "0001-macos-clang.patch"
+        ],
         "repo": "GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator",
         "version": "v3.3.0"
     },

docs/Options.md

@@ -76,7 +76,7 @@ The following options are desktop only.
 
 - `ENABLE_LIBUSB` (ON) Enable the use of the libusb input backend (HIGHLY RECOMMENDED)
 - `ENABLE_OPENGL` (ON) Enable the OpenGL graphics backend
-  - Unavailable on Windows/ARM64
+  - Unavailable on Windows/ARM64 and on Android
   - You probably shouldn't turn this off.
 
 ### Qt

externals/tz/tz/tz.h

@@ -11,6 +11,7 @@
 #include <limits>
 #include <span>
 #include <array>
+#include <algorithm>
 #include <time.h>
 
 namespace Tz {

src/common/demangle.cpp

@@ -5,6 +5,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <string>
+#include <cstdlib>
 #include <string_view>
 #ifdef _WIN32
 #include <llvm/Demangle/Demangle.h>

src/common/scratch_buffer.h

@@ -1,9 +1,13 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
 #include <iterator>
+#include <cstring>
 
 #include "common/make_unique_for_overwrite.h"
 
@@ -61,7 +65,7 @@ public:
     void resize(size_type size) {
         if (size > buffer_capacity) {
             auto new_buffer = Common::make_unique_for_overwrite<T[]>(size);
-            std::move(buffer.get(), buffer.get() + buffer_capacity, new_buffer.get());
+            std::memcpy(new_buffer.get(), buffer.get(), buffer_capacity * sizeof(T));
             buffer = std::move(new_buffer);
             buffer_capacity = size;
         }

src/common/uuid.h

@@ -10,6 +10,7 @@
 #include <functional>
 #include <span>
 #include <string>
+#include <type_traits>
 
 #include "common/common_types.h"
 

src/core/debugger/debugger.cpp

@@ -6,13 +6,13 @@
 
 #include <mutex>
 #include <utility>
+#include <type_traits>
+
 #include <boost/asio.hpp>
 #include <boost/version.hpp>
-
 #if BOOST_VERSION > 108400 && (!defined(_WINDOWS) && !defined(__ANDROID__)) || defined(YUZU_BOOST_v1)
 #define USE_BOOST_v1
 #endif
-
 #ifdef USE_BOOST_v1
 #include <boost/process/v1/async_pipe.hpp>
 #else
@@ -358,7 +358,7 @@ private:
 
         ConnectionState(boost::asio::ip::tcp::socket&& client_socket_, async_pipe signal_pipe_, Kernel::KernelCore& kernel)
             : client_socket{std::move(client_socket_)}
-            , signal_pipe{signal_pipe_}
+            , signal_pipe{std::move(signal_pipe_)}
             , active_thread{kernel, nullptr}
         {}
 

src/core/file_sys/fssystem/fs_types.h

@@ -1,4 +1,4 @@
-// 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 2023 yuzu Emulator Project
@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <type_traits>
 #include "common/common_funcs.h"
 
 namespace FileSys {

src/core/file_sys/fssystem/fssystem_aes_ctr_counter_extended_storage.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <optional>
+#include <type_traits>
 
 #include "common/literals.h"
 #include "core/file_sys/fssystem/fs_i_storage.h"

src/core/file_sys/fssystem/fssystem_bucket_tree.cpp

@@ -4,6 +4,7 @@
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <type_traits>
 #include "core/file_sys/errors.h"
 #include "core/file_sys/fssystem/fssystem_bucket_tree.h"
 #include "core/file_sys/fssystem/fssystem_bucket_tree_utils.h"

src/core/file_sys/fssystem/fssystem_bucket_tree.h

@@ -1,9 +1,13 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
 #include <mutex>
+#include <type_traits>
 
 #include "common/alignment.h"
 #include "common/common_funcs.h"

src/core/file_sys/fssystem/fssystem_bucket_tree_template_impl.h

@@ -1,4 +1,4 @@
-// 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 2023 yuzu Emulator Project
@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <type_traits>
 #include "core/file_sys/errors.h"
 #include "core/file_sys/fssystem/fssystem_bucket_tree.h"
 #include "core/file_sys/fssystem/fssystem_bucket_tree_utils.h"

src/core/file_sys/fssystem/fssystem_compressed_storage.h

@@ -6,6 +6,8 @@
 
 #pragma once
 
+#include <type_traits>
+#include <cstddef>
 #include "common/literals.h"
 
 #include "core/file_sys/errors.h"

src/core/file_sys/fssystem/fssystem_hierarchical_integrity_verification_storage.h

@@ -1,8 +1,12 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
+#include <type_traits>
 #include "common/alignment.h"
 #include "core/file_sys/fssystem/fs_i_storage.h"
 #include "core/file_sys/fssystem/fs_types.h"

src/core/file_sys/fssystem/fssystem_indirect_storage.h

@@ -6,6 +6,9 @@
 
 #pragma once
 
+#include <type_traits>
+#include <array>
+#include <cstddef>
 #include "core/file_sys/errors.h"
 #include "core/file_sys/fssystem/fs_i_storage.h"
 #include "core/file_sys/fssystem/fssystem_bucket_tree.h"

src/core/file_sys/fssystem/fssystem_integrity_verification_storage.h

@@ -1,9 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
 #include <optional>
+#include <array>
+#include <cstddef>
+#include <type_traits>
 
 #include "core/file_sys/fssystem/fs_i_storage.h"
 #include "core/file_sys/fssystem/fs_types.h"

src/core/file_sys/fssystem/fssystem_nca_file_system_driver.h

@@ -1,4 +1,4 @@
-// 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 2023 yuzu Emulator Project
@@ -6,6 +6,8 @@
 
 #pragma once
 
+#include <type_traits>
+#include <cstddef>
 #include "core/file_sys/fssystem/fssystem_compression_common.h"
 #include "core/file_sys/fssystem/fssystem_nca_header.h"
 #include "core/file_sys/vfs/vfs.h"

src/core/file_sys/fssystem/fssystem_nca_header.h

@@ -1,8 +1,14 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
+#include <type_traits>
+#include <array>
+#include <cstddef>
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/literals.h"

src/core/hle/kernel/k_page_table_base.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <memory>
+#include <type_traits>
 
 #include "common/common_funcs.h"
 #include "common/page_table.h"

src/core/hle/kernel/k_process.cpp

@@ -1021,6 +1021,9 @@ Result KProcess::Run(KernelCore& kernel, s32 priority, size_t stack_size) {
 
     // Suspend for debug, if we should.
     if (kernel.System().DebuggerEnabled()) {
+        LOG_INFO(Debug_GDBStub,
+                 "GDB stub enabled; suspending guest process until a debugger continues execution on port {}",
+                 Settings::values.gdbstub_port.GetValue());
         main_thread->RequestSuspend(kernel, SuspendType::Debug);
     }
 

src/core/hle/result.h

@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <type_traits>
 #include "common/assert.h"
 #include "common/bit_field.h"
 #include "common/common_funcs.h"

src/core/hle/service/bcat/bcat_types.h

@@ -1,9 +1,13 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2024 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #pragma once
 
 #include <array>
+#include <type_traits>
 #include <functional>
 
 #include "common/common_funcs.h"

src/core/hle/service/caps/caps_manager.cpp

@@ -175,19 +175,10 @@ Result AlbumManager::LoadAlbumScreenShotImage(LoadAlbumScreenShotImageOutput& ou
         return ResultIsNotMounted;
     }
 
-    out_image_output = {
-        .width = 1280,
-        .height = 720,
-        .attribute =
-            {
-                .unknown_0{},
-                .orientation = AlbumImageOrientation::None,
-                .unknown_1{},
-                .unknown_2{},
-                .pad163{},
-            },
-        .pad179{},
-    };
+    out_image_output = {};
+    out_image_output.width = 1280;
+    out_image_output.height = 720;
+    out_image_output.attribute.orientation = AlbumImageOrientation::None;
 
     std::filesystem::path path;
     const auto result = GetFile(path, file_id);
@@ -211,19 +202,10 @@ Result AlbumManager::LoadAlbumScreenShotThumbnail(
         return ResultIsNotMounted;
     }
 
-    out_image_output = {
-        .width = 320,
-        .height = 180,
-        .attribute =
-            {
-                .unknown_0{},
-                .orientation = AlbumImageOrientation::None,
-                .unknown_1{},
-                .unknown_2{},
-                .pad163{},
-            },
-        .pad179{},
-    };
+    out_image_output = {};
+    out_image_output.width = 320;
+    out_image_output.height = 180;
+    out_image_output.attribute.orientation = AlbumImageOrientation::None;
 
     std::filesystem::path path;
     const auto result = GetFile(path, file_id);

src/core/hle/service/caps/caps_su.cpp

@@ -73,13 +73,8 @@ void IScreenShotApplicationService::CaptureAndSaveScreenshot(AlbumReportOption r
     Layout::FramebufferLayout layout =
         Layout::DefaultFrameLayout(screenshot_width, screenshot_height);
 
-    const Capture::ScreenShotAttribute attribute{
-        .unknown_0{},
-        .orientation = Capture::AlbumImageOrientation::None,
-        .unknown_1{},
-        .unknown_2{},
-        .pad163{},
-    };
+    Capture::ScreenShotAttribute attribute{};
+    attribute.orientation = Capture::AlbumImageOrientation::None;
 
     renderer.RequestScreenshot(
         image_data.data(),

src/core/hle/service/caps/caps_types.h

@@ -1,8 +1,12 @@
+// 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
 
 #pragma once
 
+#include <type_traits>
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 

src/core/hle/service/ldn/ldn_types.h

@@ -1,4 +1,4 @@
-// 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 2022 yuzu Emulator Project
@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <type_traits>
 #include <fmt/ranges.h>
 
 #include "common/common_funcs.h"

src/core/hle/service/psc/time/common.h

@@ -8,6 +8,7 @@
 
 #include <array>
 #include <chrono>
+#include <type_traits>
 #include <fmt/ranges.h>
 
 #include "common/common_types.h"

src/core/hle/service/psc/time/shared_memory.h

@@ -1,10 +1,13 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
 #include <array>
-
+#include <type_traits>
 #include "common/common_types.h"
 #include "core/hle/service/psc/time/common.h"
 

src/core/hle/service/set/settings_types.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <array>
+#include <type_traits>
 
 #include "common/bit_field.h"
 #include "common/common_funcs.h"

src/hid_core/hid_types.h

@@ -1,8 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
+#include <cstddef>
+#include <array>
+#include <type_traits>
+
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"

src/shader_recompiler/CMakeLists.txt

@@ -6,55 +6,6 @@
 
 add_library(shader_recompiler STATIC
     backend/bindings.h
-    backend/glasm/emit_glasm.cpp
-    backend/glasm/emit_glasm.h
-    backend/glasm/emit_glasm_barriers.cpp
-    backend/glasm/emit_glasm_bitwise_conversion.cpp
-    backend/glasm/emit_glasm_composite.cpp
-    backend/glasm/emit_glasm_context_get_set.cpp
-    backend/glasm/emit_glasm_control_flow.cpp
-    backend/glasm/emit_glasm_convert.cpp
-    backend/glasm/emit_glasm_floating_point.cpp
-    backend/glasm/emit_glasm_image.cpp
-    backend/glasm/emit_glasm_instructions.h
-    backend/glasm/emit_glasm_integer.cpp
-    backend/glasm/emit_glasm_logical.cpp
-    backend/glasm/emit_glasm_memory.cpp
-    backend/glasm/emit_glasm_not_implemented.cpp
-    backend/glasm/emit_glasm_select.cpp
-    backend/glasm/emit_glasm_shared_memory.cpp
-    backend/glasm/emit_glasm_special.cpp
-    backend/glasm/emit_glasm_undefined.cpp
-    backend/glasm/emit_glasm_warp.cpp
-    backend/glasm/glasm_emit_context.cpp
-    backend/glasm/glasm_emit_context.h
-    backend/glasm/reg_alloc.cpp
-    backend/glasm/reg_alloc.h
-    backend/glsl/emit_glsl.cpp
-    backend/glsl/emit_glsl.h
-    backend/glsl/emit_glsl_atomic.cpp
-    backend/glsl/emit_glsl_barriers.cpp
-    backend/glsl/emit_glsl_bitwise_conversion.cpp
-    backend/glsl/emit_glsl_composite.cpp
-    backend/glsl/emit_glsl_context_get_set.cpp
-    backend/glsl/emit_glsl_control_flow.cpp
-    backend/glsl/emit_glsl_convert.cpp
-    backend/glsl/emit_glsl_floating_point.cpp
-    backend/glsl/emit_glsl_image.cpp
-    backend/glsl/emit_glsl_instructions.h
-    backend/glsl/emit_glsl_integer.cpp
-    backend/glsl/emit_glsl_logical.cpp
-    backend/glsl/emit_glsl_memory.cpp
-    backend/glsl/emit_glsl_not_implemented.cpp
-    backend/glsl/emit_glsl_select.cpp
-    backend/glsl/emit_glsl_shared_memory.cpp
-    backend/glsl/emit_glsl_special.cpp
-    backend/glsl/emit_glsl_undefined.cpp
-    backend/glsl/emit_glsl_warp.cpp
-    backend/glsl/glsl_emit_context.cpp
-    backend/glsl/glsl_emit_context.h
-    backend/glsl/var_alloc.cpp
-    backend/glsl/var_alloc.h
     backend/spirv/emit_spirv.cpp
     backend/spirv/emit_spirv.h
     backend/spirv/emit_spirv_atomic.cpp
@@ -239,9 +190,60 @@ add_library(shader_recompiler STATIC
     program_header.h
     runtime_info.h
     shader_info.h
-    varying_state.h
+    varying_state.h)
 
-)
+if (ENABLE_OPENGL)
+    target_sources(shader_recompiler PRIVATE
+        backend/glasm/emit_glasm.cpp
+        backend/glasm/emit_glasm.h
+        backend/glasm/emit_glasm_barriers.cpp
+        backend/glasm/emit_glasm_bitwise_conversion.cpp
+        backend/glasm/emit_glasm_composite.cpp
+        backend/glasm/emit_glasm_context_get_set.cpp
+        backend/glasm/emit_glasm_control_flow.cpp
+        backend/glasm/emit_glasm_convert.cpp
+        backend/glasm/emit_glasm_floating_point.cpp
+        backend/glasm/emit_glasm_image.cpp
+        backend/glasm/emit_glasm_instructions.h
+        backend/glasm/emit_glasm_integer.cpp
+        backend/glasm/emit_glasm_logical.cpp
+        backend/glasm/emit_glasm_memory.cpp
+        backend/glasm/emit_glasm_not_implemented.cpp
+        backend/glasm/emit_glasm_select.cpp
+        backend/glasm/emit_glasm_shared_memory.cpp
+        backend/glasm/emit_glasm_special.cpp
+        backend/glasm/emit_glasm_undefined.cpp
+        backend/glasm/emit_glasm_warp.cpp
+        backend/glasm/glasm_emit_context.cpp
+        backend/glasm/glasm_emit_context.h
+        backend/glasm/reg_alloc.cpp
+        backend/glasm/reg_alloc.h
+        backend/glsl/emit_glsl.cpp
+        backend/glsl/emit_glsl.h
+        backend/glsl/emit_glsl_atomic.cpp
+        backend/glsl/emit_glsl_barriers.cpp
+        backend/glsl/emit_glsl_bitwise_conversion.cpp
+        backend/glsl/emit_glsl_composite.cpp
+        backend/glsl/emit_glsl_context_get_set.cpp
+        backend/glsl/emit_glsl_control_flow.cpp
+        backend/glsl/emit_glsl_convert.cpp
+        backend/glsl/emit_glsl_floating_point.cpp
+        backend/glsl/emit_glsl_image.cpp
+        backend/glsl/emit_glsl_instructions.h
+        backend/glsl/emit_glsl_integer.cpp
+        backend/glsl/emit_glsl_logical.cpp
+        backend/glsl/emit_glsl_memory.cpp
+        backend/glsl/emit_glsl_not_implemented.cpp
+        backend/glsl/emit_glsl_select.cpp
+        backend/glsl/emit_glsl_shared_memory.cpp
+        backend/glsl/emit_glsl_special.cpp
+        backend/glsl/emit_glsl_undefined.cpp
+        backend/glsl/emit_glsl_warp.cpp
+        backend/glsl/glsl_emit_context.cpp
+        backend/glsl/glsl_emit_context.h
+        backend/glsl/var_alloc.cpp
+        backend/glsl/var_alloc.h)
+endif()
 
 target_link_libraries(shader_recompiler PUBLIC common fmt::fmt sirit::sirit)
 

src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp

@@ -553,17 +553,6 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateIn
     }
 }
 
-template <typename Descriptors, typename Descriptor, typename Func>
-static u32 Add(Descriptors& descriptors, const Descriptor& desc, Func&& pred) {
-    // TODO: Handle arrays
-    const auto it{std::ranges::find_if(descriptors, pred)};
-    if (it != descriptors.end()) {
-        return static_cast<u32>(std::distance(descriptors.begin(), it));
-    }
-    descriptors.push_back(desc);
-    return static_cast<u32>(descriptors.size()) - 1;
-}
-
 void JoinStorageInfo(Info& base, Info& source) {
     auto& descriptors = base.storage_buffers_descriptors;
     for (auto& desc : source.storage_buffers_descriptors) {

src/video_core/CMakeLists.txt

@@ -20,6 +20,7 @@ add_library(video_core STATIC
     buffer_cache/buffer_cache.h
     buffer_cache/memory_tracker_base.h
     buffer_cache/usage_tracker.h
+    buffer_cache/virtual_range_cache.h
     buffer_cache/word_manager.h
     cache_types.h
     capture.h
@@ -166,6 +167,8 @@ add_library(video_core STATIC
     renderer_vulkan/vk_fence_manager.h
     renderer_vulkan/vk_graphics_pipeline.cpp
     renderer_vulkan/vk_graphics_pipeline.h
+    renderer_vulkan/vk_multi_range_buffer.cpp
+    renderer_vulkan/vk_multi_range_buffer.h
     renderer_vulkan/vk_master_semaphore.cpp
     renderer_vulkan/vk_master_semaphore.h
     renderer_vulkan/vk_pipeline_cache.cpp

src/video_core/buffer_cache/buffer_base.h

@@ -20,6 +20,8 @@ namespace VideoCommon {
 
 enum class BufferFlagBits {
     Picked = 1 << 0,
+    CachedWrites = 1 << 1,
+    PreemtiveDownload = 1 << 2,
 };
 DECLARE_ENUM_FLAG_OPERATORS(BufferFlagBits)
 
@@ -56,6 +58,15 @@ public:
         flags |= BufferFlagBits::Picked;
     }
 
+    void MarkPreemtiveDownload() noexcept {
+        flags |= BufferFlagBits::PreemtiveDownload;
+    }
+
+    /// Unmark buffer as picked
+    void Unpick() noexcept {
+        flags &= ~BufferFlagBits::Picked;
+    }
+
     /// Increases the likeliness of this being a stream buffer
     void IncreaseStreamScore(int score) noexcept {
         stream_score += score;
@@ -76,6 +87,15 @@ public:
         return True(flags & BufferFlagBits::Picked);
     }
 
+    /// Returns true when the buffer has pending cached writes
+    [[nodiscard]] bool HasCachedWrites() const noexcept {
+        return True(flags & BufferFlagBits::CachedWrites);
+    }
+
+    bool IsPreemtiveDownload() const noexcept {
+        return True(flags & BufferFlagBits::PreemtiveDownload);
+    }
+
     /// Returns the base CPU address of the buffer
     [[nodiscard]] VAddr CpuAddr() const noexcept {
         return cpu_addr;

src/video_core/buffer_cache/buffer_cache.h

@@ -7,7 +7,6 @@
 #pragma once
 
 #include <algorithm>
-#include <limits>
 #include <memory>
 #include <numeric>
 
@@ -113,6 +112,13 @@ void BufferCache<P>::TickFrame() {
     async_buffers_death_ring.clear();
 }
 
+template <class P>
+void BufferCache<P>::UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size) {
+    if constexpr (requires { runtime.BindMultiRangeStorageBuffer(u64{}, bool{}); }) {
+        virtual_ranges.Unmap(as_id, gpu_addr, size);
+    }
+}
+
 template <class P>
 void BufferCache<P>::WriteMemory(DAddr device_addr, u64 size) {
     if (memory_tracker.IsRegionGpuModified(device_addr, size)) {
@@ -122,6 +128,25 @@ void BufferCache<P>::WriteMemory(DAddr device_addr, u64 size) {
     memory_tracker.MarkRegionAsCpuModified(device_addr, size);
 }
 
+template <class P>
+void BufferCache<P>::CachedWriteMemory(DAddr device_addr, u64 size) {
+    const bool is_dirty = IsRegionRegistered(device_addr, size);
+    if (!is_dirty) {
+        return;
+    }
+    DAddr aligned_start = Common::AlignDown(device_addr, DEVICE_PAGESIZE);
+    DAddr aligned_end = Common::AlignUp(device_addr + size, DEVICE_PAGESIZE);
+    if (!IsRegionGpuModified(aligned_start, aligned_end - aligned_start)) {
+        WriteMemory(device_addr, size);
+        return;
+    }
+
+    tmp_buffer.resize_destructive(size);
+    device_memory.ReadBlockUnsafe(device_addr, tmp_buffer.data(), size);
+
+    InlineMemoryImplementation(device_addr, size, tmp_buffer);
+}
+
 template <class P>
 bool BufferCache<P>::OnCPUWrite(DAddr device_addr, u64 size) {
     const bool is_dirty = IsRegionRegistered(device_addr, size);
@@ -190,8 +215,8 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     BufferId buffer_b;
     do {
         channel_state->has_deleted_buffers = false;
-        buffer_a = FindBuffer(*cpu_src_address, static_cast<u32>(amount));
-        buffer_b = FindBuffer(*cpu_dest_address, static_cast<u32>(amount));
+        buffer_a = FindBuffer(*cpu_src_address, static_cast<u32>(amount), false);
+        buffer_b = FindBuffer(*cpu_dest_address, static_cast<u32>(amount), false);
     } while (channel_state->has_deleted_buffers);
     auto& src_buffer = slot_buffers[buffer_a];
     auto& dest_buffer = slot_buffers[buffer_b];
@@ -247,7 +272,7 @@ bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) {
     ClearDownload(*cpu_dst_address, size);
     gpu_modified_ranges.Subtract(*cpu_dst_address, size);
 
-    const BufferId buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
+    const BufferId buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size), false);
     Buffer& dest_buffer = slot_buffers[buffer];
     const u32 offset = dest_buffer.Offset(*cpu_dst_address);
     runtime.ClearBuffer(dest_buffer, offset, size, value);
@@ -269,7 +294,7 @@ std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainBuffer(GPUVAddr gpu_ad
 template <class P>
 std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainCPUBuffer(
     DAddr device_addr, u32 size, ObtainBufferSynchronize sync_info, ObtainBufferOperation post_op) {
-    const BufferId buffer_id = FindBuffer(device_addr, size);
+    const BufferId buffer_id = FindBuffer(device_addr, size, false);
     Buffer& buffer = slot_buffers[buffer_id];
 
     // synchronize op
@@ -404,7 +429,7 @@ void BufferCache<P>::UnbindGraphicsStorageBuffers(size_t stage) {
 }
 
 template <class P>
-void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index,
+bool BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index,
                                                u32 cbuf_offset, bool is_written) {
     const bool already_enabled =
         ((channel_state->enabled_storage_buffers[stage] >> ssbo_index) & 1U) != 0;
@@ -415,7 +440,7 @@ void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index,
                 LOG_WARNING(HW_GPU,
                             "Skipping graphics storage buffer {} due to driver limit {}",
                             ssbo_index, max_bindings);
-                return;
+                return false;
             }
         }
     }
@@ -431,6 +456,7 @@ void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index,
     const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset;
     channel_state->storage_buffers[stage][ssbo_index] =
         StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
+    return (channel_state->storage_buffers[stage][ssbo_index].buffer_id != NULL_BUFFER_ID);
 }
 
 template <class P>
@@ -743,6 +769,16 @@ void BufferCache<P>::BindHostIndexBuffer() {
     }
 }
 
+template <class P>
+void BufferCache<P>::BindHostVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size,
+                                          u32 stride) {
+    if constexpr (IS_OPENGL) {
+        runtime.BindVertexBuffer(index, buffer, offset, size, stride);
+    } else {
+        runtime.BindVertexBuffer(index, buffer.Handle(), offset, size, stride);
+    }
+}
+
 template <class P>
 Binding& BufferCache<P>::VertexBufferSlot(u32 index) {
     ASSERT(index < NUM_VERTEX_BUFFERS);
@@ -969,11 +1005,85 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
     channel_state->fast_bound_uniform_buffers[stage] &= ~(1u << binding_index);
 }
 
+template <class P>
+void BufferCache<P>::ResolveMultiRangeStorage(Binding& binding, bool is_written,
+                                              std::vector<MultiRangeSegment>& pool) {
+    binding.segment_first = 0;
+    binding.segment_count = 0;
+    if constexpr (requires { runtime.BindMultiRangeStorageBuffer(u64{}, bool{}); }) {
+        if (binding.gpu_addr == 0 || binding.size == 0) {
+            return;
+        }
+        if (is_written && !runtime.PrefersSparseSources()) {
+            return;
+        }
+        const VirtualSegments* found =
+            virtual_ranges.Query(*gpu_memory, binding.gpu_addr, binding.size);
+        if (!found || found->size() < 2) {
+            return;
+        }
+        const VirtualSegments segments = *found;
+        const u32 first = static_cast<u32>(pool.size());
+        const bool prefer_sparse = runtime.PrefersSparseSources();
+        for (const VirtualSegment& segment : segments) {
+            const BufferId buffer_id =
+                FindBuffer(segment.device_addr, segment.size, prefer_sparse);
+            if (!buffer_id) {
+                pool.resize(first);
+                return;
+            }
+            pool.push_back(MultiRangeSegment{
+                .buffer_id = buffer_id,
+                .device_addr = segment.device_addr,
+                .size = segment.size,
+            });
+        }
+        binding.segment_first = first;
+        binding.segment_count = static_cast<u32>(segments.size());
+    }
+}
+
+template <class P>
+bool BufferCache<P>::BindMultiRangeStorage(const Binding& binding, bool is_written,
+                                           std::span<const MultiRangeSegment> pool) {
+    if constexpr (requires { runtime.BindMultiRangeStorageBuffer(u64{}, bool{}); }) {
+        if (binding.segment_count < 2) {
+            return false;
+        }
+        if (binding.segment_first + binding.segment_count > pool.size()) {
+            return false;
+        }
+        const u64 key = (static_cast<u64>(gpu_memory->GetID()) << 48) ^ binding.gpu_addr;
+        runtime.ResetMultiRange();
+        for (u32 index = 0; index < binding.segment_count; ++index) {
+            const MultiRangeSegment& segment = pool[binding.segment_first + index];
+            Buffer& buffer = slot_buffers[segment.buffer_id];
+            TouchBuffer(buffer, segment.buffer_id);
+            if (SynchronizeBuffer(buffer, segment.device_addr, segment.size)) {
+                runtime.InvalidateMultiRange(key);
+            }
+            const u32 offset = buffer.Offset(segment.device_addr);
+            buffer.MarkUsage(offset, segment.size);
+            if (is_written) {
+                MarkWrittenBuffer(segment.buffer_id, segment.device_addr, segment.size);
+            }
+            runtime.PushMultiRangeSource(buffer, offset, segment.size);
+        }
+        return runtime.BindMultiRangeStorageBuffer(key, is_written);
+    } else {
+        return false;
+    }
+}
+
 template <class P>
 void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
     u32 binding_index = 0;
     ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
         const Binding& binding = channel_state->storage_buffers[stage][index];
+        const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
+        if (BindMultiRangeStorage(binding, is_written, graphics_segments)) {
+            return;
+        }
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
@@ -981,7 +1091,6 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
 
         const u32 offset = buffer.Offset(binding.device_addr);
         buffer.MarkUsage(offset, size);
-        const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
 
         if (is_written) {
             MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
@@ -1110,6 +1219,11 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
     u32 binding_index = 0;
     ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
         const Binding& binding = channel_state->compute_storage_buffers[index];
+        const bool is_written =
+            ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
+        if (BindMultiRangeStorage(binding, is_written, compute_segments)) {
+            return;
+        }
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
@@ -1117,8 +1231,6 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
 
         const u32 offset = buffer.Offset(binding.device_addr);
         buffer.MarkUsage(offset, size);
-        const bool is_written =
-            ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
 
         if (is_written) {
             MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
@@ -1164,11 +1276,12 @@ void BufferCache<P>::BindHostComputeTextureBuffers() {
 
 template <class P>
 void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
+    graphics_segments.clear();
     BufferOperations([&]() {
         if (is_indexed) {
             UpdateIndexBuffer();
         }
-        UpdateVertexBuffers(is_indexed);
+        UpdateVertexBuffers();
         UpdateTransformFeedbackBuffers();
         for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
             UpdateUniformBuffers(stage);
@@ -1183,6 +1296,7 @@ void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
 
 template <class P>
 void BufferCache<P>::DoUpdateComputeBuffers() {
+    compute_segments.clear();
     BufferOperations([&]() {
         UpdateComputeUniformBuffers();
         UpdateComputeStorageBuffers();
@@ -1205,11 +1319,11 @@ void BufferCache<P>::UpdateIndexBuffer() {
         auto inline_index_size = static_cast<u32>(draw_state.inline_index_draw_indexes.size());
         u32 buffer_size = Common::AlignUp(inline_index_size, CACHING_PAGESIZE);
         if (inline_buffer_id == NULL_BUFFER_ID) [[unlikely]] {
-            inline_buffer_id = CreateBuffer(0, buffer_size);
+            inline_buffer_id = CreateBuffer(0, buffer_size, false);
         }
         if (slot_buffers[inline_buffer_id].SizeBytes() < buffer_size) [[unlikely]] {
             slot_buffers.erase(inline_buffer_id);
-            inline_buffer_id = CreateBuffer(0, buffer_size);
+            inline_buffer_id = CreateBuffer(0, buffer_size, false);
         }
         channel_state->index_buffer = Binding{
             .device_addr = 0,
@@ -1222,14 +1336,9 @@ void BufferCache<P>::UpdateIndexBuffer() {
     const GPUVAddr gpu_addr_begin = index_buffer_ref.StartAddress();
     const GPUVAddr gpu_addr_end = index_buffer_ref.EndAddress();
     const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin);
-    u64 address_size = 0;
-    if (gpu_addr_end > gpu_addr_begin) {
-        address_size = (std::min)(gpu_addr_end - gpu_addr_begin,
-                                  u64{(std::numeric_limits<u32>::max)()});
-    }
-    const u64 draw_size = (u64{index_buffer_ref.count} + u64{index_buffer_ref.first}) *
-                          u64{index_buffer_ref.FormatSizeInBytes()};
-    const u32 size = static_cast<u32>((std::min)(address_size, draw_size));
+    const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin);
+    const u32 draw_size = (index_buffer_ref.count + index_buffer_ref.first) * u32(index_buffer_ref.FormatSizeInBytes());
+    const u32 size = (std::min)(address_size, draw_size);
     if (size == 0 || !device_addr) {
         channel_state->index_buffer = NULL_BINDING;
         return;
@@ -1237,147 +1346,25 @@ void BufferCache<P>::UpdateIndexBuffer() {
     channel_state->index_buffer = Binding{
         .device_addr = *device_addr,
         .size = size,
-        .buffer_id = FindBuffer(*device_addr, size),
+        .buffer_id = FindBuffer(*device_addr, size, false),
     };
 }
 
 template <class P>
-u64 BufferCache<P>::DrawMaxIndex() {
-    if (max_index_scanned) {
-        return cached_max_index;
-    }
-    max_index_scanned = true;
-    cached_max_index = 0;
-    const auto& index_buffer_ref = maxwell3d->draw_manager.draw_state.index_buffer;
-    const u32 count = index_buffer_ref.count;
-    if (count == 0) {
-        return 0;
-    }
-    index_scan_buffer.resize_destructive(count);
-    gpu_memory->ReadBlockUnsafe(index_buffer_ref.IndexStart(), index_scan_buffer.data(),
-                                size_t{count} * sizeof(u32));
-    u32 restart_index = (std::numeric_limits<u32>::max)();
-    if (maxwell3d->regs.primitive_restart.enabled != 0) {
-        restart_index = maxwell3d->regs.primitive_restart.index;
-    }
-    u32 max_index = 0;
-    for (u32 i = 0; i < count; ++i) {
-        const u32 value = index_scan_buffer[i];
-        if (value == restart_index) {
-            continue;
-        }
-        max_index = (std::max)(max_index, value);
-    }
-    cached_max_index = max_index;
-    return cached_max_index;
-}
-
-template <class P>
-u64 BufferCache<P>::StreamAttributeExtent(u32 index) {
-    using VertexAttribute = typename Maxwell::VertexAttribute;
-    if (stream_extents_valid) {
-        return stream_extents[index];
-    }
-    stream_extents_valid = true;
-    stream_extents.fill(0);
-    for (size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
-        const auto& attribute = maxwell3d->regs.vertex_attrib_format[i];
-        if (attribute.constant != 0 || attribute.size == VertexAttribute::Size::Invalid) {
-            continue;
-        }
-        const u32 buffer = attribute.buffer.Value();
-        if (buffer >= NUM_VERTEX_BUFFERS) {
-            continue;
-        }
-        const u64 end = static_cast<u64>(attribute.offset.Value()) +
-                        static_cast<u64>(attribute.SizeInBytes());
-        stream_extents[buffer] = (std::max)(stream_extents[buffer], end);
-    }
-    return stream_extents[index];
-}
-
-template <class P>
-u64 BufferCache<P>::DrawVertexBound(u32 index, bool is_indexed) {
-    const auto& array = maxwell3d->regs.vertex_streams[index];
-    if (array.enable == 0) {
-        return 0;
-    }
-    const u64 extent = StreamAttributeExtent(index);
-    if (extent == 0) {
-        return 0;
-    }
-    const u64 stride = static_cast<u64>(array.stride);
-    if (stride == 0) {
-        return extent;
-    }
-    const auto& draw_state = maxwell3d->draw_manager.draw_state;
-    u64 elements = 0;
-    if (maxwell3d->regs.vertex_stream_instances.IsInstancingEnabled(index)) {
-        if (draw_instance_count == 0) {
-            return 0;
-        }
-        const u64 base_instance = static_cast<u64>(draw_state.base_instance);
-        elements = base_instance + 1;
-        if (array.frequency != 0) {
-            elements = (base_instance + static_cast<u64>(draw_instance_count) - 1) /
-                           static_cast<u64>(array.frequency) +
-                       1;
-        }
-    } else if (!is_indexed) {
-        elements = static_cast<u64>(draw_state.vertex_buffer.first) +
-                   static_cast<u64>(draw_state.vertex_buffer.count);
-    } else {
-        const auto format = draw_state.index_buffer.format;
-        u64 max_index = 0xFF;
-        if (format == Maxwell::IndexFormat::UnsignedShort) {
-            max_index = 0xFFFF;
-        } else if (format != Maxwell::IndexFormat::UnsignedByte) {
-            const auto& limit = maxwell3d->regs.vertex_stream_limits[index];
-            const GPUVAddr gpu_addr_begin = array.Address();
-            const GPUVAddr gpu_addr_end = limit.Address() + 1;
-            if (gpu_addr_end <= gpu_addr_begin) {
-                return 0;
-            }
-            const bool walks = gpu_addr_end - gpu_addr_begin >= IMPLAUSIBLE_VERTEX_SIZE ||
-                               !gpu_memory->IsWithinGPUAddressRange(gpu_addr_end);
-            if (!walks) {
-                return 0;
-            }
-            max_index = DrawMaxIndex();
-        }
-        elements = static_cast<u64>(draw_state.base_index) + max_index + 1;
-    }
-    if (elements == 0) {
-        return extent;
-    }
-    return (elements - 1) * stride + extent;
-}
-
-template <class P>
-void BufferCache<P>::UpdateVertexBuffers(bool is_indexed) {
+void BufferCache<P>::UpdateVertexBuffers() {
     auto& flags = maxwell3d->dirty.flags;
-    max_index_scanned = false;
-    stream_extents_valid = false;
-    for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
-        const u64 bound = DrawVertexBound(index, is_indexed);
-        if (bound <= last_draw_bounds[index]) {
-            continue;
-        }
-        flags[Dirty::VertexBuffer0 + index] = true;
-        flags[Dirty::VertexBuffers] = true;
-    }
     if (!maxwell3d->dirty.flags[Dirty::VertexBuffers]) {
         return;
     }
     flags[Dirty::VertexBuffers] = false;
 
     for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
-        UpdateVertexBuffer(index, is_indexed);
+        UpdateVertexBuffer(index);
     }
 }
 
 template <class P>
-void BufferCache<P>::UpdateVertexBuffer(u32 index, bool is_indexed) {
+void BufferCache<P>::UpdateVertexBuffer(u32 index) {
     if (!maxwell3d->dirty.flags[Dirty::VertexBuffer0 + index]) {
         return;
     }
@@ -1386,33 +1373,17 @@ void BufferCache<P>::UpdateVertexBuffer(u32 index, bool is_indexed) {
     const GPUVAddr gpu_addr_begin = array.Address();
     const GPUVAddr gpu_addr_end = limit.Address() + 1;
     const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin);
-    if (array.enable == 0 || !device_addr || gpu_addr_end <= gpu_addr_begin) {
+    const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin);
+    u32 size = address_size; // TODO: Analyze stride and number of vertices
+    if (array.enable == 0 || size == 0 || !device_addr) {
         channel_state->vertex_buffers[index] = NULL_BINDING;
         UpdateVertexBufferSlot(index, NULL_BINDING);
         return;
     }
-    // TODO: Analyze stride and number of vertices
-    constexpr u64 implausible_size = IMPLAUSIBLE_VERTEX_SIZE;
-    u64 address_size = gpu_addr_end - gpu_addr_begin;
-    if (address_size > u64{(std::numeric_limits<u32>::max)()}) {
-        address_size = implausible_size;
+    if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end) || size >= 64_MiB) {
+        size = static_cast<u32>(gpu_memory->MaxContinuousRange(gpu_addr_begin, size));
     }
-    const u64 draw_bound = DrawVertexBound(index, is_indexed);
-    last_draw_bounds[index] = (std::numeric_limits<u64>::max)();
-    if (draw_bound != 0) {
-        last_draw_bounds[index] = draw_bound;
-        address_size = (std::min)(address_size, draw_bound);
-    }
-    if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end) || address_size >= implausible_size) {
-        address_size = gpu_memory->MaxContinuousRange(gpu_addr_begin, address_size);
-    }
-    const u32 size = static_cast<u32>(address_size);
-    if (size == 0) {
-        channel_state->vertex_buffers[index] = NULL_BINDING;
-        UpdateVertexBufferSlot(index, NULL_BINDING);
-        return;
-    }
-    const BufferId buffer_id = FindBuffer(*device_addr, size);
+    const BufferId buffer_id = FindBuffer(*device_addr, size, false);
     const Binding binding{
         .device_addr = *device_addr,
         .size = size,
@@ -1433,7 +1404,7 @@ void BufferCache<P>::UpdateDrawIndirect() {
         binding = Binding{
             .device_addr = *device_addr,
             .size = static_cast<u32>(size),
-            .buffer_id = FindBuffer(*device_addr, static_cast<u32>(size)),
+            .buffer_id = FindBuffer(*device_addr, static_cast<u32>(size), false),
         };
     };
     if (current_draw_indirect->include_count) {
@@ -1457,7 +1428,7 @@ void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
             channel_state->dirty_uniform_buffers[stage] |= 1U << index;
         }
         // Resolve buffer
-        binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
+        binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false);
     });
 }
 
@@ -1466,8 +1437,10 @@ void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
     ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
         // Resolve buffer
         Binding& binding = channel_state->storage_buffers[stage][index];
-        const BufferId buffer_id = FindBuffer(binding.device_addr, binding.size);
+        const BufferId buffer_id = FindBuffer(binding.device_addr, binding.size, false);
         binding.buffer_id = buffer_id;
+        const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
+        ResolveMultiRangeStorage(binding, is_written, graphics_segments);
     });
 }
 
@@ -1475,7 +1448,7 @@ template <class P>
 void BufferCache<P>::UpdateTextureBuffers(size_t stage) {
     ForEachEnabledBit(channel_state->enabled_texture_buffers[stage], [&](u32 index) {
         Binding& binding = channel_state->texture_buffers[stage][index];
-        binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
+        binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false);
     });
 }
 
@@ -1499,7 +1472,7 @@ void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) {
         channel_state->transform_feedback_buffers[index] = NULL_BINDING;
         return;
     }
-    const BufferId buffer_id = FindBuffer(*device_addr, size);
+    const BufferId buffer_id = FindBuffer(*device_addr, size, false);
     channel_state->transform_feedback_buffers[index] = Binding{
         .device_addr = *device_addr,
         .size = size,
@@ -1521,7 +1494,7 @@ void BufferCache<P>::UpdateComputeUniformBuffers() {
                 binding.size = cbuf.size;
             }
         }
-        binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
+        binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false);
     });
 }
 
@@ -1530,7 +1503,10 @@ void BufferCache<P>::UpdateComputeStorageBuffers() {
     ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
         // Resolve buffer
         Binding& binding = channel_state->compute_storage_buffers[index];
-        binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
+        binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false);
+        const bool is_written =
+            ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
+        ResolveMultiRangeStorage(binding, is_written, compute_segments);
     });
 }
 
@@ -1538,7 +1514,7 @@ template <class P>
 void BufferCache<P>::UpdateComputeTextureBuffers() {
     ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) {
         Binding& binding = channel_state->compute_texture_buffers[index];
-        binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
+        binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false);
     });
 }
 
@@ -1554,7 +1530,7 @@ void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u3
 }
 
 template <class P>
-BufferId BufferCache<P>::FindBuffer(DAddr device_addr, u32 size) {
+BufferId BufferCache<P>::FindBuffer(DAddr device_addr, u32 size, bool sparse_compatible) {
     if (device_addr == 0) {
         return NULL_BUFFER_ID;
     }
@@ -1564,10 +1540,18 @@ BufferId BufferCache<P>::FindBuffer(DAddr device_addr, u32 size) {
         Buffer& buffer = slot_buffers[buffer_id];
         WaitForGpuFenceIfNeeded(buffer);
         if (buffer.IsInBounds(device_addr, size)) {
-            return buffer_id;
+            bool usable = true;
+            if constexpr (requires { buffer.IsSparseCompatible(); }) {
+                if (sparse_compatible && !buffer.IsSparseCompatible()) {
+                    usable = false;
+                }
+            }
+            if (usable) {
+                return buffer_id;
+            }
         }
     }
-    return CreateBuffer(device_addr, size);
+    return CreateBuffer(device_addr, size, sparse_compatible);
 }
 
 template <class P>
@@ -1689,14 +1673,15 @@ void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id,
 }
 
 template <class P>
-BufferId BufferCache<P>::CreateBuffer(DAddr device_addr, u32 wanted_size) {
+BufferId BufferCache<P>::CreateBuffer(DAddr device_addr, u32 wanted_size,
+                                      bool sparse_compatible) {
     DAddr device_addr_end = Common::AlignUp(device_addr + wanted_size, CACHING_PAGESIZE);
     device_addr = Common::AlignDown(device_addr, CACHING_PAGESIZE);
-    constexpr u64 max_buffer_size = u64{(std::numeric_limits<u32>::max)()};
-    wanted_size = static_cast<u32>((std::min)(device_addr_end - device_addr, max_buffer_size));
+    wanted_size = static_cast<u32>(device_addr_end - device_addr);
     const OverlapResult overlap = ResolveOverlaps(device_addr, wanted_size);
-    const u32 size = static_cast<u32>((std::min)(overlap.end - overlap.begin, max_buffer_size));
-    const BufferId new_buffer_id = slot_buffers.insert(runtime, overlap.begin, size);
+    const u32 size = static_cast<u32>(overlap.end - overlap.begin);
+    const BufferId new_buffer_id =
+        slot_buffers.insert(runtime, overlap.begin, size, sparse_compatible);
     auto& new_buffer = slot_buffers[new_buffer_id];
     const size_t size_bytes = new_buffer.SizeBytes();
     runtime.ClearBuffer(new_buffer, 0, size_bytes, 0);
@@ -1860,7 +1845,7 @@ void BufferCache<P>::InlineMemoryImplementation(DAddr dest_address, size_t copy_
     ClearDownload(dest_address, copy_size);
     gpu_modified_ranges.Subtract(dest_address, copy_size);
 
-    BufferId buffer_id = FindBuffer(dest_address, static_cast<u32>(copy_size));
+    BufferId buffer_id = FindBuffer(dest_address, static_cast<u32>(copy_size), false);
     auto& buffer = slot_buffers[buffer_id];
     SynchronizeBuffer(buffer, dest_address, static_cast<u32>(copy_size));
 
@@ -1946,6 +1931,9 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, DAddr device_addr, u64
 
 template <class P>
 void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
+    if constexpr (requires { runtime.OnBufferDeleted(slot_buffers[buffer_id]); }) {
+        runtime.OnBufferDeleted(slot_buffers[buffer_id]);
+    }
     bool dirty_index{false};
     boost::container::small_vector<u64, NUM_VERTEX_BUFFERS> dirty_vertex_buffers;
     const auto scalar_replace = [buffer_id](Binding& binding) {
@@ -2049,9 +2037,14 @@ Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
     // The end address used for size calculation does not need to be aligned
     const DAddr cpu_end = Common::AlignUp(*device_addr + size, Core::DEVICE_PAGESIZE);
 
+    u32 binding_size = static_cast<u32>(cpu_end - *aligned_device_addr);
+    if (is_written) {
+        binding_size = aligned_size;
+    }
     const Binding binding{
         .device_addr = *aligned_device_addr,
-        .size = is_written ? aligned_size : static_cast<u32>(cpu_end - *aligned_device_addr),
+        .gpu_addr = aligned_gpu_addr,
+        .size = binding_size,
         .buffer_id = BufferId{},
     };
     return binding;

src/video_core/buffer_cache/buffer_cache_base.h

@@ -29,6 +29,7 @@
 #include "common/settings.h"
 #include "common/slot_vector.h"
 #include "video_core/buffer_cache/buffer_base.h"
+#include "video_core/buffer_cache/virtual_range_cache.h"
 #include "video_core/control/channel_state_cache.h"
 #include "video_core/delayed_destruction_ring.h"
 #include "video_core/dirty_flags.h"
@@ -51,7 +52,6 @@ constexpr u32 NUM_VERTEX_BUFFERS = 16;
 #else
 constexpr u32 NUM_VERTEX_BUFFERS = 32;
 #endif
-constexpr u64 IMPLAUSIBLE_VERTEX_SIZE = 64_MiB;
 constexpr u32 NUM_TRANSFORM_FEEDBACK_BUFFERS = 4;
 constexpr u32 NUM_GRAPHICS_UNIFORM_BUFFERS = 18;
 constexpr u32 NUM_COMPUTE_UNIFORM_BUFFERS = 8;
@@ -82,8 +82,17 @@ static constexpr u32 DEFAULT_SKIP_CACHE_SIZE = static_cast<u32>(4_KiB);
 
 struct Binding {
     DAddr device_addr{};
+    GPUVAddr gpu_addr{};
     u32 size{};
     BufferId buffer_id;
+    u32 segment_first{};
+    u32 segment_count{};
+};
+
+struct MultiRangeSegment {
+    BufferId buffer_id;
+    DAddr device_addr{};
+    u32 size{};
 };
 
 struct TextureBufferBinding : Binding {
@@ -216,8 +225,18 @@ public:
 
     void TickFrame();
 
+    bool BindMultiRangeStorage(const Binding& binding, bool is_written,
+                               std::span<const MultiRangeSegment> pool);
+
+    void ResolveMultiRangeStorage(Binding& binding, bool is_written,
+                                  std::vector<MultiRangeSegment>& pool);
+
+    void UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size);
+
     void WriteMemory(DAddr device_addr, u64 size);
 
+    void CachedWriteMemory(DAddr device_addr, u64 size);
+
     bool OnCPUWrite(DAddr device_addr, u64 size);
 
     void DownloadMemory(DAddr device_addr, u64 size);
@@ -247,7 +266,7 @@ public:
 
     void UnbindGraphicsStorageBuffers(size_t stage);
 
-    void BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
+    bool BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
                                    bool is_written);
 
     void UnbindGraphicsTextureBuffers(size_t stage);
@@ -308,10 +327,6 @@ public:
         current_draw_indirect = current_draw_indirect_;
     }
 
-    void SetDrawInstanceCount(u32 draw_instance_count_) {
-        draw_instance_count = draw_instance_count_;
-    }
-
     [[nodiscard]] std::pair<Buffer*, u32> GetDrawIndirectCount();
 
     [[nodiscard]] std::pair<Buffer*, u32> GetDrawIndirectBuffer();
@@ -379,6 +394,8 @@ private:
 
     void BindHostTransformFeedbackBuffers();
 
+    void BindHostVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride);
+
     void BindHostComputeUniformBuffers();
 
     void BindHostComputeStorageBuffers();
@@ -391,15 +408,9 @@ private:
 
     void UpdateIndexBuffer();
 
-    void UpdateVertexBuffers(bool is_indexed);
+    void UpdateVertexBuffers();
 
-    void UpdateVertexBuffer(u32 index, bool is_indexed);
-
-    [[nodiscard]] u64 DrawVertexBound(u32 index, bool is_indexed);
-
-    [[nodiscard]] u64 DrawMaxIndex();
-
-    [[nodiscard]] u64 StreamAttributeExtent(u32 index);
+    void UpdateVertexBuffer(u32 index);
 
     void UpdateDrawIndirect();
 
@@ -421,7 +432,7 @@ private:
 
     void MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u32 size);
 
-    [[nodiscard]] BufferId FindBuffer(DAddr device_addr, u32 size);
+    [[nodiscard]] BufferId FindBuffer(DAddr device_addr, u32 size, bool sparse_compatible);
 
     void WaitForGpuFenceIfNeeded(Buffer& buffer);
 
@@ -429,7 +440,8 @@ private:
 
     void JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, bool accumulate_stream_score);
 
-    [[nodiscard]] BufferId CreateBuffer(DAddr device_addr, u32 wanted_size);
+    [[nodiscard]] BufferId CreateBuffer(DAddr device_addr, u32 wanted_size,
+                                        bool sparse_compatible);
 
     void Register(BufferId buffer_id);
 
@@ -491,14 +503,6 @@ private:
 
     const Tegra::Engines::Maxwell3D::DrawManager::IndirectParams* current_draw_indirect{};
 
-    u32 draw_instance_count = 0;
-    std::array<u64, NUM_VERTEX_BUFFERS> last_draw_bounds{};
-    Common::ScratchBuffer<u32> index_scan_buffer;
-    u64 cached_max_index = 0;
-    bool max_index_scanned = false;
-    std::array<u64, NUM_VERTEX_BUFFERS> stream_extents{};
-    bool stream_extents_valid = false;
-
     u32 last_index_count = 0;
 
     u32 enabled_vertex_buffers_mask = 0;
@@ -528,6 +532,9 @@ private:
         using TickType = u64;
     };
     Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
+    VirtualRangeCache virtual_ranges;
+    std::vector<MultiRangeSegment> graphics_segments;
+    std::vector<MultiRangeSegment> compute_segments;
     u64 frame_tick = 0;
     u64 total_used_memory = 0;
     u64 minimum_memory = 0;

src/video_core/buffer_cache/virtual_range_cache.h

@@ -0,0 +1,173 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <atomic>
+#include <limits>
+#include <mutex>
+#include <optional>
+#include <vector>
+
+#include <boost/container/small_vector.hpp>
+
+#include "common/common_types.h"
+#include "common/container/unordered_map.h"
+#include "video_core/memory_manager.h"
+
+namespace VideoCommon {
+
+struct VirtualSegment {
+    GPUVAddr gpu_addr;
+    DAddr device_addr;
+    u32 size;
+};
+
+using VirtualSegments = boost::container::small_vector<VirtualSegment, 8>;
+
+class VirtualRangeCache {
+public:
+    static constexpr size_t MAX_ENTRIES = 8192;
+    static constexpr size_t MAX_DEFERRED = 4096;
+
+    const VirtualSegments* Query(Tegra::MemoryManager& memory, GPUVAddr gpu_addr, u32 size) {
+        if (has_deferred.load(std::memory_order_acquire)) {
+            ApplyDeferred();
+        }
+        if (entries.size() > MAX_ENTRIES) {
+            entries.clear();
+        }
+        const size_t as_id = memory.GetID();
+        const u64 key = MakeKey(as_id, gpu_addr);
+        const auto it = entries.find(key);
+        if (it != entries.end() && it->second.as_id == as_id &&
+            it->second.gpu_addr == gpu_addr && it->second.size == size) {
+            return &it->second.segments;
+        }
+        Entry entry{};
+        entry.as_id = as_id;
+        entry.gpu_addr = gpu_addr;
+        entry.size = size;
+        const auto ranges = memory.GetSubmappedRange(gpu_addr, size);
+        GPUVAddr expected = gpu_addr;
+        bool contiguous = true;
+        for (const auto& [range_addr, range_size] : ranges) {
+            if (range_addr != expected || range_size == 0) {
+                contiguous = false;
+                break;
+            }
+            const std::optional<DAddr> device_addr = memory.GpuToCpuAddress(range_addr);
+            if (!device_addr || *device_addr == 0) {
+                contiguous = false;
+                break;
+            }
+            if (range_size > static_cast<size_t>((std::numeric_limits<u32>::max)())) {
+                contiguous = false;
+                break;
+            }
+            entry.segments.push_back(VirtualSegment{
+                .gpu_addr = range_addr,
+                .device_addr = *device_addr,
+                .size = static_cast<u32>(range_size),
+            });
+            expected += range_size;
+        }
+        if (!contiguous || expected != gpu_addr + size) {
+            entry.segments.clear();
+        }
+        const auto result = entries.insert_or_assign(key, std::move(entry));
+        return &result.first->second.segments;
+    }
+
+    void Unmap(size_t as_id, GPUVAddr gpu_addr, u64 size) {
+        if (size == 0) {
+            return;
+        }
+        {
+            std::scoped_lock lock{deferred_mutex};
+            if (!deferred.empty()) {
+                DeferredUnmap& last = deferred.back();
+                if (last.as_id == as_id && last.gpu_addr + last.size == gpu_addr) {
+                    last.size += size;
+                    has_deferred.store(true, std::memory_order_release);
+                    return;
+                }
+            }
+            if (deferred.size() >= MAX_DEFERRED) {
+                deferred.clear();
+                deferred_overflow = true;
+            } else {
+                deferred.push_back(DeferredUnmap{
+                    .as_id = as_id,
+                    .gpu_addr = gpu_addr,
+                    .size = size,
+                });
+            }
+        }
+        has_deferred.store(true, std::memory_order_release);
+    }
+
+private:
+    struct Entry {
+        VirtualSegments segments;
+        size_t as_id{};
+        GPUVAddr gpu_addr{};
+        u32 size{};
+    };
+
+    struct DeferredUnmap {
+        size_t as_id;
+        GPUVAddr gpu_addr;
+        u64 size;
+    };
+
+    static u64 MakeKey(size_t as_id, GPUVAddr gpu_addr) {
+        return (static_cast<u64>(as_id) << 48) ^ gpu_addr;
+    }
+
+    void ApplyDeferred() {
+        std::vector<DeferredUnmap> pending;
+        bool overflow = false;
+        {
+            std::scoped_lock lock{deferred_mutex};
+            has_deferred.store(false, std::memory_order_release);
+            pending.swap(deferred);
+            overflow = deferred_overflow;
+            deferred_overflow = false;
+        }
+        if (overflow) {
+            entries.clear();
+            return;
+        }
+        if (pending.empty() || entries.empty()) {
+            return;
+        }
+        for (auto it = entries.begin(); it != entries.end();) {
+            const Entry& entry = it->second;
+            const GPUVAddr entry_end = entry.gpu_addr + entry.size;
+            bool overlaps = false;
+            for (const DeferredUnmap& unmap : pending) {
+                if (unmap.as_id != entry.as_id) {
+                    continue;
+                }
+                if (entry.gpu_addr < unmap.gpu_addr + unmap.size && unmap.gpu_addr < entry_end) {
+                    overlaps = true;
+                    break;
+                }
+            }
+            if (overlaps) {
+                it = entries.erase(it);
+            } else {
+                ++it;
+            }
+        }
+    }
+
+    ::Common::unordered_map<u64, Entry> entries;
+    std::vector<DeferredUnmap> deferred;
+    std::mutex deferred_mutex;
+    std::atomic<bool> has_deferred{false};
+    bool deferred_overflow{};
+};
+
+} // namespace VideoCommon

src/video_core/host1x/codec_types.h

@@ -8,6 +8,7 @@
 
 #include <array>
 #include <vector>
+#include <type_traits>
 
 #include "common/bit_field.h"
 #include "common/common_funcs.h"

src/video_core/host1x/nvdec_common.h

@@ -1,8 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
+#include <array>
+#include <cstddef>
+#include <type_traits>
+
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"

src/video_core/host1x/vic.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <memory>
+#include <type_traits>
 
 #include "common/common_types.h"
 #include "common/scratch_buffer.h"

src/video_core/host_shaders/CMakeLists.txt

@@ -206,40 +206,6 @@ foreach(VARIANT IN ITEMS ${SHADER_TYPE_VARIANTS})
     set(SHADER_HEADERS ${SHADER_HEADERS} ${VARIANT_HEADER_FILE})
 endforeach()
 
-set(SHADER_DEFINE_VARIANTS
-    "block_linear_unswizzle_2d.comp|nonarrow|HAS_EXTENDED_TYPES=0"
-    "pitch_unswizzle.comp|nonarrow|HAS_EXTENDED_TYPES=0"
-    "block_linear_unswizzle_3d.comp|nonarrow|HAS_EXTENDED_TYPES=0"
-)
-
-foreach(VARIANT IN ITEMS ${SHADER_DEFINE_VARIANTS})
-    string(REPLACE "|" ";" VARIANT_PARTS ${VARIANT})
-    list(GET VARIANT_PARTS 0 VARIANT_FILENAME)
-    list(GET VARIANT_PARTS 1 VARIANT_SUFFIX)
-    list(GET VARIANT_PARTS 2 VARIANT_DEFINE)
-
-    set(VARIANT_SOURCE ${CMAKE_CURRENT_SOURCE_DIR}/${VARIANT_FILENAME})
-    get_filename_component(VARIANT_STEM ${VARIANT_FILENAME} NAME_WE)
-    get_filename_component(VARIANT_EXT ${VARIANT_FILENAME} EXT)
-    string(REPLACE "." "" VARIANT_EXT ${VARIANT_EXT})
-    set(VARIANT_NAME ${VARIANT_STEM}_${VARIANT_SUFFIX}_${VARIANT_EXT})
-
-    string(TOUPPER ${VARIANT_NAME}_SPV VARIANT_VARIABLE_NAME)
-    set(VARIANT_HEADER_FILE ${SHADER_DIR}/${VARIANT_NAME}_spv.h)
-    add_custom_command(
-        OUTPUT
-            ${VARIANT_HEADER_FILE}
-        COMMAND
-            ${GLSLANGVALIDATOR} -V ${QUIET_FLAG} -I"${FIDELITYFX_INCLUDE_DIR}" ${GLSL_FLAGS}
-            -D${VARIANT_DEFINE}
-            --variable-name ${VARIANT_VARIABLE_NAME} -o ${VARIANT_HEADER_FILE} ${VARIANT_SOURCE}
-            --target-env ${SPIR_V_VERSION}
-        MAIN_DEPENDENCY
-            ${VARIANT_SOURCE}
-    )
-    set(SHADER_HEADERS ${SHADER_HEADERS} ${VARIANT_HEADER_FILE})
-endforeach()
-
 foreach(FILEPATH IN ITEMS ${FIDELITYFX_FILES})
     get_filename_component(FILENAME ${FILEPATH} NAME)
     string(REPLACE "." "_" HEADER_NAME ${FILENAME})

src/video_core/host_shaders/block_linear_unswizzle_2d.comp

@@ -5,13 +5,9 @@
 
 #ifdef VULKAN
 
-#ifndef HAS_EXTENDED_TYPES
-#define HAS_EXTENDED_TYPES 1
-#endif
-#if HAS_EXTENDED_TYPES
 #extension GL_EXT_shader_16bit_storage : require
 #extension GL_EXT_shader_8bit_storage : require
-#endif
+#define HAS_EXTENDED_TYPES 1
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)

src/video_core/host_shaders/block_linear_unswizzle_3d.comp

@@ -5,13 +5,9 @@
 
 #ifdef VULKAN
 
-#ifndef HAS_EXTENDED_TYPES
-#define HAS_EXTENDED_TYPES 1
-#endif
-#if HAS_EXTENDED_TYPES
 #extension GL_EXT_shader_16bit_storage : require
 #extension GL_EXT_shader_8bit_storage : require
-#endif
+#define HAS_EXTENDED_TYPES 1
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)

src/video_core/host_shaders/pitch_unswizzle.comp

@@ -5,13 +5,9 @@
 
 #ifdef VULKAN
 
-#ifndef HAS_EXTENDED_TYPES
-#define HAS_EXTENDED_TYPES 1
-#endif
-#if HAS_EXTENDED_TYPES
 #extension GL_EXT_shader_16bit_storage : require
 #extension GL_EXT_shader_8bit_storage : require
-#endif
+#define HAS_EXTENDED_TYPES 1
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -52,7 +52,7 @@ constexpr std::array PROGRAM_LUT{
 Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
     : VideoCommon::BufferBase(null_params) {}
 
-Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_)
+Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_, bool)
     : VideoCommon::BufferBase(cpu_addr_, size_bytes_) {
     buffer.Create();
     if (runtime.device.HasDebuggingToolAttached()) {
@@ -226,6 +226,22 @@ void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) {
     }
 }
 
+void BufferCacheRuntime::BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size,
+                                          u32 stride) {
+    if (index >= max_attributes) {
+        return;
+    }
+    if (has_unified_vertex_buffers) {
+        buffer.MakeResident(GL_READ_ONLY);
+        glBindVertexBuffer(index, 0, 0, static_cast<GLsizei>(stride));
+        glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, index,
+                               buffer.HostGpuAddr() + offset, static_cast<GLsizeiptr>(size));
+    } else {
+        glBindVertexBuffer(index, buffer.Handle(), static_cast<GLintptr>(offset),
+                           static_cast<GLsizei>(stride));
+    }
+}
+
 void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings) {
     // TODO: Should HostBindings provide the correct runtime types to avoid these transforms?
     std::array<GLuint, 32> buffer_handles;

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -23,7 +23,8 @@ class BufferCacheRuntime;
 
 class Buffer : public VideoCommon::BufferBase {
 public:
-    explicit Buffer(BufferCacheRuntime&, DAddr cpu_addr, u64 size_bytes);
+    explicit Buffer(BufferCacheRuntime&, DAddr cpu_addr, u64 size_bytes,
+                    bool sparse_compatible);
     explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams);
 
     void ImmediateUpload(size_t offset, std::span<const u8> data) noexcept;
@@ -99,6 +100,8 @@ public:
 
     void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size);
 
+    void BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride);
+
     void BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings);
 
     void BindUniformBuffer(size_t stage, u32 binding_index, Buffer& buffer, u32 offset, u32 size);

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -259,7 +259,6 @@ void RasterizerOpenGL::PrepareDraw(bool is_indexed, Func&& draw_func) {
 }
 
 void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
-    buffer_cache.SetDrawInstanceCount(instance_count);
     PrepareDraw(is_indexed, [this, is_indexed, instance_count](GLenum primitive_mode) {
         const auto& draw_state = maxwell3d->draw_manager.draw_state;
         const GLuint base_instance = GLuint(draw_state.base_instance);
@@ -305,7 +304,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
 void RasterizerOpenGL::DrawIndirect() {
     const auto& params = maxwell3d->draw_manager.indirect_state;
     buffer_cache.SetDrawIndirect(&params);
-    buffer_cache.SetDrawInstanceCount(0);
     PrepareDraw(params.is_indexed, [this, &params](GLenum primitive_mode) {
         if (params.is_byte_count) {
             const GPUVAddr tfb_object_base_addr = params.indirect_start_address - 4U;

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -56,7 +56,8 @@ size_t BytesPerIndex(VkIndexType index_type) {
     }
 }
 
-vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allocator, u64 size) {
+vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allocator, u64 size,
+                        VkDeviceSize sparse_alignment) {
     VkBufferUsageFlags flags =
         VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
         VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT |
@@ -82,6 +83,9 @@ vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allo
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,
     };
+    if (sparse_alignment > 1) {
+        return memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal, sparse_alignment);
+    }
     return memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal);
 }
 } // Anonymous namespace
@@ -99,10 +103,14 @@ Buffer::Buffer(BufferCacheRuntime& runtime, VideoCommon::NullBufferParams null_p
     }
 }
 
-Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_)
+Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_,
+               bool sparse_compatible_)
     : VideoCommon::BufferBase(cpu_addr_, size_bytes_), device{&runtime.device},
       scheduler{&runtime.scheduler},
-      buffer{CreateBuffer(*device, runtime.memory_allocator, SizeBytes())}, tracker{SizeBytes()} {
+      buffer{CreateBuffer(*device, runtime.memory_allocator, SizeBytes(),
+                          runtime.SparseAlignmentFor(sparse_compatible_))},
+      tracker{SizeBytes()} {
+    sparse_compatible = sparse_compatible_;
     if (runtime.device.HasDebuggingToolAttached()) {
         buffer.SetObjectNameEXT(fmt::format("Buffer {:#x}", CpuAddr()).c_str());
     }
@@ -348,7 +356,8 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& m
     : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
       staging_pool{staging_pool_}, guest_descriptor_queue{guest_descriptor_queue_},
       quad_index_pass(device, scheduler, descriptor_pool, staging_pool,
-                      compute_pass_descriptor_queue) {
+                      compute_pass_descriptor_queue),
+      multi_range_buffers(device_) {
     const VkDriverIdKHR driver_id = device.GetDriverID();
     limit_dynamic_storage_buffers = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY ||
                                     driver_id == VK_DRIVER_ID_ARM_PROPRIETARY;
@@ -536,6 +545,37 @@ void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t si
     });
 }
 
+bool BufferCacheRuntime::BindMultiRangeStorageBuffer(u64 key, bool is_written) {
+    if (multi_range_sources.empty() || multi_range_total == 0) {
+        return false;
+    }
+    const MultiRangeRef ref = multi_range_buffers.Get(device, scheduler, memory_allocator, key,
+                                                     multi_range_sources, multi_range_total);
+    if (ref.handle == VK_NULL_HANDLE) {
+        return false;
+    }
+    if (is_written && !ref.sparse) {
+        return false;
+    }
+    if (ref.needs_gather) {
+        PreCopyBarrier();
+        VkDeviceSize dst_offset = 0;
+        for (const MultiRangeSource& source : multi_range_sources) {
+            const std::array<VideoCommon::BufferCopy, 1> copy{VideoCommon::BufferCopy{
+                .src_offset = u64(source.offset),
+                .dst_offset = u64(dst_offset),
+                .size = size_t(source.size),
+            }};
+            CopyBuffer(ref.handle, source.handle, copy, false);
+            dst_offset += source.size;
+        }
+        PostCopyBarrier();
+        multi_range_buffers.MarkGathered(key);
+    }
+    guest_descriptor_queue.AddBuffer(ref.handle, ref.address, 0, ref.size);
+    return true;
+}
+
 void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format,
                                          u32 base_vertex, u32 num_indices, VkBuffer buffer,
                                          u32 offset, [[maybe_unused]] u32 size) {
@@ -585,6 +625,29 @@ void BufferCacheRuntime::BindQuadIndexBuffer(PrimitiveTopology topology, u32 fir
     }
 }
 
+void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride) {
+    if (index >= device.GetMaxVertexInputBindings()) {
+        return;
+    }
+    if (device.IsExtExtendedDynamicStateSupported()) {
+        scheduler.Record([index, buffer, offset, size, stride](vk::CommandBuffer cmdbuf) {
+            const VkDeviceSize vk_offset = buffer != VK_NULL_HANDLE ? offset : 0;
+            const VkDeviceSize vk_size = buffer != VK_NULL_HANDLE ? size : VK_WHOLE_SIZE;
+            const VkDeviceSize vk_stride = stride;
+            cmdbuf.BindVertexBuffers2EXT(index, 1, &buffer, &vk_offset, &vk_size, &vk_stride);
+        });
+    } else {
+        if (!device.HasNullDescriptor() && buffer == VK_NULL_HANDLE) {
+            ReserveNullBuffer();
+            buffer = *null_buffer;
+            offset = 0;
+        }
+        scheduler.Record([index, buffer, offset](vk::CommandBuffer cmdbuf) {
+            cmdbuf.BindVertexBuffer(index, buffer, offset);
+        });
+    }
+}
+
 void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings) {
     boost::container::static_vector<VkBuffer, VideoCommon::NUM_VERTEX_BUFFERS> buffer_handles(bindings.buffers.size());
     for (u32 i = 0; i < bindings.buffers.size(); ++i) {

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -8,11 +8,14 @@
 
 #include <limits>
 
+#include <boost/container/small_vector.hpp>
+
 #include "video_core/buffer_cache/buffer_cache_base.h"
 #include "video_core/buffer_cache/memory_tracker_base.h"
 #include "video_core/buffer_cache/usage_tracker.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
+#include "video_core/renderer_vulkan/vk_multi_range_buffer.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/surface.h"
@@ -31,7 +34,8 @@ class BufferCacheRuntime;
 class Buffer : public VideoCommon::BufferBase {
 public:
     explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params);
-    explicit Buffer(BufferCacheRuntime& runtime, VAddr cpu_addr_, u64 size_bytes_);
+    explicit Buffer(BufferCacheRuntime& runtime, VAddr cpu_addr_, u64 size_bytes_,
+                    bool sparse_compatible_);
 
     [[nodiscard]] VkBufferView View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format);
 
@@ -43,6 +47,14 @@ public:
         return device_address;
     }
 
+    [[nodiscard]] bool IsSparseCompatible() const noexcept {
+        return sparse_compatible;
+    }
+
+    [[nodiscard]] vk::MemoryLocation Location() const noexcept {
+        return buffer.Location();
+    }
+
     [[nodiscard]] bool IsRegionUsed(u64 offset, u64 size) const noexcept {
         return tracker.IsUsed(offset, size);
     }
@@ -77,6 +89,7 @@ private:
     VkDeviceAddress device_address{};
     u64 last_usage_tick{};
     bool is_null{};
+    bool sparse_compatible{};
 };
 
 class QuadArrayIndexBuffer;
@@ -125,7 +138,7 @@ public:
 
     void PreCopyBarrier();
 
-    void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer,
+    void CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
                     std::span<const VideoCommon::BufferCopy> copies, bool barrier,
                     bool can_reorder_upload = false);
 
@@ -138,6 +151,8 @@ public:
 
     void BindQuadIndexBuffer(PrimitiveTopology topology, u32 first, u32 count);
 
+    void BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride);
+
     void BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings);
 
     void BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size);
@@ -153,6 +168,46 @@ public:
         return ref.mapped_span;
     }
 
+    [[nodiscard]] VkDeviceSize SparseAlignmentFor(bool sparse_compatible) const noexcept {
+        if (!sparse_compatible || !multi_range_buffers.use_sparse) {
+            return 0;
+        }
+        return multi_range_buffers.block_size;
+    }
+
+    [[nodiscard]] bool PrefersSparseSources() const noexcept {
+        return multi_range_buffers.use_sparse;
+    }
+
+    void ResetMultiRange() noexcept {
+        multi_range_sources.clear();
+        multi_range_total = 0;
+    }
+
+    void PushMultiRangeSource(const Buffer& buffer, u32 offset, u32 size) {
+        const vk::MemoryLocation location = buffer.Location();
+        multi_range_sources.push_back(MultiRangeSource{
+            .handle = buffer.Handle(),
+            .memory = location.memory,
+            .memory_offset = location.offset,
+            .offset = offset,
+            .size = size,
+            .write_tick = buffer.getWriteTick(),
+            .memory_type = location.memory_type,
+        });
+        multi_range_total += size;
+    }
+
+    bool BindMultiRangeStorageBuffer(u64 key, bool is_written);
+
+    void InvalidateMultiRange(u64 key) {
+        multi_range_buffers.Invalidate(key);
+    }
+
+    void OnBufferDeleted(const Buffer& buffer) {
+        multi_range_buffers.DropOwner(scheduler, buffer.Handle());
+    }
+
     void BindUniformBuffer(const Buffer& buffer, u32 offset, u32 size) {
         BindBuffer(buffer, offset, size);
     }
@@ -206,6 +261,10 @@ private:
     std::unique_ptr<Uint8Pass> uint8_pass;
     QuadIndexedPass quad_index_pass;
 
+    MultiRangeBufferCache multi_range_buffers;
+    boost::container::small_vector<MultiRangeSource, 16> multi_range_sources;
+    VkDeviceSize multi_range_total{};
+
     bool limit_dynamic_storage_buffers = false;
     u32 max_dynamic_storage_buffers = (std::numeric_limits<u32>::max)();
 };

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -22,13 +22,7 @@
 #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_comp_spv.h"
-#include "video_core/host_shaders/block_linear_unswizzle_2d_nonarrow_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_comp_spv.h"
-#include "video_core/host_shaders/block_linear_unswizzle_3d_nonarrow_comp_spv.h"
-#include "video_core/host_shaders/pitch_unswizzle_comp_spv.h"
-#include "video_core/host_shaders/pitch_unswizzle_nonarrow_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"
@@ -878,291 +872,4 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
     });
 }
 
-namespace {
-
-constexpr u32 UNSWIZZLE_BINDING_INPUT_BUFFER = 0;
-constexpr u32 UNSWIZZLE_BINDING_OUTPUT_IMAGE = 1;
-constexpr size_t UNSWIZZLE_NUM_BINDINGS = 2;
-
-constexpr std::array<VkDescriptorSetLayoutBinding, UNSWIZZLE_NUM_BINDINGS>
-    UNSWIZZLE_DESCRIPTOR_SET_BINDINGS{{
-        {
-            .binding = UNSWIZZLE_BINDING_INPUT_BUFFER,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .descriptorCount = 1,
-            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-            .pImmutableSamplers = nullptr,
-        },
-        {
-            .binding = UNSWIZZLE_BINDING_OUTPUT_IMAGE,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
-            .descriptorCount = 1,
-            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-            .pImmutableSamplers = nullptr,
-        },
-    }};
-
-constexpr std::array<VkDescriptorUpdateTemplateEntry, UNSWIZZLE_NUM_BINDINGS>
-    UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE{{
-        {
-            .dstBinding = UNSWIZZLE_BINDING_INPUT_BUFFER,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = UNSWIZZLE_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
-        {
-            .dstBinding = UNSWIZZLE_BINDING_OUTPUT_IMAGE,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
-            .offset = UNSWIZZLE_BINDING_OUTPUT_IMAGE * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
-    }};
-
-constexpr DescriptorBankInfo UNSWIZZLE_BANK_INFO{
-    .uniform_buffers = 0,
-    .storage_buffers = 1,
-    .texture_buffers = 0,
-    .image_buffers = 0,
-    .textures = 0,
-    .images = 1,
-    .score = 2,
-};
-
-[[nodiscard]] std::span<const u32> UnswizzleSpv(const Device& device,
-                                                std::span<const u32> extended,
-                                                std::span<const u32> narrow) {
-    if (device.IsStorageBuffer8BitAccessSupported() &&
-        device.IsStorageBuffer16BitAccessSupported()) {
-        return extended;
-    }
-    return narrow;
-}
-
-struct PitchUnswizzlePushConstants {
-    alignas(8) std::array<u32, 2> origin;
-    alignas(8) std::array<s32, 2> destination;
-    u32 bytes_per_block;
-    u32 pitch;
-};
-
-void RecordUnswizzleEntryBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, VkImage vk_image,
-                                 VkImageAspectFlags aspect_mask, bool is_initialized) {
-    scheduler.Record([vk_pipeline, vk_image, aspect_mask,
-                      is_initialized](vk::CommandBuffer cmdbuf) {
-        VkAccessFlags src_access = VK_ACCESS_NONE;
-        VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED;
-        if (is_initialized) {
-            src_access = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT |
-                         VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
-            old_layout = VK_IMAGE_LAYOUT_GENERAL;
-        }
-        const VkImageMemoryBarrier image_barrier{
-            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = src_access,
-            .dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
-            .oldLayout = old_layout,
-            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
-            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .image = vk_image,
-            .subresourceRange{
-                .aspectMask = aspect_mask,
-                .baseMipLevel = 0,
-                .levelCount = VK_REMAINING_MIP_LEVELS,
-                .baseArrayLayer = 0,
-                .layerCount = VK_REMAINING_ARRAY_LAYERS,
-            },
-        };
-        VkPipelineStageFlags src_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
-        if (is_initialized) {
-            src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER;
-        }
-        cmdbuf.PipelineBarrier(src_stage, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
-        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
-    });
-}
-
-void RecordUnswizzleExitBarrier(Scheduler& scheduler, VkImage vk_image,
-                                VkImageAspectFlags aspect_mask) {
-    scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) {
-        const VkImageMemoryBarrier image_barrier{
-            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT |
-                             VK_ACCESS_COLOR_ATTACHMENT_READ_BIT,
-            .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
-            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
-            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .image = vk_image,
-            .subresourceRange{
-                .aspectMask = aspect_mask,
-                .baseMipLevel = 0,
-                .levelCount = VK_REMAINING_MIP_LEVELS,
-                .baseArrayLayer = 0,
-                .layerCount = VK_REMAINING_ARRAY_LAYERS,
-            },
-        };
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                               vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier);
-    });
-}
-
-} // Anonymous namespace
-
-BlockLinearUnswizzle2DPass::BlockLinearUnswizzle2DPass(
-    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
-    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
-    : ComputePass(device_, scheduler_, descriptor_pool_, UNSWIZZLE_DESCRIPTOR_SET_BINDINGS,
-                  UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE, UNSWIZZLE_BANK_INFO,
-                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(
-                      VideoCommon::Accelerated::BlockLinearSwizzle2DParams)>,
-                  UnswizzleSpv(device_, BLOCK_LINEAR_UNSWIZZLE_2D_COMP_SPV,
-                               BLOCK_LINEAR_UNSWIZZLE_2D_NONARROW_COMP_SPV)),
-      scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
-
-BlockLinearUnswizzle2DPass::~BlockLinearUnswizzle2DPass() = default;
-
-void BlockLinearUnswizzle2DPass::Unswizzle(
-    Image& image, const StagingBufferRef& map,
-    std::span<const VideoCommon::SwizzleParameters> swizzles) {
-    using namespace VideoCommon::Accelerated;
-    scheduler.RequestOutsideRenderPassOperationContext();
-    const VkPipeline vk_pipeline = *pipeline;
-    const VkImageAspectFlags aspect_mask = image.AspectMask();
-    const VkImage vk_image = image.Handle();
-    const bool is_initialized = image.ExchangeInitialization();
-    RecordUnswizzleEntryBarrier(scheduler, vk_pipeline, vk_image, aspect_mask, is_initialized);
-
-    const u32 num_layers = static_cast<u32>(image.info.resources.layers);
-    for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
-        const size_t input_offset = swizzle.buffer_offset + map.offset;
-        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
-        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
-
-        compute_pass_descriptor_queue.Acquire(scheduler, 2);
-        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
-                                                image.guest_size_bytes - swizzle.buffer_offset);
-        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
-        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
-
-        const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
-        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_layers, params,
-                          descriptor_data](vk::CommandBuffer cmdbuf) {
-            const VkDescriptorSet set = descriptor_allocator.Commit();
-            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
-            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
-            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
-            cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_layers);
-        });
-    }
-    RecordUnswizzleExitBarrier(scheduler, vk_image, aspect_mask);
-}
-
-BlockLinearUnswizzleImage3DPass::BlockLinearUnswizzleImage3DPass(
-    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
-    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
-    : ComputePass(device_, scheduler_, descriptor_pool_, UNSWIZZLE_DESCRIPTOR_SET_BINDINGS,
-                  UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE, UNSWIZZLE_BANK_INFO,
-                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearSwizzle3DParams)>,
-                  UnswizzleSpv(device_, BLOCK_LINEAR_UNSWIZZLE_3D_COMP_SPV,
-                               BLOCK_LINEAR_UNSWIZZLE_3D_NONARROW_COMP_SPV)),
-      scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
-
-BlockLinearUnswizzleImage3DPass::~BlockLinearUnswizzleImage3DPass() = default;
-
-void BlockLinearUnswizzleImage3DPass::Unswizzle(
-    Image& image, const StagingBufferRef& map,
-    std::span<const VideoCommon::SwizzleParameters> swizzles) {
-    using namespace VideoCommon::Accelerated;
-    scheduler.RequestOutsideRenderPassOperationContext();
-    const VkPipeline vk_pipeline = *pipeline;
-    const VkImageAspectFlags aspect_mask = image.AspectMask();
-    const VkImage vk_image = image.Handle();
-    const bool is_initialized = image.ExchangeInitialization();
-    RecordUnswizzleEntryBarrier(scheduler, vk_pipeline, vk_image, aspect_mask, is_initialized);
-
-    for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
-        const size_t input_offset = swizzle.buffer_offset + map.offset;
-        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 16U);
-        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
-        const u32 num_dispatches_z = Common::DivCeil(swizzle.num_tiles.depth, 8U);
-
-        compute_pass_descriptor_queue.Acquire(scheduler, 2);
-        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
-                                                image.guest_size_bytes - swizzle.buffer_offset);
-        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
-        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
-
-        const auto params = MakeBlockLinearSwizzle3DParams(swizzle, image.info);
-        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params,
-                          descriptor_data](vk::CommandBuffer cmdbuf) {
-            const VkDescriptorSet set = descriptor_allocator.Commit();
-            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
-            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
-            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
-            cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z);
-        });
-    }
-    RecordUnswizzleExitBarrier(scheduler, vk_image, aspect_mask);
-}
-
-PitchUnswizzlePass::PitchUnswizzlePass(
-    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
-    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
-    : ComputePass(device_, scheduler_, descriptor_pool_, UNSWIZZLE_DESCRIPTOR_SET_BINDINGS,
-                  UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE, UNSWIZZLE_BANK_INFO,
-                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(PitchUnswizzlePushConstants)>,
-                  UnswizzleSpv(device_, PITCH_UNSWIZZLE_COMP_SPV,
-                               PITCH_UNSWIZZLE_NONARROW_COMP_SPV)),
-      scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
-
-PitchUnswizzlePass::~PitchUnswizzlePass() = default;
-
-void PitchUnswizzlePass::Unswizzle(Image& image, const StagingBufferRef& map,
-                                   std::span<const VideoCommon::SwizzleParameters> swizzles) {
-    scheduler.RequestOutsideRenderPassOperationContext();
-    const VkPipeline vk_pipeline = *pipeline;
-    const VkImageAspectFlags aspect_mask = image.AspectMask();
-    const VkImage vk_image = image.Handle();
-    const bool is_initialized = image.ExchangeInitialization();
-    RecordUnswizzleEntryBarrier(scheduler, vk_pipeline, vk_image, aspect_mask, is_initialized);
-
-    const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(image.info.format);
-    const u32 pitch = image.info.pitch;
-    for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
-        const size_t input_offset = swizzle.buffer_offset + map.offset;
-        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
-        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
-
-        compute_pass_descriptor_queue.Acquire(scheduler, 2);
-        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
-                                                image.guest_size_bytes - swizzle.buffer_offset);
-        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
-        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
-
-        const PitchUnswizzlePushConstants params{
-            .origin{0, 0},
-            .destination{0, 0},
-            .bytes_per_block = bytes_per_block,
-            .pitch = pitch,
-        };
-        scheduler.Record([this, num_dispatches_x, num_dispatches_y, params,
-                          descriptor_data](vk::CommandBuffer cmdbuf) {
-            const VkDescriptorSet set = descriptor_allocator.Commit();
-            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
-            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
-            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
-            cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1);
-        });
-    }
-    RecordUnswizzleExitBarrier(scheduler, vk_image, aspect_mask);
-}
-
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -164,49 +164,4 @@ private:
     ComputePassDescriptorQueue& compute_pass_descriptor_queue;
 };
 
-class BlockLinearUnswizzle2DPass final : public ComputePass {
-public:
-    explicit BlockLinearUnswizzle2DPass(
-        const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
-        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
-    ~BlockLinearUnswizzle2DPass();
-
-    void Unswizzle(Image& image, const StagingBufferRef& map,
-                   std::span<const VideoCommon::SwizzleParameters> swizzles);
-
-private:
-    Scheduler& scheduler;
-    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
-};
-
-class BlockLinearUnswizzleImage3DPass final : public ComputePass {
-public:
-    explicit BlockLinearUnswizzleImage3DPass(
-        const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
-        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
-    ~BlockLinearUnswizzleImage3DPass();
-
-    void Unswizzle(Image& image, const StagingBufferRef& map,
-                   std::span<const VideoCommon::SwizzleParameters> swizzles);
-
-private:
-    Scheduler& scheduler;
-    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
-};
-
-class PitchUnswizzlePass final : public ComputePass {
-public:
-    explicit PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_,
-                                DescriptorPool& descriptor_pool_,
-                                ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
-    ~PitchUnswizzlePass();
-
-    void Unswizzle(Image& image, const StagingBufferRef& map,
-                   std::span<const VideoCommon::SwizzleParameters> swizzles);
-
-private:
-    Scheduler& scheduler;
-    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
-};
-
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -694,11 +694,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
         const size_t num_vertex_arrays = (std::min)(
             Maxwell::NumVertexArrays, static_cast<size_t>(device.GetMaxVertexInputBindings()));
         for (size_t index = 0; index < num_vertex_arrays; ++index) {
-            const bool instanced = ((key.state.enabled_divisors >> index) & 1) != 0;
-            auto rate = VK_VERTEX_INPUT_RATE_VERTEX;
-            if (instanced) {
-                rate = VK_VERTEX_INPUT_RATE_INSTANCE;
-            }
+            const bool instanced = key.state.binding_divisors[index] != 0;
+            const auto rate =
+                instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX;
             vertex_bindings.push_back({
                 .binding = static_cast<u32>(index),
                 .stride = key.state.vertex_strides[index],
@@ -707,7 +705,7 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
             if (instanced) {
                 vertex_binding_divisors.push_back({
                     .binding = static_cast<u32>(index),
-                    .divisor = device.GetVertexAttribDivisor(key.state.binding_divisors[index]),
+                    .divisor = key.state.binding_divisors[index],
                 });
             }
         }

src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp

@@ -0,0 +1,338 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include <algorithm>
+#include <mutex>
+#include <utility>
+
+#include "video_core/renderer_vulkan/vk_multi_range_buffer.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+
+namespace Vulkan {
+
+MultiRangeBufferCache::MultiRangeBufferCache(const Device& device) {
+    sparse_usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
+                   VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
+    if (device.IsBufferDeviceAddressSupported()) {
+        sparse_usage |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
+    }
+    if (!device.IsSparseBindingSupported()) {
+        return;
+    }
+    u32 memory_type_bits = 0;
+    const VkDeviceSize queried = QueryBlockSize(device, memory_type_bits);
+    if (queried == 0 || memory_type_bits == 0) {
+        return;
+    }
+    block_size = queried;
+    sparse_memory_type_bits = memory_type_bits;
+    use_sparse = true;
+}
+
+VkDeviceSize MultiRangeBufferCache::QueryBlockSize(const Device& device,
+                                                   u32& memory_type_bits) const {
+    const VkDevice logical = *device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    const VkBufferCreateInfo probe_ci{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = VK_BUFFER_CREATE_SPARSE_BINDING_BIT | VK_BUFFER_CREATE_SPARSE_ALIASED_BIT,
+        .size = DEFAULT_BLOCK_SIZE,
+        .usage = sparse_usage,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    };
+    VkBuffer probe{};
+    if (dld.vkCreateBuffer(logical, &probe_ci, nullptr, &probe) != VK_SUCCESS) {
+        return 0;
+    }
+    const SparseBuffer owned{probe, logical, dld};
+    const VkBufferMemoryRequirementsInfo2 reqs_info{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2,
+        .pNext = nullptr,
+        .buffer = probe,
+    };
+    VkMemoryRequirements2 reqs2{
+        .sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2,
+        .pNext = nullptr,
+        .memoryRequirements = {},
+    };
+    dld.vkGetBufferMemoryRequirements2(logical, &reqs_info, &reqs2);
+    memory_type_bits = reqs2.memoryRequirements.memoryTypeBits;
+    return reqs2.memoryRequirements.alignment;
+}
+
+u64 MultiRangeBufferCache::HashSources(std::span<const MultiRangeSource> sources) const {
+    u64 hash = 0xcbf29ce484222325ULL;
+    const auto mix = [&hash](u64 value) {
+        hash ^= value;
+        hash *= 0x100000001b3ULL;
+    };
+    for (const MultiRangeSource& source : sources) {
+        mix(u64(source.handle));
+        mix(u64(source.offset));
+        mix(u64(source.size));
+    }
+    return hash;
+}
+
+u64 MultiRangeBufferCache::HashContent(std::span<const MultiRangeSource> sources) const {
+    u64 hash = 0xcbf29ce484222325ULL;
+    for (const MultiRangeSource& source : sources) {
+        hash ^= source.write_tick;
+        hash *= 0x100000001b3ULL;
+    }
+    return hash;
+}
+
+bool MultiRangeBufferCache::CanBindSparse(std::span<const MultiRangeSource> sources) const {
+    return use_sparse &&
+           std::none_of(sources.begin(), sources.end(),
+                        [block = block_size, bits = sparse_memory_type_bits](auto const& e) {
+                            const VkDeviceSize memory_offset = e.memory_offset + e.offset;
+                            return e.memory == VK_NULL_HANDLE || e.memory_type >= 32 ||
+                                   ((bits >> e.memory_type) & 1) == 0 ||
+                                   (memory_offset % block) != 0 || (e.size % block) != 0;
+                        });
+}
+
+SparseBuffer MultiRangeBufferCache::CreateSparse(const Device& device, Scheduler& scheduler,
+                                                 std::span<const MultiRangeSource> sources,
+                                                 VkDeviceSize total) {
+    const VkDevice logical = *device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    const VkBufferCreateInfo buffer_ci{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = VK_BUFFER_CREATE_SPARSE_BINDING_BIT | VK_BUFFER_CREATE_SPARSE_ALIASED_BIT,
+        .size = total,
+        .usage = sparse_usage,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    };
+    VkBuffer raw{};
+    if (dld.vkCreateBuffer(logical, &buffer_ci, nullptr, &raw) != VK_SUCCESS) {
+        return SparseBuffer{};
+    }
+    SparseBuffer handle{raw, logical, dld};
+    std::vector<VkSparseMemoryBind> binds;
+    binds.reserve(sources.size());
+    VkDeviceSize resource_offset = 0;
+    for (const MultiRangeSource& source : sources) {
+        binds.push_back(VkSparseMemoryBind{
+            .resourceOffset = resource_offset,
+            .size = source.size,
+            .memory = source.memory,
+            .memoryOffset = source.memory_offset + source.offset,
+            .flags = 0,
+        });
+        resource_offset += source.size;
+    }
+    const VkSparseBufferMemoryBindInfo buffer_bind{
+        .buffer = raw,
+        .bindCount = static_cast<u32>(binds.size()),
+        .pBinds = binds.data(),
+    };
+    const VkBindSparseInfo bind_info{
+        .sType = VK_STRUCTURE_TYPE_BIND_SPARSE_INFO,
+        .pNext = nullptr,
+        .waitSemaphoreCount = 0,
+        .pWaitSemaphores = nullptr,
+        .bufferBindCount = 1,
+        .pBufferBinds = &buffer_bind,
+        .imageOpaqueBindCount = 0,
+        .pImageOpaqueBinds = nullptr,
+        .imageBindCount = 0,
+        .pImageBinds = nullptr,
+        .signalSemaphoreCount = 0,
+        .pSignalSemaphores = nullptr,
+    };
+    const VkFenceCreateInfo fence_ci{
+        .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+    };
+    vk::Fence fence = device.GetLogical().CreateFence(fence_ci);
+    VkResult bind_result = VK_ERROR_UNKNOWN;
+    {
+        std::scoped_lock lock{scheduler.submit_mutex};
+        bind_result = device.GetGraphicsQueue().BindSparse(bind_info, *fence);
+    }
+    if (bind_result != VK_SUCCESS) {
+        return SparseBuffer{};
+    }
+    fence.Wait();
+    return handle;
+}
+
+void MultiRangeBufferCache::RetireEntry(Scheduler& scheduler, Entry& entry) {
+    if (!entry.sparse_handle && !entry.gathered) {
+        return;
+    }
+    if (retired.size() == retired.capacity()) {
+        DrainRetired(scheduler);
+    }
+    if (retired.size() == retired.capacity()) {
+        u64 oldest = retired.front().tick;
+        for (const Retired& item : retired) {
+            if (item.tick < oldest) {
+                oldest = item.tick;
+            }
+        }
+        scheduler.Wait(oldest);
+        DrainRetired(scheduler);
+    }
+    retired.push_back(Retired{
+        .handle = std::move(entry.sparse_handle),
+        .gathered = std::move(entry.gathered),
+        .tick = scheduler.CurrentTick(),
+    });
+}
+
+void MultiRangeBufferCache::DrainRetired(Scheduler& scheduler) {
+    size_t index = 0;
+    while (index < retired.size()) {
+        if (scheduler.IsFree(retired[index].tick)) {
+            if (index + 1 != retired.size()) {
+                retired[index] = std::move(retired.back());
+            }
+            retired.pop_back();
+        } else {
+            ++index;
+        }
+    }
+}
+
+MultiRangeRef MultiRangeBufferCache::Get(const Device& device, Scheduler& scheduler,
+                                         MemoryAllocator& memory_allocator, u64 key,
+                                         std::span<const MultiRangeSource> sources,
+                                         VkDeviceSize total) {
+    if (sources.empty() || total == 0) {
+        return MultiRangeRef{};
+    }
+    if (!retired.empty()) {
+        DrainRetired(scheduler);
+    }
+    const u64 geometry = HashSources(sources);
+    const u64 content = HashContent(sources);
+    const auto it = entries.find(key);
+    if (it != entries.end() && it->second.geometry == geometry && it->second.size == total) {
+        Entry& entry = it->second;
+        if (entry.content != content) {
+            entry.content = content;
+            entry.dirty = true;
+        }
+        MultiRangeRef ref{
+            .handle = *entry.sparse_handle,
+            .address = entry.address,
+            .size = entry.size,
+            .sparse = true,
+            .needs_gather = false,
+        };
+        if (!entry.sparse_handle) {
+            ref.handle = *entry.gathered;
+            ref.sparse = false;
+            ref.needs_gather = entry.dirty;
+        }
+        return ref;
+    }
+    if (it != entries.end()) {
+        RetireEntry(scheduler, it->second);
+        entries.erase(it);
+    }
+
+    Entry entry{};
+    entry.geometry = geometry;
+    entry.content = content;
+    entry.size = total;
+    if (CanBindSparse(sources)) {
+        entry.sparse_handle = CreateSparse(device, scheduler, sources, total);
+        if (entry.sparse_handle) {
+            entry.owners.reserve(sources.size());
+            for (const MultiRangeSource& source : sources) {
+                entry.owners.push_back(source.handle);
+            }
+        }
+    }
+    if (!entry.sparse_handle) {
+        VkBufferUsageFlags flags = VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
+                                   VK_BUFFER_USAGE_TRANSFER_DST_BIT |
+                                   VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
+        if (device.IsBufferDeviceAddressSupported()) {
+            flags |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
+        }
+        const VkBufferCreateInfo gather_ci{
+            .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .size = total,
+            .usage = flags,
+            .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+            .queueFamilyIndexCount = 0,
+            .pQueueFamilyIndices = nullptr,
+        };
+        entry.gathered = memory_allocator.CreateBuffer(gather_ci, MemoryUsage::DeviceLocal);
+        entry.dirty = true;
+    }
+    if (device.IsBufferDeviceAddressSupported()) {
+        VkBuffer address_handle = *entry.sparse_handle;
+        if (!entry.sparse_handle) {
+            address_handle = *entry.gathered;
+        }
+        entry.address = device.GetLogical().GetBufferDeviceAddress(address_handle);
+    }
+
+    MultiRangeRef ref{
+        .handle = *entry.sparse_handle,
+        .address = entry.address,
+        .size = entry.size,
+        .sparse = true,
+        .needs_gather = false,
+    };
+    if (!entry.sparse_handle) {
+        ref.handle = *entry.gathered;
+        ref.sparse = false;
+        ref.needs_gather = true;
+    }
+    entries.emplace(key, std::move(entry));
+    return ref;
+}
+
+void MultiRangeBufferCache::MarkGathered(u64 key) {
+    if (auto const it = entries.find(key); it != entries.end()) {
+        it->second.dirty = false;
+    }
+}
+
+void MultiRangeBufferCache::DropOwner(Scheduler& scheduler, VkBuffer owner) {
+    if (owner == VK_NULL_HANDLE) {
+        return;
+    }
+    for (auto it = entries.begin(); it != entries.end();) {
+        Entry& entry = it->second;
+        bool owned = false;
+        for (const VkBuffer handle : entry.owners) {
+            if (handle == owner) {
+                owned = true;
+                break;
+            }
+        }
+        if (!owned) {
+            ++it;
+            continue;
+        }
+        RetireEntry(scheduler, entry);
+        it = entries.erase(it);
+    }
+}
+
+void MultiRangeBufferCache::Invalidate(u64 key) {
+    if (auto const it = entries.find(key); it != entries.end()) {
+        it->second.dirty = true;
+    }
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_multi_range_buffer.h

@@ -0,0 +1,105 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <span>
+#include <vector>
+
+#include <boost/container/static_vector.hpp>
+
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "common/container/unordered_map.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
+#include "video_core/vulkan_common/vulkan_wrapper.h"
+
+namespace Vulkan {
+
+using SparseBuffer = vk::Handle<VkBuffer, VkDevice, vk::DeviceDispatch>;
+
+class Device;
+class Scheduler;
+
+struct MultiRangeSource {
+    VkBuffer handle{};
+    VkDeviceMemory memory{};
+    VkDeviceSize memory_offset{};
+    VkDeviceSize offset{};
+    VkDeviceSize size{};
+    u64 write_tick{};
+    u32 memory_type{};
+};
+
+struct MultiRangeRef {
+    VkBuffer handle{};
+    VkDeviceAddress address{};
+    VkDeviceSize size{};
+    bool sparse{};
+    bool needs_gather{};
+};
+
+class MultiRangeBufferCache final {
+public:
+    static constexpr VkDeviceSize DEFAULT_BLOCK_SIZE = 64 * 1024;
+    static constexpr size_t MAX_RETIRED = 256;
+
+    explicit MultiRangeBufferCache(const Device& device);
+
+    YUZU_NON_COPYABLE(MultiRangeBufferCache);
+
+    [[nodiscard]] MultiRangeRef Get(const Device& device, Scheduler& scheduler,
+                                    MemoryAllocator& memory_allocator, u64 key,
+                                    std::span<const MultiRangeSource> sources,
+                                    VkDeviceSize total);
+
+    void MarkGathered(u64 key);
+
+    void Invalidate(u64 key);
+
+    void DropOwner(Scheduler& scheduler, VkBuffer owner);
+
+    VkDeviceSize block_size{DEFAULT_BLOCK_SIZE};
+    bool use_sparse{};
+
+private:
+    struct Retired {
+        SparseBuffer handle;
+        vk::Buffer gathered;
+        u64 tick{};
+    };
+
+    struct Entry {
+        vk::Buffer gathered;
+        SparseBuffer sparse_handle;
+        std::vector<VkBuffer> owners;
+        VkDeviceAddress address{};
+        VkDeviceSize size{};
+        u64 geometry{};
+        u64 content{};
+        bool dirty{true};
+    };
+
+    [[nodiscard]] u64 HashSources(std::span<const MultiRangeSource> sources) const;
+
+    [[nodiscard]] u64 HashContent(std::span<const MultiRangeSource> sources) const;
+
+    [[nodiscard]] bool CanBindSparse(std::span<const MultiRangeSource> sources) const;
+
+    [[nodiscard]] SparseBuffer CreateSparse(const Device& device, Scheduler& scheduler,
+                                            std::span<const MultiRangeSource> sources,
+                                            VkDeviceSize total);
+
+    [[nodiscard]] VkDeviceSize QueryBlockSize(const Device& device, u32& memory_type_bits) const;
+
+    void RetireEntry(Scheduler& scheduler, Entry& entry);
+
+    void DrainRetired(Scheduler& scheduler);
+
+    ::Common::unordered_map<u64, Entry> entries;
+    boost::container::static_vector<Retired, MAX_RETIRED> retired;
+    u32 sparse_memory_type_bits{};
+    VkBufferUsageFlags sparse_usage{};
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_present_manager.cpp

@@ -338,7 +338,7 @@ void PresentManager::PresentThread(std::stop_token token) {
             // By exchanging the lock ownership we take the swapchain lock
             // before the queue lock goes out of scope. This way the swapchain
             // lock in WaitPresent is guaranteed to occur after here.
-            std::exchange(lock, std::unique_lock{swapchain_mutex});
+            void(std::exchange(lock, std::unique_lock{swapchain_mutex}));
             CopyToSwapchain(frame);
 
             // Free the frame for reuse

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -137,12 +137,10 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u3
     max_y = (std::max)(max_y, 0);
 
     if (src.enable) {
-        const s32 min_x = static_cast<s32>(src.min_x.Value());
-        const s32 max_x = static_cast<s32>(src.max_x.Value());
-        scissor.offset.x = scale_up(min_x);
+        scissor.offset.x = scale_up(src.min_x);
         scissor.offset.y = scale_up(min_y);
-        scissor.extent.width = scale_up((std::max)(max_x - min_x, 0));
-        scissor.extent.height = scale_up((std::max)(max_y - min_y, 0));
+        scissor.extent.width = scale_up(src.max_x - src.min_x);
+        scissor.extent.height = scale_up(max_y - min_y);
     } else {
         scissor.offset.x = 0;
         scissor.offset.y = 0;
@@ -262,7 +260,6 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
 }
 
 void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
-    buffer_cache.SetDrawInstanceCount(instance_count);
     PrepareDraw(is_indexed, [this, is_indexed, instance_count] {
         const auto& draw_state = maxwell3d->draw_manager.draw_state;
         const u32 num_instances{instance_count};
@@ -298,7 +295,6 @@ void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
 void RasterizerVulkan::DrawIndirect() {
     const auto& params = maxwell3d->draw_manager.indirect_state;
     buffer_cache.SetDrawIndirect(&params);
-    buffer_cache.SetDrawInstanceCount(0);
     PrepareDraw(params.is_indexed, [this, &params] {
         const auto indirect_buffer = buffer_cache.GetDrawIndirectBuffer();
         const auto& buffer = indirect_buffer.first;
@@ -823,6 +819,7 @@ void RasterizerVulkan::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) {
         std::scoped_lock lock{texture_cache.mutex};
         texture_cache.UnmapGPUMemory(as_id, addr, size);
     }
+    buffer_cache.UnmapGPUMemory(as_id, addr, size);
 }
 
 void RasterizerVulkan::SignalFence(std::function<void()>&& func) {
@@ -1921,19 +1918,13 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs)
     for (u32 binding = 0; binding < max_bindings; ++binding) {
         const auto& input_binding{regs.vertex_streams[binding]};
         const bool is_instanced{regs.vertex_stream_instances.IsInstancingEnabled(binding)};
-        auto input_rate = VK_VERTEX_INPUT_RATE_VERTEX;
-        u32 divisor = 1;
-        if (is_instanced) {
-            input_rate = VK_VERTEX_INPUT_RATE_INSTANCE;
-            divisor = device.GetVertexAttribDivisor(input_binding.frequency);
-        }
         bindings.push_back({
             .sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_BINDING_DESCRIPTION_2_EXT,
             .pNext = nullptr,
             .binding = binding,
             .stride = input_binding.stride,
-            .inputRate = input_rate,
-            .divisor = divisor,
+            .inputRate = is_instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX,
+            .divisor = is_instanced ? input_binding.frequency : 1,
         });
     }
 

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -300,7 +300,7 @@ void Scheduler::WorkerThread(std::stop_token stop_token) {
             // Exchange lock ownership so that we take the execution lock before
             // the queue lock goes out of scope. This allows us to force execution
             // to complete in the next step.
-            std::exchange(lk, std::unique_lock{execution_mutex});
+            void(std::exchange(lk, std::unique_lock{execution_mutex}));
 
             // Perform the work, tracking whether the chunk was a submission
             // before executing.
@@ -449,9 +449,7 @@ void Scheduler::EndRenderPass()
                                          | 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
-                                         | VK_ACCESS_TRANSFER_READ_BIT
-                                         | VK_ACCESS_TRANSFER_WRITE_BIT,
+                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
                         .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                         .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -462,7 +460,7 @@ 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_TRANSFER,
+                                   VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
                                    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

@@ -160,55 +160,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
           info.size.depth == 1;
 }
 
-[[nodiscard]] PixelFormat UnswizzleViewFormat(u32 bytes_per_block) {
-    switch (bytes_per_block) {
-    case 1:
-        return PixelFormat::R8_UINT;
-    case 2:
-        return PixelFormat::R16_UINT;
-    case 4:
-        return PixelFormat::R32_UINT;
-    case 8:
-        return PixelFormat::R32G32_UINT;
-    case 16:
-        return PixelFormat::R32G32B32A32_UINT;
-    default:
-        return PixelFormat::Invalid;
-    }
-}
-
-constexpr u32 UNSWIZZLE_WORKGROUP_INVOCATIONS = 32 * 32;
-
-[[nodiscard]] bool SupportsAcceleratedUnswizzleDevice(const Device& device) {
-    return device.IsKhrImageFormatListSupported() &&
-           device.GetMaxComputeWorkGroupInvocations() >= UNSWIZZLE_WORKGROUP_INVOCATIONS;
-}
-
-[[nodiscard]] bool SupportsAcceleratedUnswizzle(const Device& device, const ImageInfo& info) {
-    if (!SupportsAcceleratedUnswizzleDevice(device)) {
-        return false;
-    }
-    if (info.num_samples > 1) {
-        return false;
-    }
-    if (info.type != ImageType::e2D && info.type != ImageType::e3D &&
-        info.type != ImageType::Linear) {
-        return false;
-    }
-    const PixelFormat view_format =
-        UnswizzleViewFormat(VideoCore::Surface::BytesPerBlock(info.format));
-    if (view_format == PixelFormat::Invalid) {
-        return false;
-    }
-    if (!VideoCore::Surface::IsViewCompatible(info.format, view_format, false, true)) {
-        return false;
-    }
-    const auto host_format =
-        MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, view_format);
-    return device.IsFormatSupported(host_format.format, VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT,
-                                    FormatType::Optimal);
-}
-
 [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info,
                                                     std::optional<VkFormat> format_override = {}) {
     auto format_info =
@@ -297,18 +248,8 @@ constexpr u32 UNSWIZZLE_WORKGROUP_INVOCATIONS = 32 * 32;
     return allocator.CreateImage(image_ci);
 }
 
-[[nodiscard]] VkImageViewType StorageViewType(ImageType type) {
-    if (type == ImageType::e3D) {
-        return VK_IMAGE_VIEW_TYPE_3D;
-    }
-    if (type == ImageType::Linear) {
-        return VK_IMAGE_VIEW_TYPE_2D;
-    }
-    return VK_IMAGE_VIEW_TYPE_2D_ARRAY;
-}
-
 [[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image,
-                                            VkFormat format, VkImageViewType view_type) {
+                                            VkFormat format) {
     static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
         .pNext = nullptr,
@@ -319,7 +260,7 @@ constexpr u32 UNSWIZZLE_WORKGROUP_INVOCATIONS = 32 * 32;
         .pNext = &storage_image_view_usage_create_info,
         .flags = 0,
         .image = image,
-        .viewType = view_type,
+        .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
         .format = format,
         .components{
             .r = VK_COMPONENT_SWIZZLE_IDENTITY,
@@ -717,11 +658,18 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
             .subresourceRange = subresource_range,
     };
 
-    cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
+    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,
                            read_barrier);
     cmdbuf.CopyBufferToImage(src_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copies);
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0,
-                           nullptr, nullptr, write_barrier);
+    // 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);
 }
 
 [[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region,
@@ -1020,14 +968,6 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
         bl3d_unswizzle_pass.emplace(device, scheduler, descriptor_pool,
                                    staging_buffer_pool, compute_pass_descriptor_queue);
     }
-    if (SupportsAcceleratedUnswizzleDevice(device)) {
-        bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool,
-                                     compute_pass_descriptor_queue);
-        bl_unswizzle_image_3d_pass.emplace(device, scheduler, descriptor_pool,
-                                           compute_pass_descriptor_queue);
-        pitch_unswizzle_pass.emplace(device, scheduler, descriptor_pool,
-                                     compute_pass_descriptor_queue);
-    }
 }
 
 void TextureCacheRuntime::Finish() {
@@ -1962,12 +1902,16 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
     if (runtime->device.HasDebuggingToolAttached()) {
         original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
     }
-    if (False(flags & VideoCommon::ImageFlagBits::Converted) &&
-        SupportsAcceleratedUnswizzle(runtime->device, info)) {
-        flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
-    }
     current_image = &Image::original_image;
     storage_image_views.resize(info.resources.levels);
+    if (WillUseAcceleratedAstcDecode(runtime->device, info)) {
+        const auto& device = runtime->device.GetLogical();
+        const VkFormat storage_format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
+        for (s32 level = 0; level < info.resources.levels; ++level) {
+            storage_image_views[level] =
+                MakeStorageView(device, level, *original_image, storage_format);
+        }
+    }
 }
 
 Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {}
@@ -2091,7 +2035,7 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
                                                 temp_vk_image, info.format, info.num_samples,
                                                 {image_copies.data(), image_copies.size()}, false);
         }
-        InitializationFor(current_image) = true;
+        initialized = true;
         runtime->ReleaseMsaaScratchImage(temp_vk_image);
 
         if (is_rescaled) {
@@ -2112,7 +2056,7 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
     const VkBuffer src_buffer = buffer;
     const VkImage vk_image = *original_image;
     const VkImageAspectFlags vk_aspect_mask = aspect_mask;
-    const bool was_initialized = std::exchange(InitializationFor(&Image::original_image), true);
+    const bool was_initialized = std::exchange(initialized, true);
 
     scheduler->Record([src_buffer, vk_image, vk_aspect_mask, was_initialized,
                        vk_copies](vk::CommandBuffer cmdbuf) {
@@ -2377,46 +2321,16 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm
     DownloadMemory(buffers, offsets, copies);
 }
 
-std::vector<vk::ImageView>& Image::StorageViewsFor(vk::Image Image::*image) {
-    if (image == &Image::scaled_image) {
-        if (scaled_storage_image_views.empty()) {
-            scaled_storage_image_views.resize(info.resources.levels);
-        }
-        return scaled_storage_image_views;
-    }
-    return storage_image_views;
-}
-
-bool& Image::InitializationFor(vk::Image Image::*image) noexcept {
-    if (image == &Image::scaled_image) {
-        return scaled_initialized;
-    }
-    return original_initialized;
-}
-
 VkImageView Image::StorageImageView(s32 level) noexcept {
-    const bool astc_decode = WillUseAcceleratedAstcDecode(runtime->device, info);
-    const bool unswizzle_upload =
-        !astc_decode && True(flags & ImageFlagBits::AcceleratedUpload);
-    vk::Image Image::*target = current_image;
-    if (astc_decode || unswizzle_upload) {
-        target = &Image::original_image;
-    }
-    auto& view = StorageViewsFor(target)[level];
+    auto& view = storage_image_views[level];
     if (!view) {
         auto format_info =
             MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format);
-        if (astc_decode) {
+        if (WillUseAcceleratedAstcDecode(runtime->device, info)) {
             format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
         }
-        if (unswizzle_upload) {
-            const PixelFormat view_format =
-                UnswizzleViewFormat(VideoCore::Surface::BytesPerBlock(info.format));
-            format_info = MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, false,
-                                                     view_format);
-        }
-        view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*target),
-                               format_info.format, StorageViewType(info.type));
+        view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*current_image),
+                               format_info.format);
     }
     return *view;
 }
@@ -2456,7 +2370,6 @@ bool Image::ScaleUp(bool ignore) {
     }
     if (NeedsScaleHelper()) {
         if (!BlitScaleHelper(true)) {
-            flags &= ~ImageFlagBits::Rescaled;
             current_image = &Image::original_image;
             return false;
         }
@@ -2646,10 +2559,6 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
     if (device->IsExtAstcDecodeModeSupported() && IsLdrAstcFormat(format_info.format)) {
         view_next = &astc_decode_mode;
     }
-    auto subresource_range = MakeSubresourceRange(aspect_mask, info.range);
-    if (True(flags & VideoCommon::ImageViewFlagBits::Slice)) {
-        subresource_range.levelCount = 1;
-    }
     const VkImageViewCreateInfo create_info{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
         .pNext = view_next,
@@ -2658,7 +2567,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
         .viewType = VkImageViewType{},
         .format = format_info.format,
         .components = swizzle_mapping,
-        .subresourceRange = subresource_range,
+        .subresourceRange = MakeSubresourceRange(aspect_mask, info.range),
     };
     const auto create = [&](TextureType tex_type, std::optional<u32> num_layers) {
         VkImageViewCreateInfo ci{create_info};
@@ -3232,19 +3141,6 @@ void TextureCacheRuntime::AccelerateImageUpload(
         return astc_decoder_pass->Assemble(image, map, swizzles);
     }
 
-    if (bl_unswizzle_2d_pass && image.info.type == ImageType::e2D) {
-        return bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles);
-    }
-
-    if (bl_unswizzle_image_3d_pass && image.info.type == ImageType::e3D &&
-        !IsPixelFormatBCn(image.info.format)) {
-        return bl_unswizzle_image_3d_pass->Unswizzle(image, map, swizzles);
-    }
-
-    if (pitch_unswizzle_pass && image.info.type == ImageType::Linear) {
-        return pitch_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

@@ -159,9 +159,6 @@ public:
     std::optional<ASTCDecoderPass> astc_decoder_pass;
 
     std::optional<BlockLinearUnswizzle3DPass> bl3d_unswizzle_pass;
-    std::optional<BlockLinearUnswizzle2DPass> bl_unswizzle_2d_pass;
-    std::optional<BlockLinearUnswizzleImage3DPass> bl_unswizzle_image_3d_pass;
-    std::optional<PitchUnswizzlePass> pitch_unswizzle_pass;
     const Settings::ResolutionScalingInfo& resolution;
     std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;
 
@@ -353,7 +350,7 @@ public:
 
     /// Returns true when the image is already initialized and mark it as initialized
     [[nodiscard]] bool ExchangeInitialization() noexcept {
-        return std::exchange(InitializationFor(current_image), true);
+        return std::exchange(initialized, true);
     }
 
     VkImageView StorageImageView(s32 level) noexcept;
@@ -373,10 +370,6 @@ private:
 
     bool NeedsScaleHelper() const;
 
-    std::vector<vk::ImageView>& StorageViewsFor(vk::Image Image::*image);
-
-    bool& InitializationFor(vk::Image Image::*image) noexcept;
-
     Scheduler* scheduler{};
     TextureCacheRuntime* runtime{};
 
@@ -394,10 +387,8 @@ private:
     vk::Image Image::*current_image{};
 
     std::vector<vk::ImageView> storage_image_views;
-    std::vector<vk::ImageView> scaled_storage_image_views;
     VkImageAspectFlags aspect_mask = 0;
-    bool original_initialized = false;
-    bool scaled_initialized = false;
+    bool initialized = false;
 
     std::optional<Framebuffer> scale_framebuffer;
     std::optional<Framebuffer> normal_framebuffer;

src/video_core/texture_cache/accelerated_swizzle.h

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -26,8 +23,8 @@ struct BlockLinearSwizzle2DParams {
 };
 
 struct BlockLinearSwizzle3DParams {
-    alignas(16) std::array<u32, 3> origin;
-    alignas(16) std::array<s32, 3> destination;
+    std::array<u32, 3> origin;
+    std::array<s32, 3> destination;
     u32 bytes_per_block_log2;
     u32 slice_size;
     u32 block_size;

src/video_core/texture_cache/texture_cache.h

@@ -20,7 +20,6 @@
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/guest_memory.h"
 #include "video_core/host1x/gpu_device_memory_manager.h"
-#include "video_core/texture_cache/accelerated_swizzle.h"
 #include "video_core/texture_cache/image_view_base.h"
 #include "video_core/texture_cache/samples_helper.h"
 #include "video_core/texture_cache/texture_cache_base.h"
@@ -280,11 +279,11 @@ void TextureCache<P>::CheckFeedbackLoop(std::span<const ImageViewInOut> views) {
 
             const ImageId view_image_id = slot_image_views[view.id].image_id;
             {
-                bool is_feedback = false;
+                bool is_continue = false;
                 for (size_t i = 0; i < 8; ++i)
-                    is_feedback |= (rt_active_mask & (1u << i)) && view_image_id == rt_image_id[i];
-                if (is_feedback)
-                    return true;
+                    is_continue |= (rt_active_mask & (1u << i)) && view_image_id == rt_image_id[i];
+                if (is_continue)
+                    continue;
             }
             if (depth_active && view_image_id == rt_depth_image_id) {
                 return true;
@@ -627,25 +626,14 @@ void TextureCache<P>::DownloadMemory(DAddr cpu_addr, size_t size) {
     std::ranges::sort(images, [this](ImageId lhs, ImageId rhs) {
         return slot_images[lhs].modification_tick < slot_images[rhs].modification_tick;
     });
-    size_t total_size_bytes = 0;
-    for (const ImageId image_id : images) {
-        total_size_bytes += slot_images[image_id].unswizzled_size_bytes;
-    }
-    auto download_map = runtime.DownloadStagingBuffer(total_size_bytes);
     for (const ImageId image_id : images) {
         Image& image = slot_images[image_id];
+        auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes);
         const auto copies = FixSmallVectorADL(FullDownloadCopies(image.info));
-        image.DownloadMemory(download_map, copies);
-        download_map.offset += image.unswizzled_size_bytes;
-    }
-    runtime.Finish();
-    std::span<u8> download_span = download_map.mapped_span;
-    for (const ImageId image_id : images) {
-        const ImageBase& image = slot_images[image_id];
-        const auto copies = FixSmallVectorADL(FullDownloadCopies(image.info));
-        SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span,
+        image.DownloadMemory(map, copies);
+        runtime.Finish();
+        SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span,
                      swizzle_data_buffer);
-        download_span = download_span.subspan(image.unswizzled_size_bytes);
     }
 }
 
@@ -1134,12 +1122,6 @@ void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {
 
     TrackImage(image, image_id);
 
-    if (image.info.rescaleable &&
-        IsRegionGpuModified(image.cpu_addr, image.guest_size_bytes)) {
-        runtime.TransitionImageLayout(image);
-        return;
-    }
-
     if (image.info.num_samples > 1 && !runtime.CanUploadMSAA()) {
         LOG_WARNING(HW_GPU, "MSAA image uploads are not implemented");
         runtime.TransitionImageLayout(image);
@@ -1175,7 +1157,8 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
     const GPUVAddr gpu_addr = image.gpu_addr;
 
     if (True(image.flags & ImageFlagBits::AcceleratedUpload)) {
-        gpu_memory->ReadBlockUnsafe(gpu_addr, mapped_span.data(), image.guest_size_bytes);
+        gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
+                              VideoCommon::CacheType::NoTextureCache);
         const auto uploads = FullUploadSwizzles(image.info);
         runtime.AccelerateImageUpload(image, staging, FixSmallVectorADL(uploads), 0, 0);
         return;
@@ -1282,9 +1265,6 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
 
 template <class P>
 bool TextureCache<P>::ImageCanRescale(ImageBase& image) {
-    if (!Settings::values.resolution_info.active) {
-        return false;
-    }
     if (!image.info.rescaleable) {
         return false;
     }
@@ -1372,12 +1352,12 @@ void TextureCache<P>::QueueAsyncDecode(Image& image, ImageId image_id) {
     decode->image_id = image_id;
     async_decodes.push_back(std::move(decode));
 
-    Common::ScratchBuffer<u8> local_unswizzle_data_buffer(image.unswizzled_size_bytes);
+    std::vector<u8> local_unswizzle_data_buffer(image.unswizzled_size_bytes, 0);
     Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::UnsafeRead> swizzle_data(*gpu_memory, image.gpu_addr, image.guest_size_bytes, &swizzle_data_buffer);
     auto copies = UnswizzleImage(*gpu_memory, image.gpu_addr, image.info, swizzle_data, local_unswizzle_data_buffer);
     const size_t out_size = MapSizeBytes(image);
 
-    auto func = [out_size, copies = std::move(copies), info = image.info,
+    auto func = [out_size, copies, info = image.info,
                  input = std::move(local_unswizzle_data_buffer),
                  async_decode = decode_ptr]() mutable {
         async_decode->decoded_data.resize_destructive(out_size);
@@ -1446,16 +1426,18 @@ void TextureCache<P>::TickAsyncUnswizzle() {
     Image& image = slot_images[task.image_id];
 
     if (!task.initialized) {
-        task.total_size = image.guest_size_bytes;
+        task.total_size = MapSizeBytes(image);
         task.staging_buffer = runtime.UploadStagingBuffer(task.total_size, true);
 
-        const auto layout = FullUploadSwizzles(task.info);
-        const auto params =
-            VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams(layout.front(), task.info);
-        task.bytes_per_slice = params.slice_size;
-        task.chunked = task.info.block.depth == 0;
+        const auto& info = image.info;
+        const u32 bytes_per_block = BytesPerBlock(info.format);
+        const u32 width_blocks = Common::DivCeil(info.size.width, 4u);
+        const u32 height_blocks = Common::DivCeil(info.size.height, 4u);
+
+        const u32 stride = width_blocks * bytes_per_block;
+        const u32 aligned_height = height_blocks;
+        task.bytes_per_slice = static_cast<size_t>(stride) * aligned_height;
         task.last_submitted_offset = 0;
-        task.slices_submitted = 0;
         task.initialized = true;
     }
 
@@ -1470,39 +1452,31 @@ void TextureCache<P>::TickAsyncUnswizzle() {
             if (copy_amount == 0) copy_amount = task.bytes_per_slice;
         }
 
-        gpu_memory->ReadBlockUnsafe(image.gpu_addr + task.current_offset,
-                                    task.staging_buffer.mapped_span.data() + task.current_offset,
-                                    copy_amount);
+        gpu_memory->ReadBlock(image.gpu_addr + task.current_offset,
+                              task.staging_buffer.mapped_span.data() + task.current_offset,
+                              copy_amount);
         task.current_offset += copy_amount;
     }
 
     const bool is_final_batch = task.current_offset >= task.total_size;
+    const size_t bytes_ready = task.current_offset - task.last_submitted_offset;
+    const u32 complete_slices = static_cast<u32>(bytes_ready / task.bytes_per_slice);
 
-    if (task.chunked) {
-        const size_t bytes_ready = task.current_offset - task.last_submitted_offset;
-        const u32 complete_slices = static_cast<u32>(bytes_ready / task.bytes_per_slice);
+    if (complete_slices >= swizzle_slices_per_batch || (is_final_batch && complete_slices > 0)) {
+        const u32 z_start = static_cast<u32>(task.last_submitted_offset / task.bytes_per_slice);
+        const u32 slices_to_process = (std::min)(complete_slices, swizzle_slices_per_batch);
+        const u32 z_count = (std::min)(slices_to_process, image.info.size.depth - z_start);
 
-        if (complete_slices >= swizzle_slices_per_batch || (is_final_batch && complete_slices > 0)) {
-            const u32 z_start = task.slices_submitted;
-            const u32 slices_to_process = (std::min)(complete_slices, swizzle_slices_per_batch);
-            const u32 z_count = (std::min)(slices_to_process, image.info.size.depth - z_start);
-
-            if (z_count > 0) {
-                const auto uploads = FullUploadSwizzles(task.info);
-                runtime.AccelerateImageUpload(image, task.staging_buffer,
-                                              FixSmallVectorADL(uploads), z_start, z_count);
-                task.last_submitted_offset += static_cast<size_t>(z_count) * task.bytes_per_slice;
-                task.slices_submitted += z_count;
-            }
+        if (z_count > 0) {
+            const auto uploads = FullUploadSwizzles(task.info);
+            runtime.AccelerateImageUpload(image, task.staging_buffer, FixSmallVectorADL(uploads), z_start, z_count);
+            task.last_submitted_offset += (static_cast<size_t>(z_count) * task.bytes_per_slice);
         }
-    } else if (is_final_batch && task.slices_submitted == 0) {
-        const auto uploads = FullUploadSwizzles(task.info);
-        runtime.AccelerateImageUpload(image, task.staging_buffer, FixSmallVectorADL(uploads), 0,
-                                      image.info.size.depth);
-        task.slices_submitted = image.info.size.depth;
     }
 
-    const bool all_slices_submitted = task.slices_submitted >= image.info.size.depth;
+    // Check if complete
+    const u32 slices_submitted = static_cast<u32>(task.last_submitted_offset / task.bytes_per_slice);
+    const bool all_slices_submitted = slices_submitted >= image.info.size.depth;
 
     if (is_final_batch && all_slices_submitted) {
         runtime.FreeDeferredStagingBuffer(task.staging_buffer);

src/video_core/texture_cache/texture_cache_base.h

@@ -139,8 +139,6 @@ class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelI
         AsyncBuffer staging_buffer;
         size_t last_submitted_offset = 0;
         size_t bytes_per_slice;
-        u32 slices_submitted = 0;
-        bool chunked = false;
         bool initialized = false;
     };
 

src/video_core/vulkan_common/vulkan_device.cpp

@@ -1219,11 +1219,6 @@ bool Device::GetSuitability(bool requires_swapchain) {
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT;
         SetNext(next, properties.transform_feedback);
     }
-    if (extensions.vertex_attribute_divisor) {
-        properties.vertex_attribute_divisor.sType =
-            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT;
-        SetNext(next, properties.vertex_attribute_divisor);
-    }
     if (extensions.maintenance5) {
         properties.maintenance5.sType =
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_5_PROPERTIES_KHR;
@@ -1575,6 +1570,8 @@ void Device::SetupFamilies(VkSurfaceKHR surface) {
     }
     if (graphics) {
         graphics_family = *graphics;
+        graphics_family_sparse_binding =
+            (queue_family_properties[*graphics].queueFlags & VK_QUEUE_SPARSE_BINDING_BIT) != 0;
     }
     if (present) {
         present_family = *present;

src/video_core/vulkan_common/vulkan_device.h

@@ -70,7 +70,6 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE(EXT, ProvokingVertex, PROVOKING_VERTEX, provoking_vertex)                              \
     FEATURE(EXT, Robustness2, ROBUSTNESS_2, robustness2)                                           \
     FEATURE(EXT, TransformFeedback, TRANSFORM_FEEDBACK, transform_feedback)                        \
-    FEATURE(EXT, VertexAttributeDivisor, VERTEX_ATTRIBUTE_DIVISOR, vertex_attribute_divisor)       \
     FEATURE(EXT, VertexInputDynamicState, VERTEX_INPUT_DYNAMIC_STATE, vertex_input_dynamic_state)  \
     FEATURE(KHR, Maintenance5, MAINTENANCE_5, maintenance5)                                        \
     FEATURE(KHR, Maintenance6, MAINTENANCE_6, maintenance6)                                        \
@@ -93,6 +92,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
     EXTENSION(EXT, SHADER_STENCIL_EXPORT, shader_stencil_export)                                   \
     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)                                       \
@@ -317,6 +317,10 @@ public:
         return properties.driver.driverID;
     }
 
+    bool IsSparseBindingSupported() const {
+        return features.features.sparseBinding && graphics_family_sparse_binding;
+    }
+
     /// Returns true for tile-based deferred renderers.
     bool IsTiler() const {
         switch (GetDriverID()) {
@@ -359,7 +363,6 @@ public:
 
 #define FN_MAX_LIMIT_LIST \
     FN_MAX_LIMIT_ELEM(ComputeSharedMemorySize) \
-    FN_MAX_LIMIT_ELEM(ComputeWorkGroupInvocations) \
     FN_MAX_LIMIT_ELEM(PerStageDescriptorSampledImages) \
     FN_MAX_LIMIT_ELEM(PerStageResources) \
     FN_MAX_LIMIT_ELEM(DescriptorSetSamplers) \
@@ -690,32 +693,6 @@ FN_MAX_LIMIT_LIST
         return features.host_query_reset.hostQueryReset != VK_FALSE;
     }
 
-    u32 GetMaxVertexAttribDivisor() const {
-        const u32 reported = properties.vertex_attribute_divisor.maxVertexAttribDivisor;
-        if (reported == 0) {
-            return 1;
-        }
-        return reported;
-    }
-
-    bool IsVertexAttributeInstanceRateZeroDivisorSupported() const {
-        return features.vertex_attribute_divisor.vertexAttributeInstanceRateZeroDivisor == VK_TRUE;
-    }
-
-    u32 GetVertexAttribDivisor(u32 frequency) const {
-        const u32 max_divisor = GetMaxVertexAttribDivisor();
-        if (frequency == 0) {
-            if (IsVertexAttributeInstanceRateZeroDivisorSupported()) {
-                return 0;
-            }
-            return max_divisor;
-        }
-        if (frequency > max_divisor) {
-            return max_divisor;
-        }
-        return frequency;
-    }
-
     /// Returns true if the device supports VK_EXT_transform_feedback.
     bool IsExtTransformFeedbackSupported() const {
         return extensions.transform_feedback;
@@ -1174,6 +1151,7 @@ private:
     u32 instance_version{};      ///< Vulkan instance version.
     u32 graphics_family{};       ///< Main graphics queue family index.
     u32 present_family{};        ///< Main present queue family index.
+    bool graphics_family_sparse_binding{};
 
     struct Extensions {
 #define EXTENSION(prefix, macro_name, var_name) bool var_name{};
@@ -1216,7 +1194,6 @@ private:
         VkPhysicalDeviceDescriptorBufferPropertiesEXT descriptor_buffer{};
         VkPhysicalDeviceSubgroupSizeControlProperties subgroup_size_control{};
         VkPhysicalDeviceTransformFeedbackPropertiesEXT transform_feedback{};
-        VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT vertex_attribute_divisor{};
         VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{};
         VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{};
         VkPhysicalDeviceCustomBorderColorPropertiesEXT custom_border_color{};

src/video_core/vulkan_common/vulkan_memory_allocator.cpp

@@ -26,350 +26,383 @@
 #include "common/settings.h"
 
 namespace Vulkan {
-    namespace {
+namespace {
 
 // Helpers translating MemoryUsage to flags/usage
 
-        [[maybe_unused]] VkMemoryPropertyFlags MemoryUsagePropertyFlags(MemoryUsage usage) {
-            switch (usage) {
-                case MemoryUsage::DeviceLocal:
-                    return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
-                case MemoryUsage::Upload:
-                    return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
-                case MemoryUsage::Download:
-                    return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
-                           VK_MEMORY_PROPERTY_HOST_CACHED_BIT;
-                case MemoryUsage::Stream:
-                    return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
-                           VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
-            }
-            ASSERT_MSG(false, "Invalid memory usage={}", usage);
-            return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+    [[maybe_unused]] VkMemoryPropertyFlags MemoryUsagePropertyFlags(MemoryUsage usage) {
+        switch (usage) {
+            case MemoryUsage::DeviceLocal:
+                return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+            case MemoryUsage::Upload:
+                return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                        VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+            case MemoryUsage::Download:
+                return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                        VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
+                        VK_MEMORY_PROPERTY_HOST_CACHED_BIT;
+            case MemoryUsage::Stream:
+                return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
+                        VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                        VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
         }
+        ASSERT_MSG(false, "Invalid memory usage={}", usage);
+        return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+    }
 
-        [[nodiscard]] VkMemoryPropertyFlags MemoryUsagePreferredVmaFlags(MemoryUsage usage) {
-            if (usage == MemoryUsage::Download) {
-                return VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
-            }
-            return usage != MemoryUsage::DeviceLocal ? VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
-                                                     : VkMemoryPropertyFlagBits{};
+    [[nodiscard]] VkMemoryPropertyFlags MemoryUsagePreferredVmaFlags(MemoryUsage usage) {
+        if (usage == MemoryUsage::Download) {
+            return VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
         }
+        return usage != MemoryUsage::DeviceLocal ? VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
+                                                    : VkMemoryPropertyFlagBits{};
+    }
 
-        [[nodiscard]] VmaAllocationCreateFlags MemoryUsageVmaFlags(MemoryUsage usage) {
-            switch (usage) {
-                case MemoryUsage::Upload:
-                case MemoryUsage::Stream:
-                    return VMA_ALLOCATION_CREATE_MAPPED_BIT |
-                           VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
-                case MemoryUsage::Download:
-                    return VMA_ALLOCATION_CREATE_MAPPED_BIT |
-                           VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT;
-                case MemoryUsage::DeviceLocal:
-                    return {};
-            }
-            return {};
+    [[nodiscard]] VmaAllocationCreateFlags MemoryUsageVmaFlags(MemoryUsage usage) {
+        switch (usage) {
+            case MemoryUsage::Upload:
+            case MemoryUsage::Stream:
+                return VMA_ALLOCATION_CREATE_MAPPED_BIT |
+                        VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
+            case MemoryUsage::Download:
+                return VMA_ALLOCATION_CREATE_MAPPED_BIT |
+                        VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT;
+            case MemoryUsage::DeviceLocal:
+                return {};
         }
+        return {};
+    }
 
-        [[nodiscard]] VmaMemoryUsage MemoryUsageVma(MemoryUsage usage) {
-            switch (usage) {
-                case MemoryUsage::DeviceLocal:
-                case MemoryUsage::Stream:
-                    return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
-                case MemoryUsage::Upload:
-                case MemoryUsage::Download:
-                    return VMA_MEMORY_USAGE_AUTO_PREFER_HOST;
-            }
-            return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
+    [[nodiscard]] VmaMemoryUsage MemoryUsageVma(MemoryUsage usage) {
+        switch (usage) {
+            case MemoryUsage::DeviceLocal:
+            case MemoryUsage::Stream:
+                return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
+            case MemoryUsage::Upload:
+            case MemoryUsage::Download:
+                return VMA_MEMORY_USAGE_AUTO_PREFER_HOST;
         }
-
-
-// This avoids calling vkGetBufferMemoryRequirements* directly.
-        template<typename T>
-        static VkBuffer GetVkHandleFromBuffer(const T &buf) {
-            if constexpr (requires { static_cast<VkBuffer>(buf); }) {
-                return static_cast<VkBuffer>(buf);
-            } else if constexpr (requires {{ buf.GetHandle() } -> std::convertible_to<VkBuffer>; }) {
-                return buf.GetHandle();
-            } else if constexpr (requires {{ buf.Handle() } -> std::convertible_to<VkBuffer>; }) {
-                return buf.Handle();
-            } else if constexpr (requires {{ buf.vk_handle() } -> std::convertible_to<VkBuffer>; }) {
-                return buf.vk_handle();
-            } else {
-                static_assert(sizeof(T) == 0, "Cannot extract VkBuffer handle from vk::Buffer");
-                return VK_NULL_HANDLE;
-            }
-        }
-
-    } // namespace
+        return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
+    }
+} // namespace
 
 //MemoryCommit is now VMA-backed
-    MemoryCommit::MemoryCommit(VmaAllocator alloc, VmaAllocation a,
-                               const VmaAllocationInfo &info) noexcept
-            : allocator{alloc}, allocation{a}, memory{info.deviceMemory},
-              offset{info.offset}, size{info.size}, mapped_ptr{info.pMappedData} {
-        // Log GPU memory allocation
+MemoryCommit::MemoryCommit(VmaAllocator alloc, VmaAllocation a,
+                            const VmaAllocationInfo &info) noexcept
+        : allocator{alloc}, allocation{a}, memory{info.deviceMemory},
+            offset{info.offset}, size{info.size}, mapped_ptr{info.pMappedData} {
+    // Log GPU memory allocation
+    if (GPU::Logging::IsActive() &&
+        Settings::values.gpu_log_memory_tracking.GetValue()) {
+        GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
+            reinterpret_cast<uintptr_t>(memory),
+            static_cast<u64>(size),
+            0  // Memory property flags (not easily available from VMA)
+        );
+    }
+}
+
+MemoryCommit::~MemoryCommit() { Release(); }
+
+MemoryCommit::MemoryCommit(MemoryCommit &&rhs) noexcept
+        : allocator{std::exchange(rhs.allocator, nullptr)},
+            allocation{std::exchange(rhs.allocation, nullptr)},
+            memory{std::exchange(rhs.memory, VK_NULL_HANDLE)},
+            offset{std::exchange(rhs.offset, 0)},
+            size{std::exchange(rhs.size, 0)},
+            mapped_ptr{std::exchange(rhs.mapped_ptr, nullptr)} {}
+
+MemoryCommit &MemoryCommit::operator=(MemoryCommit &&rhs) noexcept {
+    if (this != &rhs) {
+        Release();
+        allocator = std::exchange(rhs.allocator, nullptr);
+        allocation = std::exchange(rhs.allocation, nullptr);
+        memory = std::exchange(rhs.memory, VK_NULL_HANDLE);
+        offset = std::exchange(rhs.offset, 0);
+        size = std::exchange(rhs.size, 0);
+        mapped_ptr = std::exchange(rhs.mapped_ptr, nullptr);
+    }
+    return *this;
+}
+
+std::span<u8> MemoryCommit::Map()
+{
+    if (!allocation) return {};
+    if (!mapped_ptr) {
+        if (vmaMapMemory(allocator, allocation, &mapped_ptr) != VK_SUCCESS) return {};
+    }
+    const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
+                                                                (std::numeric_limits<size_t>::max)()));
+    return std::span<u8>{static_cast<u8 *>(mapped_ptr), n};
+}
+
+std::span<const u8> MemoryCommit::Map() const
+{
+    if (!allocation) return {};
+    if (!mapped_ptr) {
+        void *p = nullptr;
+        if (vmaMapMemory(allocator, allocation, &p) != VK_SUCCESS) return {};
+        const_cast<MemoryCommit *>(this)->mapped_ptr = p;
+    }
+    const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
+                                                                (std::numeric_limits<size_t>::max)()));
+    return std::span<const u8>{static_cast<const u8 *>(mapped_ptr), n};
+}
+
+void MemoryCommit::Unmap()
+{
+    if (allocation && mapped_ptr) {
+        vmaUnmapMemory(allocator, allocation);
+        mapped_ptr = nullptr;
+    }
+}
+
+void MemoryCommit::Release() {
+    if (allocation && allocator) {
+        // Log GPU memory deallocation
         if (GPU::Logging::IsActive() &&
-            Settings::values.gpu_log_memory_tracking.GetValue()) {
-            GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
-                reinterpret_cast<uintptr_t>(memory),
-                static_cast<u64>(size),
-                0  // Memory property flags (not easily available from VMA)
+            Settings::values.gpu_log_memory_tracking.GetValue() &&
+            memory != VK_NULL_HANDLE) {
+            GPU::Logging::GPULogger::GetInstance().LogMemoryDeallocation(
+                reinterpret_cast<uintptr_t>(memory)
             );
         }
-    }
 
-    MemoryCommit::~MemoryCommit() { Release(); }
-
-    MemoryCommit::MemoryCommit(MemoryCommit &&rhs) noexcept
-            : allocator{std::exchange(rhs.allocator, nullptr)},
-              allocation{std::exchange(rhs.allocation, nullptr)},
-              memory{std::exchange(rhs.memory, VK_NULL_HANDLE)},
-              offset{std::exchange(rhs.offset, 0)},
-              size{std::exchange(rhs.size, 0)},
-              mapped_ptr{std::exchange(rhs.mapped_ptr, nullptr)} {}
-
-    MemoryCommit &MemoryCommit::operator=(MemoryCommit &&rhs) noexcept {
-        if (this != &rhs) {
-            Release();
-            allocator = std::exchange(rhs.allocator, nullptr);
-            allocation = std::exchange(rhs.allocation, nullptr);
-            memory = std::exchange(rhs.memory, VK_NULL_HANDLE);
-            offset = std::exchange(rhs.offset, 0);
-            size = std::exchange(rhs.size, 0);
-            mapped_ptr = std::exchange(rhs.mapped_ptr, nullptr);
-        }
-        return *this;
-    }
-
-    std::span<u8> MemoryCommit::Map()
-    {
-        if (!allocation) return {};
-        if (!mapped_ptr) {
-            if (vmaMapMemory(allocator, allocation, &mapped_ptr) != VK_SUCCESS) return {};
-        }
-        const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
-                                                                    (std::numeric_limits<size_t>::max)()));
-        return std::span<u8>{static_cast<u8 *>(mapped_ptr), n};
-    }
-
-    std::span<const u8> MemoryCommit::Map() const
-    {
-        if (!allocation) return {};
-        if (!mapped_ptr) {
-            void *p = nullptr;
-            if (vmaMapMemory(allocator, allocation, &p) != VK_SUCCESS) return {};
-            const_cast<MemoryCommit *>(this)->mapped_ptr = p;
-        }
-        const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
-                                                                    (std::numeric_limits<size_t>::max)()));
-        return std::span<const u8>{static_cast<const u8 *>(mapped_ptr), n};
-    }
-
-    void MemoryCommit::Unmap()
-    {
-        if (allocation && mapped_ptr) {
+        if (mapped_ptr) {
             vmaUnmapMemory(allocator, allocation);
             mapped_ptr = nullptr;
         }
+        vmaFreeMemory(allocator, allocation);
     }
+    allocation = nullptr;
+    allocator = nullptr;
+    memory = VK_NULL_HANDLE;
+    offset = 0;
+    size = 0;
+}
 
-    void MemoryCommit::Release() {
-        if (allocation && allocator) {
-            // Log GPU memory deallocation
-            if (GPU::Logging::IsActive() &&
-                Settings::values.gpu_log_memory_tracking.GetValue() &&
-                memory != VK_NULL_HANDLE) {
-                GPU::Logging::GPULogger::GetInstance().LogMemoryDeallocation(
-                    reinterpret_cast<uintptr_t>(memory)
-                );
-            }
+MemoryAllocator::MemoryAllocator(const Device &device_)
+        : device{device_}, allocator{device.GetAllocator()},
+            properties{device_.GetPhysical().GetMemoryProperties().memoryProperties},
+            buffer_image_granularity{
+                    device_.GetPhysical().GetProperties().limits.bufferImageGranularity} {
 
-            if (mapped_ptr) {
-                vmaUnmapMemory(allocator, allocation);
-                mapped_ptr = nullptr;
-            }
-            vmaFreeMemory(allocator, allocation);
-        }
-        allocation = nullptr;
-        allocator = nullptr;
-        memory = VK_NULL_HANDLE;
-        offset = 0;
-        size = 0;
-    }
-
-    MemoryAllocator::MemoryAllocator(const Device &device_)
-            : device{device_}, allocator{device.GetAllocator()},
-              properties{device_.GetPhysical().GetMemoryProperties().memoryProperties},
-              buffer_image_granularity{
-                      device_.GetPhysical().GetProperties().limits.bufferImageGranularity} {
-
-        // Preserve the previous "RenderDoc small heap" trimming behavior that we had in original vma minus the heap bug
-        if (device.HasDebuggingToolAttached())
-        {
-            using namespace Common::Literals;
-            ForEachDeviceLocalHostVisibleHeap(device, [this](size_t heap_idx, VkMemoryHeap &heap) {
-                if (heap.size <= 256_MiB) {
-                    for (u32 t = 0; t < properties.memoryTypeCount; ++t) {
-                        if (properties.memoryTypes[t].heapIndex == heap_idx) {
-                            valid_memory_types &= ~(1u << t);
-                        }
+    // Preserve the previous "RenderDoc small heap" trimming behavior that we had in original vma minus the heap bug
+    if (device.HasDebuggingToolAttached())
+    {
+        using namespace Common::Literals;
+        ForEachDeviceLocalHostVisibleHeap(device, [this](size_t heap_idx, VkMemoryHeap &heap) {
+            if (heap.size <= 256_MiB) {
+                for (u32 t = 0; t < properties.memoryTypeCount; ++t) {
+                    if (properties.memoryTypes[t].heapIndex == heap_idx) {
+                        valid_memory_types &= ~(1u << t);
                     }
                 }
-            });
-        }
+            }
+        });
     }
+}
 
-    MemoryAllocator::~MemoryAllocator() = default;
+MemoryAllocator::~MemoryAllocator() = default;
 
-    vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const
-    {
-        const VmaAllocationCreateInfo alloc_ci = {
-                .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT,
-                .usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE,
-                .requiredFlags = 0,
-                .preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
-                .memoryTypeBits = 0,
-                .pool = VK_NULL_HANDLE,
-                .pUserData = nullptr,
-                .priority = 0.f,
-        };
-
-        VkImage handle{};
-        VmaAllocation allocation{};
-        VmaAllocationInfo alloc_info{};
-        vk::Check(vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
-
-        // Log GPU memory allocation for images
-        if (GPU::Logging::IsActive() &&
-            Settings::values.gpu_log_memory_tracking.GetValue()) {
-            GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
-                reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
-                static_cast<u64>(alloc_info.size),
-                VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
-            );
-        }
-
-        return vk::Image(handle, ci.usage, *device.GetLogical(), allocator, allocation,
-                         device.GetDispatchLoader());
-    }
-
-    vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) const {
-        // MESA will do memcpy() if not marked as host cached, so just force mark it for most buffers
-        auto const anv_flags = (usage == MemoryUsage::Stream
-            && device.GetDriverID() == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA)
-            ? VK_MEMORY_PROPERTY_HOST_CACHED_BIT : 0;
-        const VmaAllocationCreateInfo alloc_ci = {
-            .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage),
-            .usage = MemoryUsageVma(usage),
+vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const
+{
+    const VmaAllocationCreateInfo alloc_ci = {
+            .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT,
+            .usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE,
             .requiredFlags = 0,
-            .preferredFlags = MemoryUsagePreferredVmaFlags(usage) | anv_flags,
-            .memoryTypeBits = usage == MemoryUsage::Stream ? 0u : valid_memory_types,
+            .preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
+            .memoryTypeBits = 0,
             .pool = VK_NULL_HANDLE,
             .pUserData = nullptr,
             .priority = 0.f,
-        };
+    };
 
-        VkBuffer handle{};
-        VmaAllocationInfo alloc_info{};
-        VmaAllocation allocation{};
-        VkMemoryPropertyFlags property_flags{};
+    VkImage handle{};
+    VmaAllocation allocation{};
+    VmaAllocationInfo alloc_info{};
+    vk::Check(vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
 
-        vk::Check(vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
-        vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
-
-        // Log GPU memory allocation for buffers
-        if (GPU::Logging::IsActive() &&
-            Settings::values.gpu_log_memory_tracking.GetValue()) {
-            GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
-                reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
-                static_cast<u64>(alloc_info.size),
-                property_flags
-            );
-        }
-
-        u8 *data = reinterpret_cast<u8 *>(alloc_info.pMappedData);
-        const std::span<u8> mapped_data = data ? std::span<u8>{data, ci.size} : std::span<u8>{};
-        const bool is_coherent = (property_flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
-
-        return vk::Buffer(handle, *device.GetLogical(), allocator, allocation, mapped_data,
-                          is_coherent,
-                          device.GetDispatchLoader());
+    // Log GPU memory allocation for images
+    if (GPU::Logging::IsActive() &&
+        Settings::values.gpu_log_memory_tracking.GetValue()) {
+        GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
+            reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
+            static_cast<u64>(alloc_info.size),
+            VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
+        );
     }
 
-    MemoryCommit MemoryAllocator::Commit(const VkMemoryRequirements &reqs, MemoryUsage usage)
-    {
-        const auto vma_usage = MemoryUsageVma(usage);
-        VmaAllocationCreateInfo ci{};
-        ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
-        ci.usage = vma_usage;
-        ci.memoryTypeBits = reqs.memoryTypeBits & valid_memory_types;
-        ci.requiredFlags = 0;
-        ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
+    return vk::Image(handle, ci.usage, *device.GetLogical(), allocator, allocation,
+                        device.GetDispatchLoader());
+}
 
-        VmaAllocation a{};
-        VmaAllocationInfo info{};
+vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) const {
+    // MESA will do memcpy() if not marked as host cached, so just force mark it for most buffers
+    auto const anv_flags = (usage == MemoryUsage::Stream
+        && device.GetDriverID() == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA)
+        ? VK_MEMORY_PROPERTY_HOST_CACHED_BIT : 0;
+    const VmaAllocationCreateInfo alloc_ci = {
+        .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage),
+        .usage = MemoryUsageVma(usage),
+        .requiredFlags = 0,
+        .preferredFlags = MemoryUsagePreferredVmaFlags(usage) | anv_flags,
+        .memoryTypeBits = usage == MemoryUsage::Stream ? 0u : valid_memory_types,
+        .pool = VK_NULL_HANDLE,
+        .pUserData = nullptr,
+        .priority = 0.f,
+    };
 
-        VkResult res = vmaAllocateMemory(allocator, &reqs, &ci, &a, &info);
+    VkBuffer handle{};
+    VmaAllocationInfo alloc_info{};
+    VmaAllocation allocation{};
+    VkMemoryPropertyFlags property_flags{};
 
-        if (res != VK_SUCCESS) {
-            // Relax 1: drop budget constraint
-            auto ci2 = ci;
-            ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
-            res = vmaAllocateMemory(allocator, &reqs, &ci2, &a, &info);
+    vk::Check(vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
+    vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
 
-            // Relax 2: if we preferred DEVICE_LOCAL, drop that preference
-            if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
-                auto ci3 = ci2;
-                ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
-                res = vmaAllocateMemory(allocator, &reqs, &ci3, &a, &info);
-            }
-        }
-
-        vk::Check(res);
-        return MemoryCommit(allocator, a, info);
+    // Log GPU memory allocation for buffers
+    if (GPU::Logging::IsActive() &&
+        Settings::values.gpu_log_memory_tracking.GetValue()) {
+        GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
+            reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
+            static_cast<u64>(alloc_info.size),
+            property_flags
+        );
     }
 
-    MemoryCommit MemoryAllocator::Commit(const vk::Buffer &buffer, MemoryUsage usage) {
-        // Allocate memory appropriate for this buffer automatically
-        const auto vma_usage = MemoryUsageVma(usage);
+    u8 *data = reinterpret_cast<u8 *>(alloc_info.pMappedData);
+    const std::span<u8> mapped_data = data ? std::span<u8>{data, ci.size} : std::span<u8>{};
+    const bool is_coherent = (property_flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
 
-        VmaAllocationCreateInfo ci{};
-        ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
-        ci.usage = vma_usage;
-        ci.requiredFlags = 0;
-        ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
-        ci.pool = VK_NULL_HANDLE;
-        ci.pUserData = nullptr;
-        ci.priority = 0.0f;
+    const vk::MemoryLocation location{
+        .memory = alloc_info.deviceMemory,
+        .offset = alloc_info.offset,
+        .memory_type = alloc_info.memoryType,
+    };
+    return vk::Buffer(handle, *device.GetLogical(), allocator, allocation, mapped_data, is_coherent,
+                      location, device.GetDispatchLoader());
+}
 
-        const VkBuffer raw = *buffer;
+vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage,
+                                         VkDeviceSize min_alignment) const {
+    if (min_alignment <= 1) {
+        return CreateBuffer(ci, usage);
+    }
+    VkMemoryPropertyFlags anv_flags = 0;
+    if (usage == MemoryUsage::Stream &&
+        device.GetDriverID() == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA) {
+        anv_flags = VK_MEMORY_PROPERTY_HOST_CACHED_BIT;
+    }
+    u32 memory_type_bits = valid_memory_types;
+    if (usage == MemoryUsage::Stream) {
+        memory_type_bits = 0u;
+    }
+    const VmaAllocationCreateInfo alloc_ci = {
+        .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage),
+        .usage = MemoryUsageVma(usage),
+        .requiredFlags = 0,
+        .preferredFlags = MemoryUsagePreferredVmaFlags(usage) | anv_flags,
+        .memoryTypeBits = memory_type_bits,
+        .pool = VK_NULL_HANDLE,
+        .pUserData = nullptr,
+        .priority = 0.f,
+    };
 
-        VmaAllocation a{};
-        VmaAllocationInfo info{};
+    VkBuffer handle{};
+    VmaAllocationInfo alloc_info{};
+    VmaAllocation allocation{};
+    VkMemoryPropertyFlags property_flags{};
 
-        // Let VMA infer memory requirements from the buffer
-        VkResult res = vmaAllocateMemoryForBuffer(allocator, raw, &ci, &a, &info);
+    vk::Check(vmaCreateBufferWithAlignment(allocator, &ci, &alloc_ci, min_alignment, &handle,
+                                           &allocation, &alloc_info));
+    vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
 
-        if (res != VK_SUCCESS) {
-            auto ci2 = ci;
-            ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
-            res = vmaAllocateMemoryForBuffer(allocator, raw, &ci2, &a, &info);
+    u8 *data = reinterpret_cast<u8 *>(alloc_info.pMappedData);
+    std::span<u8> mapped_data{};
+    if (data) {
+        mapped_data = std::span<u8>{data, ci.size};
+    }
+    const bool is_coherent = (property_flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
 
-            if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
-                auto ci3 = ci2;
-                ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
-                res = vmaAllocateMemoryForBuffer(allocator, raw, &ci3, &a, &info);
-            }
+    const vk::MemoryLocation location{
+        .memory = alloc_info.deviceMemory,
+        .offset = alloc_info.offset,
+        .memory_type = alloc_info.memoryType,
+    };
+    return vk::Buffer(handle, *device.GetLogical(), allocator, allocation, mapped_data, is_coherent,
+                      location, device.GetDispatchLoader());
+}
+
+MemoryCommit MemoryAllocator::Commit(const VkMemoryRequirements &reqs, MemoryUsage usage)
+{
+    const auto vma_usage = MemoryUsageVma(usage);
+    VmaAllocationCreateInfo ci{};
+    ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
+    ci.usage = vma_usage;
+    ci.memoryTypeBits = reqs.memoryTypeBits & valid_memory_types;
+    ci.requiredFlags = 0;
+    ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
+
+    VmaAllocation a{};
+    VmaAllocationInfo info{};
+
+    VkResult res = vmaAllocateMemory(allocator, &reqs, &ci, &a, &info);
+
+    if (res != VK_SUCCESS) {
+        // Relax 1: drop budget constraint
+        auto ci2 = ci;
+        ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
+        res = vmaAllocateMemory(allocator, &reqs, &ci2, &a, &info);
+
+        // Relax 2: if we preferred DEVICE_LOCAL, drop that preference
+        if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
+            auto ci3 = ci2;
+            ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+            res = vmaAllocateMemory(allocator, &reqs, &ci3, &a, &info);
         }
-
-        vk::Check(res);
-        vk::Check(vmaBindBufferMemory2(allocator, a, 0, raw, nullptr));
-        return MemoryCommit(allocator, a, info);
     }
 
+    vk::Check(res);
+    return MemoryCommit(allocator, a, info);
+}
 
+MemoryCommit MemoryAllocator::Commit(const vk::Buffer &buffer, MemoryUsage usage) {
+    // Allocate memory appropriate for this buffer automatically
+    const auto vma_usage = MemoryUsageVma(usage);
+
+    VmaAllocationCreateInfo ci{};
+    ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
+    ci.usage = vma_usage;
+    ci.requiredFlags = 0;
+    ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
+    ci.pool = VK_NULL_HANDLE;
+    ci.pUserData = nullptr;
+    ci.priority = 0.0f;
+
+    const VkBuffer raw = *buffer;
+
+    VmaAllocation a{};
+    VmaAllocationInfo info{};
+
+    // Let VMA infer memory requirements from the buffer
+    VkResult res = vmaAllocateMemoryForBuffer(allocator, raw, &ci, &a, &info);
+
+    if (res != VK_SUCCESS) {
+        auto ci2 = ci;
+        ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
+        res = vmaAllocateMemoryForBuffer(allocator, raw, &ci2, &a, &info);
+
+        if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
+            auto ci3 = ci2;
+            ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+            res = vmaAllocateMemoryForBuffer(allocator, raw, &ci3, &a, &info);
+        }
+    }
+
+    vk::Check(res);
+    vk::Check(vmaBindBufferMemory2(allocator, a, 0, raw, nullptr));
+    return MemoryCommit(allocator, a, info);
+}
 
 } // namespace Vulkan

src/video_core/vulkan_common/vulkan_memory_allocator.h

@@ -1,4 +1,4 @@
-// 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 2019 yuzu Emulator Project
@@ -107,6 +107,9 @@ namespace Vulkan {
 
         vk::Buffer CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) const;
 
+        vk::Buffer CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage,
+                                VkDeviceSize min_alignment) const;
+
         /**
          * Commits a memory with the specified requirements.
          *

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -229,6 +229,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkGetPipelineExecutableStatisticsKHR);
     X(vkGetSemaphoreCounterValue);
     X(vkMapMemory);
+    X(vkQueueBindSparse);
     X(vkQueueSubmit);
     X(vkQueueSubmit2);
     X(vkResetFences);

src/video_core/vulkan_common/vulkan_wrapper.h

@@ -345,6 +345,7 @@ struct DeviceDispatch : InstanceDispatch {
     PFN_vkGetQueryPoolResults vkGetQueryPoolResults{};
     PFN_vkGetSemaphoreCounterValue vkGetSemaphoreCounterValue{};
     PFN_vkMapMemory vkMapMemory{};
+    PFN_vkQueueBindSparse vkQueueBindSparse{};
     PFN_vkQueueSubmit vkQueueSubmit{};
     PFN_vkQueueSubmit2 vkQueueSubmit2{};
     PFN_vkResetFences vkResetFences{};
@@ -740,13 +741,20 @@ private:
     const DeviceDispatch* dld = nullptr;
 };
 
+struct MemoryLocation {
+    VkDeviceMemory memory{};
+    VkDeviceSize offset{};
+    u32 memory_type{};
+};
+
 class Buffer {
 public:
     explicit Buffer(VkBuffer handle_, VkDevice owner_, VmaAllocator allocator_,
                     VmaAllocation allocation_, std::span<u8> mapped_, bool is_coherent_,
-                    const DeviceDispatch& dld_) noexcept
+                    MemoryLocation location_, const DeviceDispatch& dld_) noexcept
         : handle{handle_}, owner{owner_}, allocator{allocator_},
-          allocation{allocation_}, mapped{mapped_}, is_coherent{is_coherent_}, dld{&dld_} {}
+          allocation{allocation_}, mapped{mapped_}, location{location_},
+          is_coherent{is_coherent_}, dld{&dld_} {}
     Buffer() = default;
 
     Buffer(const Buffer&) = delete;
@@ -754,7 +762,7 @@ public:
 
     Buffer(Buffer&& rhs) noexcept
         : handle{std::exchange(rhs.handle, VkBuffer{})}, owner{rhs.owner}, allocator{rhs.allocator},
-          allocation{rhs.allocation}, mapped{rhs.mapped},
+          allocation{rhs.allocation}, mapped{rhs.mapped}, location{rhs.location},
           is_coherent{rhs.is_coherent}, dld{rhs.dld} {}
 
     Buffer& operator=(Buffer&& rhs) noexcept {
@@ -764,6 +772,7 @@ public:
         allocator = rhs.allocator;
         allocation = rhs.allocation;
         mapped = rhs.mapped;
+        location = rhs.location;
         is_coherent = rhs.is_coherent;
         dld = rhs.dld;
         return *this;
@@ -811,6 +820,10 @@ public:
 
     void SetObjectNameEXT(const char* name) const;
 
+    MemoryLocation Location() const noexcept {
+        return location;
+    }
+
 private:
     void Release() const noexcept;
 
@@ -819,6 +832,7 @@ private:
     VmaAllocator allocator = nullptr;
     VmaAllocation allocation = nullptr;
     std::span<u8> mapped = {};
+    MemoryLocation location{};
     bool is_coherent = false;
     const DeviceDispatch* dld = nullptr;
 };
@@ -843,6 +857,11 @@ public:
         return dld->vkQueueSubmit2(queue, submit_infos.size(), submit_infos.data(), fence);
     }
 
+    VkResult BindSparse(Span<VkBindSparseInfo> bind_infos,
+                        VkFence fence = VK_NULL_HANDLE) const noexcept {
+        return dld->vkQueueBindSparse(queue, bind_infos.size(), bind_infos.data(), fence);
+    }
+
     VkResult Present(const VkPresentInfoKHR& present_info) const noexcept {
         return dld->vkQueuePresentKHR(queue, &present_info);
     }

src/yuzu/main_window.cpp

@@ -4796,6 +4796,6 @@ void VolumeButton::ResetMultiplier() {
 #endif
 
 #if !defined(QT_STATICPLUGIN) || defined(__APPLE__)
-#define VMA_IMPLEMENTATION
+#define VMA_IMPLEMENTATION 1
 #include "video_core/vulkan_common/vma.h"
 #endif

tools/cpm/cmake/package/version.cmake

@@ -65,7 +65,7 @@ else()
 endif()
 
 # update cached cpmfile content
-string(JSON cpmfile SET "${cpmfile}" "${key}" "${new_object}")
+string(JSON cpmfile SET "${cpmfile}" "${KEY}" "${new_object}")
 
 # write cached cpmfile
 get_cpmfile_path(file)