Compare commits

..

5 Commits

Author SHA1 Message Date
CamilleLaVey 49e96351b1 Just a quick test 2026-09-05 21:51:37 -04:00
CamilleLaVey 29b2b04a2d [shader_recompiler] Trackers on phi operands 2026-09-05 19:44:21 -04:00
CamilleLaVey 76b7a561ba [vulkan, buffer_cache] Enable sparse binding on buffer cache 2026-09-05 19:35:37 -04:00
lizzie 11de264541 [cmake] Fix some macOS compilation errors (#4350)
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>
2026-09-05 23:43:46 +02:00
lizzie f6e7686038 [ci] Fix macOS "no type named 'free' in namespace 'std' (#4347)
Signed-off-by: lizzie <lizzie@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.

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

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4347
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
2026-09-04 13:32:36 +02:00
62 changed files with 1467 additions and 447 deletions
@@ -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)
{
+2 -1
View File
@@ -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 "${CMAKE_CURRENT_SOURCE_DIR}/CMakeModules")
@@ -512,7 +513,7 @@ endfunction()
# =============================================
if (APPLE)
foreach(fw Carbon Metal Cocoa IOKit CoreVideo CoreMedia Security UniformTypeIdentifiers)
foreach(fw Carbon Metal Cocoa IOKit CoreVideo CoreMedia Security UniformTypeIdentifiers Foundation)
find_library(${fw}_LIBRARY ${fw} REQUIRED)
list(APPEND PLATFORM_LIBRARIES ${${fw}_LIBRARY})
endforeach()
+2 -2
View File
@@ -22,9 +22,9 @@ Eden is free, open-source, copyleft software, licensed under the terms of the [G
- No LLM or AI usage, *period*, for patches, pull requests, issues, comments, debugging, brainstorming, etc.
- For details on why, see the [detailed AI policy](docs/policies/AI.md).
- Usage of any form of profanity or otherwise unsavory language is prohibited.
- Usage of any form of profanity or otherwise unsavory language is generally discouraged.
- This is primarily because it rarely helps to actually understand what's going on.
- Remember that your comments should be focused and actually address what's happening.
- Remember that your comments should be focused and actually address what's happening. With very few exceptions, expletives are actively detrimental at best.
- New code must follow the same general style as the surrounding codebase. Exceptions may be granted in certain cases.
- Maintainers reserve the right to change your patches and pull requests at will. We will try to avoid this.
- You should generally respect all decisions made by the [code owners](docs/CODEOWNERS) in your particular subsystem.
+5 -10
View File
@@ -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"
},
+2 -2
View File
@@ -239,6 +239,6 @@ find ./*/ -name "*.dll" | while read -r dll; do deps "$dll"; done
## RedoxOS
Package installs randomly hang at times, you may need to restart it about 3-5 times to complete. Always do `sudo pkg update` or it will hang. Try installing each package individually.
The package install may randomly hang at times, in which case it has to be restarted. ALWAYS do a `sudo pkg update` or the chances of it hanging will be close to 90%. If "multiple" installs fail at once, try installing 1 by 1 the packages.
When CMake invokes certain file syscalls - most of the time it causes crashes, and corruptions on the address space. Reboot if there's a freeze over 5 minutes.
When CMake invokes certain file syscalls - it may sometimes cause crashes or corruptions on the (kernel?) address space - so reboot the system if there is a "hang" in CMake.
+4 -4
View File
@@ -38,7 +38,7 @@ Pull requests are only to be merged by core developers when properly tested and
- The level of namespacing is generally left to the committer's choice.
- However, we never recommend going more than two levels *except* in `hle`, in which case you may go as many as four levels depending on the specificity of your changes.
- Occasionally, up to two additional namespaces may be provided for more clarity.
- Ocassionally, up to two additional namespaces may be provided for more clarity.
- Changes that affect the entire project (sans CMake changes) should be namespaced as `meta`.
- Maintainers are permitted to change namespaces at will.
- Commits within PRs are not required to be namespaced, but it is highly recommended.
@@ -51,8 +51,8 @@ When adding new settings, use `tr("Setting:")` if the setting is meant to be a f
- Debug settings must never be turned on by default.
- Provide reasonable bounds (for example, a setting controlling the amount of VRAM should never be 0).
- The description of the setting must be short and concise, if the setting "does a lot of things" consider splitting the setting into multiple if possible.
- Try to avoid excessive/redundant explanations "recommended for most users and games" can just be "(recommended)".
- Try to not write "slow/fast" options unless it clearly degrades/increases performance for a given case, as most options may modify behavior that result in different metrics across different systems. If for example the option is an "accuracy" option, writing "High" is sufficient to imply "Slow". No need to write "High (Slow)".
- Try to avoid excessive/redundant explainations "recommended for most users and games" can just be "(recommended)".
- Try to not write "slow/fast" options unless it clearly degrades/increases performance for a given case, as most options may modify behaviour that result in different metrics accross different systems. If for example the option is an "accuracy" option, writing "High" is sufficient to imply "Slow". No need to write "High (Slow)".
Some examples:
@@ -91,7 +91,7 @@ You may additionally need the `Qt Extension Pack` extension if building Qt.
# Build speedup
If you have an HDD, use ramdisk (build in RAM); you need about 4GB for a `RelWithDebInfo` build:
If you have an HDD, use ramdisk (build in RAM), approximately you need 4GB for a full build with debug symbols:
```sh
mkdir /tmp/ramdisk
+2 -2
View File
@@ -3,9 +3,9 @@
> [!WARNING]
> This guide is intended for developers ONLY. If you're looking for configuring the emulator itself, read **[the user handbook](./user/README.md)**.
Settings have their own dedicated subsystem. Toggles can be added to gate features and fixes, as not all of them apply to every game.
Settings on the emulator are very important, toggles and such can be used to guard and/or add branches to paths where some games may crash while others won't, and viceversa.
However, this process can be tedious for those unfamiliar; this document serves as an outline for the settings subsystem.
However, this process can be tedious for those unfamiliar; this document serves as an outline/documentation for the settings subsystem.
## Index
+5 -5
View File
@@ -12,7 +12,7 @@ AMD64, aka x86_64, is the most tested and supported architecture for desktop tar
### Caveats
AMD64 systems are almost always limited by the CPU. For example, a Zen 5/RX 6600 system will often hit max CPU usage before the GPU ever reaches 70% usage, with minimal exceptions (that tend to pop up only at >200fps). JIT is slow.
AMD64 systems are almost always limited by the CPU. For example, a Zen 5/RX 6600 system will often hit max CPU usage before the GPU ever reaches 70% usage, with minimal exceptions (that tend to pop up only at >200fps). JIT is slow!
Computers on Linux will almost always run Eden strictly better than an equivalent machine on Windows. This is largely due to the way the Linux kernel handles memory management (and the lack of Microsoft spyware).
@@ -24,7 +24,7 @@ ARM64, aka aarch64, is the only supported architecture for Android, with limited
### Caveats
NCE (Native Code Execution) is currently only available on Android and (experimentally) Linux. Support for macOS is in the works, but Windows is extremely unlikely to ever happen (if you want it -- submit patches). Generally, if NCE is available, you should pretty much always use it due to the massive performance hit JIT has.
NCE (Native Code Execution) is currently only available on Android and (experimentally) Linux. Support for macOS is in the works, but Windows is extremely unlikely to ever happen (if you want it--submit patches!). Generally, if NCE is available, you should pretty much always use it due to the massive performance hit JIT has.
When NCE is enabled, do note that the GPU will almost always be the limiting factor. This is especially the case for Android, as well as desktops that lack dedicated GPUs; Adreno, Mali, PowerVR, etc. GPUs are generally significantly weaker relative to their respective CPUs.
@@ -44,7 +44,7 @@ Only Fedora/riscv64 has been tested, but in theory, every riscv64 distribution t
## Other
Other architectures, such as SPARC, MIPS, PowerPC, Loong, and all 32-bit architectures are completely unsupported, as there is no JIT backend or emitter thereof. If you want support for it -- submit patches.
Other architectures, such as SPARC, MIPS, PowerPC, Loong, and all 32-bit architectures are completely unsupported, as there is no JIT backend or emitter thereof. If you want support for it -- submit patches!
IA-64 (Itanium) support is completely unknown. Existing amd64 packages will not run on IA-64 (assuming you can even find a supported Windows/Linux distribution)
@@ -106,14 +106,14 @@ Qualcomm Snapdragon SoCs are generally the most well supported.
* A good base to compare to is the Snapdragon 865--e.g. [Tensor vs SD865](https://archive.is/M1P58)
* Some benchmarks may be misleading due to thermal throttling OR RAM requirements.
- For example, a Pixel 6a (Tensor G1) performs about 1/3 as well as an 865 due to its lack of RAM and poor thermals.
* Remember--always use a cooler if you can, and you MUST have *at least* 8GB of RAM.
* Remember--always use a cooler if you can, and you MUST have *at least* 8GB of RAM!
- If you're not sure what SoC you have, check [GSMArena](https://www.gsmarena.com) - e.g. [Pixel 9 Pro](https://archive.ph/91VhA)
Custom ROMs are recommended, *as long as* you know what you're doing.
- For most devices, [LineageOS](https://lineageos.org/) is preferred.
- [CalyxOS](https://calyxos.org/) is available as well.
- For Google Pixel devices ONLY... and [soon another OEM](https://archive.ph/cPpMd)... [GrapheneOS](https://grapheneos.org/) is highly recommended.
* As of October 5, 2025, the Pixel 10 line is unsupported, however, [it will be](https://archive.is/viAUl) in the very near future.
* As of October 5, 2025, the Pixel 10 line is unsupported, however, [it will be](https://archive.is/viAUl) in the very near future!
* Keep checking the [FAQ page](https://grapheneos.org/faq#supported-devices) for news.
- Custom ROMs will likely be exclusively recommended in the future due to Google's upcoming [draconian](https://archive.is/hGIjZ), [anti-privacy, anti-user](https://archive.is/mc1CJ) verification requirements.
+1
View File
@@ -11,6 +11,7 @@
#include <limits>
#include <span>
#include <array>
#include <algorithm>
#include <time.h>
namespace Tz {
+1
View File
@@ -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>
+5 -1
View File
@@ -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;
}
+1
View File
@@ -10,6 +10,7 @@
#include <functional>
#include <span>
#include <string>
#include <type_traits>
#include "common/common_types.h"
+2 -2
View File
@@ -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
+2 -1
View File
@@ -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 {
@@ -7,6 +7,7 @@
#pragma once
#include <optional>
#include <type_traits>
#include "common/literals.h"
#include "core/file_sys/fssystem/fs_i_storage.h"
@@ -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"
@@ -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"
@@ -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"
@@ -6,6 +6,8 @@
#pragma once
#include <type_traits>
#include <cstddef>
#include "common/literals.h"
#include "core/file_sys/errors.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"
@@ -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"
@@ -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"
@@ -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"
@@ -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"
+1
View File
@@ -7,6 +7,7 @@
#pragma once
#include <memory>
#include <type_traits>
#include "common/common_funcs.h"
#include "common/page_table.h"
+1
View File
@@ -6,6 +6,7 @@
#pragma once
#include <type_traits>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
+4
View File
@@ -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"
+8 -26
View File
@@ -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);
+2 -7
View File
@@ -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(),
+4
View File
@@ -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"
+2 -1
View File
@@ -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"
+1
View File
@@ -8,6 +8,7 @@
#include <array>
#include <chrono>
#include <type_traits>
#include <fmt/ranges.h>
#include "common/common_types.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"
@@ -7,6 +7,7 @@
#pragma once
#include <array>
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_funcs.h"
+7
View File
@@ -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"
+53 -51
View File
@@ -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)
@@ -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) {
+59 -3
View File
@@ -312,10 +312,66 @@ static inline std::optional<ConstBufferAddr> TrackCached(const IR::Value& v, Env
std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env, const HostTranslateInfo& host_info);
bool IsSameConstBufferAddr(const ConstBufferAddr& lhs, const ConstBufferAddr& rhs) {
return lhs.index == rhs.index && lhs.offset == rhs.offset &&
lhs.shift_left == rhs.shift_left && lhs.secondary_index == rhs.secondary_index &&
lhs.secondary_offset == rhs.secondary_offset &&
lhs.secondary_shift_left == rhs.secondary_shift_left && lhs.count == rhs.count &&
lhs.has_secondary == rhs.has_secondary && lhs.dynamic_offset == rhs.dynamic_offset;
}
std::optional<ConstBufferAddr> TrackPhi(const IR::Inst* phi, Environment& env,
const HostTranslateInfo& host_info, bool& ambiguous) {
std::optional<ConstBufferAddr> agreed;
const size_t num_args{phi->NumArgs()};
for (size_t index = 0; index < num_args; ++index) {
const IR::Value arg{phi->Arg(index).Resolve()};
if (arg.IsImmediate()) {
ambiguous = true;
return std::nullopt;
}
const IR::Inst* arg_inst{arg.InstRecursive()};
if (arg_inst == phi) {
continue;
}
if (arg_inst->GetOpcode() == IR::Opcode::Phi) {
ambiguous = true;
return std::nullopt;
}
const std::optional<ConstBufferAddr> operand{TrackCached(arg, env, host_info)};
if (!operand) {
ambiguous = true;
return std::nullopt;
}
if (!agreed) {
agreed = operand;
continue;
}
if (!IsSameConstBufferAddr(*agreed, *operand)) {
ambiguous = true;
return std::nullopt;
}
}
if (!agreed) {
ambiguous = true;
}
return agreed;
}
std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env, const HostTranslateInfo& host_info) {
return IR::BreadthFirstSearch(value, [&env, &host_info](const IR::Inst* inst) {
return TryGetConstBuffer(inst, env, host_info);
});
bool ambiguous = false;
const std::optional<ConstBufferAddr> result{IR::BreadthFirstSearch(
value, [&env, &host_info, &ambiguous](const IR::Inst* inst)
-> std::optional<ConstBufferAddr> {
if (inst->GetOpcode() == IR::Opcode::Phi) {
return TrackPhi(inst, env, host_info, ambiguous);
}
return TryGetConstBuffer(inst, env, host_info);
})};
if (ambiguous) {
return std::nullopt;
}
return result;
}
std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) {
+3
View File
@@ -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
+76 -9
View File
@@ -112,6 +112,11 @@ 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) {
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)) {
@@ -998,11 +1003,56 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
channel_state->fast_bound_uniform_buffers[stage] &= ~(1u << binding_index);
}
template <class P>
bool BufferCache<P>::BindMultiRangeStorage(const Binding& binding, bool is_written) {
if constexpr (requires { runtime.BindMultiRangeStorageBuffer(u64{}); }) {
if (binding.gpu_addr == 0 || binding.size == 0) {
return false;
}
if (is_written && !runtime.PrefersSparseSources()) {
return false;
}
const VirtualSegments* segments =
virtual_ranges.Query(*gpu_memory, binding.gpu_addr, binding.size);
if (!segments || segments->size() < 2) {
return false;
}
const u64 key = (static_cast<u64>(gpu_memory->GetID()) << 48) ^ binding.gpu_addr;
const bool prefer_sparse = runtime.PrefersSparseSources();
runtime.ResetMultiRange();
for (const VirtualSegment& segment : *segments) {
const BufferId buffer_id =
FindBuffer(segment.device_addr, segment.size, prefer_sparse);
if (!buffer_id) {
return false;
}
Buffer& buffer = slot_buffers[buffer_id];
TouchBuffer(buffer, 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(buffer_id, segment.device_addr, segment.size);
}
runtime.PushMultiRangeSource(buffer, offset, segment.size);
}
return runtime.BindMultiRangeStorageBuffer(key);
} 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)) {
return;
}
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer, binding.buffer_id);
const u32 size = binding.size;
@@ -1010,7 +1060,6 @@ 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);
@@ -1139,6 +1188,11 @@ 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)) {
return;
}
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer, binding.buffer_id);
const u32 size = binding.size;
@@ -1146,8 +1200,6 @@ 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);
@@ -1440,7 +1492,7 @@ 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;
}
@@ -1450,10 +1502,18 @@ 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>
@@ -1575,13 +1635,15 @@ 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);
wanted_size = static_cast<u32>(device_addr_end - device_addr);
const OverlapResult overlap = ResolveOverlaps(device_addr, wanted_size);
const u32 size = static_cast<u32>(overlap.end - overlap.begin);
const BufferId new_buffer_id = slot_buffers.insert(runtime, overlap.begin, size);
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);
@@ -1934,10 +1996,15 @@ 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),
.size = binding_size,
.buffer_id = BufferId{},
.gpu_addr = aligned_gpu_addr,
};
return binding;
}
@@ -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"
@@ -83,6 +84,7 @@ struct Binding {
DAddr device_addr{};
u32 size{};
BufferId buffer_id;
GPUVAddr gpu_addr{};
};
struct TextureBufferBinding : Binding {
@@ -215,6 +217,10 @@ public:
void TickFrame();
bool BindMultiRangeStorage(const Binding& binding, bool is_written);
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);
@@ -414,7 +420,8 @@ 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 = false);
void WaitForGpuFenceIfNeeded(Buffer& buffer);
@@ -422,7 +429,8 @@ 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 = false);
void Register(BufferId buffer_id);
@@ -514,6 +522,7 @@ private:
};
Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
u64 frame_tick = 0;
VirtualRangeCache virtual_ranges;
u64 total_used_memory = 0;
u64 minimum_memory = 0;
u64 critical_memory = 0;
@@ -0,0 +1,152 @@
// 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 <unordered_map>
#include <vector>
#include <boost/container/small_vector.hpp>
#include "common/common_types.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:
const VirtualSegments* Query(Tegra::MemoryManager& memory, GPUVAddr gpu_addr, u32 size) {
if (has_deferred.load(std::memory_order_acquire)) {
ApplyDeferred();
}
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);
for (const auto& [range_addr, range_size] : ranges) {
const std::optional<DAddr> device_addr = memory.GpuToCpuAddress(range_addr);
if (!device_addr) {
break;
}
u32 segment_size = (std::numeric_limits<u32>::max)();
if (range_size < static_cast<size_t>(segment_size)) {
segment_size = static_cast<u32>(range_size);
}
entry.segments.push_back(VirtualSegment{
.gpu_addr = range_addr,
.device_addr = *device_addr,
.size = segment_size,
});
}
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;
}
}
deferred.push_back(DeferredUnmap{
.as_id = as_id,
.gpu_addr = gpu_addr,
.size = size,
});
}
has_deferred.store(true, std::memory_order_release);
}
void Clear() {
{
std::scoped_lock lock{deferred_mutex};
deferred.clear();
}
has_deferred.store(false, std::memory_order_release);
entries.clear();
}
private:
struct Entry {
size_t as_id{};
GPUVAddr gpu_addr{};
u32 size{};
VirtualSegments segments;
};
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;
{
std::scoped_lock lock{deferred_mutex};
has_deferred.store(false, std::memory_order_release);
pending.swap(deferred);
}
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;
}
}
}
std::unordered_map<u64, Entry> entries;
std::vector<DeferredUnmap> deferred;
std::mutex deferred_mutex;
std::atomic<bool> has_deferred{false};
};
} // namespace VideoCommon
+1
View File
@@ -8,6 +8,7 @@
#include <array>
#include <vector>
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_funcs.h"
+7
View File
@@ -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"
+1
View File
@@ -7,6 +7,7 @@
#pragma once
#include <memory>
#include <type_traits>
#include "common/common_types.h"
#include "common/scratch_buffer.h"
@@ -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()) {
@@ -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 = false);
explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams);
void ImmediateUpload(size_t offset, std::span<const u8> data) noexcept;
@@ -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_, memory_allocator_, scheduler_) {
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 +545,33 @@ void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t si
});
}
bool BufferCacheRuntime::BindMultiRangeStorageBuffer(u64 key) {
if (multi_range_sources.empty() || multi_range_total == 0) {
return false;
}
const MultiRangeRef ref = multi_range_buffers.Get(key, multi_range_sources, multi_range_total);
if (ref.handle == VK_NULL_HANDLE) {
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 = static_cast<u64>(source.offset),
.dst_offset = static_cast<u64>(dst_offset),
.size = static_cast<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) {
@@ -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_ = false);
[[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);
@@ -155,6 +168,40 @@ public:
return ref.mapped_span;
}
[[nodiscard]] VkDeviceSize SparseAlignmentFor(bool sparse_compatible) const noexcept {
if (!sparse_compatible || !multi_range_buffers.UsesSparse()) {
return 0;
}
return multi_range_buffers.BlockSize();
}
[[nodiscard]] bool PrefersSparseSources() const noexcept {
return multi_range_buffers.UsesSparse();
}
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,
});
multi_range_total += size;
}
bool BindMultiRangeStorageBuffer(u64 key);
void InvalidateMultiRange(u64 key) {
multi_range_buffers.Invalidate(key);
}
void BindUniformBuffer(const Buffer& buffer, u32 offset, u32 size) {
BindBuffer(buffer, offset, size);
}
@@ -208,6 +255,10 @@ 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)();
};
@@ -0,0 +1,304 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include <mutex>
#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_,
MemoryAllocator& memory_allocator_,
Scheduler& scheduler_)
: device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_} {
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;
}
const VkDeviceSize queried = QueryBlockSize();
if (queried == 0) {
return;
}
block_size = queried;
use_sparse = true;
}
MultiRangeBufferCache::~MultiRangeBufferCache() {
const VkDevice logical = *device.GetLogical();
const auto& dld = device.GetDispatchLoader();
for (auto& [key, entry] : entries) {
if (entry.sparse_handle != VK_NULL_HANDLE) {
dld.vkDestroyBuffer(logical, entry.sparse_handle, nullptr);
}
}
entries.clear();
for (const Retired& item : retired) {
dld.vkDestroyBuffer(logical, item.handle, nullptr);
}
retired.clear();
}
VkDeviceSize MultiRangeBufferCache::QueryBlockSize() 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 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);
dld.vkDestroyBuffer(logical, probe, nullptr);
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(reinterpret_cast<u64>(source.handle));
mix(static_cast<u64>(source.offset));
mix(static_cast<u64>(source.size));
}
return hash;
}
bool MultiRangeBufferCache::CanBindSparse(std::span<const MultiRangeSource> sources) const {
if (!use_sparse) {
return false;
}
for (const MultiRangeSource& source : sources) {
if (source.memory == VK_NULL_HANDLE) {
return false;
}
const VkDeviceSize memory_offset = source.memory_offset + source.offset;
if ((memory_offset % block_size) != 0) {
return false;
}
if ((source.size % block_size) != 0) {
return false;
}
}
return true;
}
VkBuffer MultiRangeBufferCache::CreateSparse(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 handle{};
if (dld.vkCreateBuffer(logical, &buffer_ci, nullptr, &handle) != VK_SUCCESS) {
return VK_NULL_HANDLE;
}
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 = handle,
.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) {
dld.vkDestroyBuffer(logical, handle, nullptr);
return VK_NULL_HANDLE;
}
fence.Wait();
return handle;
}
void MultiRangeBufferCache::DestroySparse(VkBuffer handle) {
if (handle == VK_NULL_HANDLE) {
return;
}
retired.push_back(Retired{
.handle = handle,
.tick = scheduler.CurrentTick(),
});
}
void MultiRangeBufferCache::DrainRetired() {
const VkDevice logical = *device.GetLogical();
const auto& dld = device.GetDispatchLoader();
size_t index = 0;
while (index < retired.size()) {
if (scheduler.IsFree(retired[index].tick)) {
dld.vkDestroyBuffer(logical, retired[index].handle, nullptr);
retired[index] = retired.back();
retired.pop_back();
} else {
++index;
}
}
}
MultiRangeRef MultiRangeBufferCache::Get(u64 key, std::span<const MultiRangeSource> sources,
VkDeviceSize total) {
if (sources.empty() || total == 0) {
return MultiRangeRef{};
}
if (!retired.empty()) {
DrainRetired();
}
const u64 geometry = HashSources(sources);
const auto it = entries.find(key);
if (it != entries.end() && it->second.geometry == geometry && it->second.size == total) {
Entry& entry = it->second;
MultiRangeRef ref{
.handle = entry.sparse_handle,
.address = entry.address,
.size = entry.size,
.needs_gather = false,
};
if (entry.sparse_handle == VK_NULL_HANDLE) {
ref.handle = *entry.gathered;
ref.needs_gather = entry.dirty;
}
return ref;
}
if (it != entries.end()) {
DestroySparse(it->second.sparse_handle);
entries.erase(it);
}
Entry entry;
entry.geometry = geometry;
entry.size = total;
if (CanBindSparse(sources)) {
entry.sparse_handle = CreateSparse(sources, total);
}
if (entry.sparse_handle == VK_NULL_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 (address_handle == VK_NULL_HANDLE) {
address_handle = *entry.gathered;
}
entry.address = device.GetLogical().GetBufferDeviceAddress(address_handle);
}
MultiRangeRef ref{
.handle = entry.sparse_handle,
.address = entry.address,
.size = entry.size,
.needs_gather = false,
};
if (entry.sparse_handle == VK_NULL_HANDLE) {
ref.handle = *entry.gathered;
ref.needs_gather = true;
}
entries.emplace(key, std::move(entry));
return ref;
}
void MultiRangeBufferCache::MarkGathered(u64 key) {
const auto it = entries.find(key);
if (it != entries.end()) {
it->second.dirty = false;
}
}
void MultiRangeBufferCache::Invalidate(u64 key) {
const auto it = entries.find(key);
if (it != entries.end()) {
it->second.dirty = true;
}
}
void MultiRangeBufferCache::Clear() {
for (auto& [key, entry] : entries) {
DestroySparse(entry.sparse_handle);
}
entries.clear();
}
} // namespace Vulkan
@@ -0,0 +1,100 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <span>
#include <unordered_map>
#include <vector>
#include "common/common_types.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
class Device;
class Scheduler;
struct MultiRangeSource {
VkBuffer handle{};
VkDeviceMemory memory{};
VkDeviceSize memory_offset{};
VkDeviceSize offset{};
VkDeviceSize size{};
};
struct MultiRangeRef {
VkBuffer handle{};
VkDeviceAddress address{};
VkDeviceSize size{};
bool needs_gather{};
};
class MultiRangeBufferCache final {
public:
static constexpr VkDeviceSize DEFAULT_BLOCK_SIZE = 64 * 1024;
explicit MultiRangeBufferCache(const Device& device_, MemoryAllocator& memory_allocator_,
Scheduler& scheduler_);
~MultiRangeBufferCache();
MultiRangeBufferCache(const MultiRangeBufferCache&) = delete;
MultiRangeBufferCache& operator=(const MultiRangeBufferCache&) = delete;
[[nodiscard]] bool UsesSparse() const noexcept {
return use_sparse;
}
[[nodiscard]] VkDeviceSize BlockSize() const noexcept {
return block_size;
}
[[nodiscard]] MultiRangeRef Get(u64 key, std::span<const MultiRangeSource> sources,
VkDeviceSize total);
void MarkGathered(u64 key);
void Invalidate(u64 key);
void Clear();
private:
struct Retired {
VkBuffer handle{};
u64 tick{};
};
struct Entry {
vk::Buffer gathered;
VkBuffer sparse_handle{};
VkDeviceAddress address{};
VkDeviceSize size{};
u64 geometry{};
bool dirty{true};
};
[[nodiscard]] u64 HashSources(std::span<const MultiRangeSource> sources) const;
[[nodiscard]] bool CanBindSparse(std::span<const MultiRangeSource> sources) const;
[[nodiscard]] VkBuffer CreateSparse(std::span<const MultiRangeSource> sources,
VkDeviceSize total);
[[nodiscard]] VkDeviceSize QueryBlockSize() const;
void DestroySparse(VkBuffer handle);
void DrainRetired();
const Device& device;
MemoryAllocator& memory_allocator;
Scheduler& scheduler;
bool use_sparse{};
VkDeviceSize block_size{DEFAULT_BLOCK_SIZE};
VkBufferUsageFlags sparse_usage{};
std::unordered_map<u64, Entry> entries;
std::vector<Retired> retired;
};
} // namespace Vulkan
@@ -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
@@ -819,6 +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) {
@@ -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.
@@ -1570,6 +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;
@@ -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()) {
@@ -1146,6 +1150,7 @@ private:
bool owns_static_pipeline_cache{};
u32 instance_version{}; ///< Vulkan instance version.
u32 graphics_family{}; ///< Main graphics queue family index.
bool graphics_family_sparse_binding{};
u32 present_family{}; ///< Main present queue family index.
struct Extensions {
@@ -26,350 +26,408 @@
#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);
}
}
// 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());
}
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,
};
const VkDevice logical = *device.GetLogical();
const auto &dld = device.GetDispatchLoader();
VkBuffer handle{};
vk::Check(dld.vkCreateBuffer(logical, &ci, nullptr, &handle));
const VkBufferMemoryRequirementsInfo2 reqs_info{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2,
.pNext = nullptr,
.buffer = handle,
};
VkMemoryRequirements2 reqs2{
.sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2,
.pNext = nullptr,
.memoryRequirements = {},
};
dld.vkGetBufferMemoryRequirements2(logical, &reqs_info, &reqs2);
VkMemoryRequirements reqs = reqs2.memoryRequirements;
reqs.alignment = (std::max)(reqs.alignment, min_alignment);
reqs.memoryTypeBits &= alloc_ci.memoryTypeBits;
VmaAllocation allocation{};
VmaAllocationInfo alloc_info{};
VkResult res = vmaAllocateMemory(allocator, &reqs, &alloc_ci, &allocation, &alloc_info);
if (res != VK_SUCCESS) {
auto relaxed = alloc_ci;
relaxed.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
res = vmaAllocateMemory(allocator, &reqs, &relaxed, &allocation, &alloc_info);
}
if (res != VK_SUCCESS) {
dld.vkDestroyBuffer(logical, handle, nullptr);
vk::Check(res);
return MemoryCommit(allocator, a, info);
}
const VkResult bind_res = vmaBindBufferMemory(allocator, allocation, handle);
if (bind_res != VK_SUCCESS) {
vmaFreeMemory(allocator, allocation);
dld.vkDestroyBuffer(logical, handle, nullptr);
vk::Check(bind_res);
}
MemoryCommit MemoryAllocator::Commit(const vk::Buffer &buffer, MemoryUsage usage) {
// Allocate memory appropriate for this buffer automatically
const auto vma_usage = MemoryUsageVma(usage);
VkMemoryPropertyFlags property_flags{};
vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
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;
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;
const VkBuffer raw = *buffer;
return vk::Buffer(handle, logical, allocator, allocation, mapped_data, is_coherent, dld);
}
VmaAllocation a{};
VmaAllocationInfo info{};
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);
// Let VMA infer memory requirements from the buffer
VkResult res = vmaAllocateMemoryForBuffer(allocator, raw, &ci, &a, &info);
VmaAllocation a{};
VmaAllocationInfo info{};
if (res != VK_SUCCESS) {
auto ci2 = ci;
ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
res = vmaAllocateMemoryForBuffer(allocator, raw, &ci2, &a, &info);
VkResult res = vmaAllocateMemory(allocator, &reqs, &ci, &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);
}
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
@@ -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.
*
@@ -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);
@@ -539,6 +540,18 @@ void Buffer::SetObjectNameEXT(const char* name) const {
SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_BUFFER, name);
}
MemoryLocation Buffer::Location() const noexcept {
if (!allocation) {
return MemoryLocation{};
}
VmaAllocationInfo info{};
vmaGetAllocationInfo(allocator, allocation, &info);
return MemoryLocation{
.memory = info.deviceMemory,
.offset = info.offset,
};
}
void Buffer::Release() const noexcept {
if (handle) {
vmaDestroyBuffer(allocator, handle, allocation);
@@ -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{};
@@ -740,6 +741,11 @@ private:
const DeviceDispatch* dld = nullptr;
};
struct MemoryLocation {
VkDeviceMemory memory{};
VkDeviceSize offset{};
};
class Buffer {
public:
explicit Buffer(VkBuffer handle_, VkDevice owner_, VmaAllocator allocator_,
@@ -811,6 +817,8 @@ public:
void SetObjectNameEXT(const char* name) const;
MemoryLocation Location() const noexcept;
private:
void Release() const noexcept;
@@ -843,6 +851,11 @@ 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);
}
+1 -1
View File
@@ -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
+1 -1
View File
@@ -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)