This can be inlined by the compiler much easier which helps perf a fair bit due to the number of times buffers are looked up, also avoids the need for small vector construction that was done in the previous fast-path.
This isn't a guarantee provided by actual HW so we don't need to provide it either, the sync can be skipped once the buffer already been synced at least once within the execution.
Constructing the GPU copy callback in `ConstantBuffers::Load()` ended up taking a fair amount of time despite it almost never being used in practice. By making it optional it can be skipped most of the time and only done when it's actually neccessary by calling `Write()` again if the initial call returned true.
Buffer views creation was a significant pain point, requiring several layers of caching to reduce the number of creations that introduced a lot of complexity. By reworking delegates to be per-buffer rather than per-view and then linearly allocating delegates (without ever freeing) views can be reduced to just {delegatePtr, offset, size}, avoiding the need for any allocations or set operations in GetView. The one difficulty with this is the need to support buffer recreation, which is achived by allowing delegates to be chained - during recreation all source buffers have their delegates modified to point to the newly created buffer's delegate. Upon accessing a view with such a chained delegate the view will be modified to point directly to the end delegate with offset being updated accordingly, skipping the need to traverse the chain for future accesses.
In the upcoming GPU code each state member will hold a reference to its corresponding Maxwell 3D regs, this helper is needed to allow easy transformation from the the main 3D register struct into them.
Example:
```c++
struct Regs {
std::array<View, 10> viewRegs;
u32 enable;
} regs;
struct ViewState {
const View &view;
const u32 &enable;
size_t index;
};
std::array<ViewState, 10> viewStates{MergeInto<ViewState, 10>(regs.viewRegs, regs.enable, IncrementingT{})
```
Useful for cases where allocations are guaranteed to be unused by the time `Reset()` is called and calling `Free()` would be difficult or add extra performance cost due to how the allocation is used.
In some games performing the binary search in `TranslateRange()` ended up taking a fairly large (~8%) proportion of GPFIFO time. By using a segment table for O(1) lookups this is reduced to <2% for non-split mappings at the cost of slightly increased memory usage (2GiB in the absolute worse case but more like 50MiB in real world situations).
In addition to adapting `TranslateRange()` to use the segment table, a new function `LookupBlock()` for cases where only a single mapping would ever be looked up so the small_vector handling and fallback paths can be skipped and the entire lookup be inlined.
Forward this function to OpenSaveDataFileSystem for now. A proper implementation should wrap the underlying filesystem with nn::fs::ReadOnlyFileSystem.
We want to know when the `KProcess` is being killed and flushing log during it is important since it can often result in hangs due to joining not working correctly.
We currently don't wait on a slot to be freed if none are free, this worked prior to async presentation as GBP's slots wouldn't change their state until other commands were called but now slots can be held by the presentation engine. As a result, we now have to wait on the presentation engine to free up slots.
This commit also fixes the behavior of the `async` flag in `DequeueBuffer` as it was treated as a non-blocking flag but isn't supposed to do anything on HOS.
Needed for games such as AC:NH.
The `Auto` option automatically selects a region based on the currently selected system language.
Co-Authored-By: Timotej Leginus <35149140+timleg002@users.noreply.github.com>
As part of this commit, a new preference category for debug settings is being introduced. All future settings only relevant for debugging purposes will be put there. The category is hidden on release builds.
Host synchronization of a guest texture with a different guest format represents a valid use case where the host doesn't support the guest format and conversion to a host-compatible format must be performed. The issue is most evident on Mali GPUs, as they don't support BCn texture formats thus needing manual decoding before submission. It was disabled by mistake in a previous commit, this commit re-enables it.
Unindexed quad draws were broken when multiple draw calls were done on the same vertex buffer, with a non-zero `first` index.
Indexed quad draws also suffered from the same issue, but was never encountered in games.
This commit fixes both cases by accounting for the `first` drawn index when generating conversion index buffers.
TIPC is a much lighter layer ontop of the Horizon IPC system than CMIF and is used by SM in 12.0.0+. This implementation is slightly hacky since it doesn't really keep a seperation between the underlying kernel IPC stuff and userspace like CMIF/TIPC, this should be fixed eventually, probably together with an IPC dispatch rewrite to avoid the mess of frozen maps.
Tested with Hentai Uni, which now crashes needing 'ldr:ro'.
Tapping anything in titles that supported touch (such as Puyo Puyo Tetris or Sonic Mania) wouldn't work due to the first touch point never being removed from the screen, it is supposed to be removed after a 3 frame delay from the touch ending.
This commit introduces a mechanism to "time-out" touch points which counts down during the shared memory updates and removes them from the screen after a specified timeout duration.
Certain titles depend on HID LIFO entries being written out at a fixed frequency rather than on actual state change, not doing this can lead to applications freezing till the LIFO is filled up to maximum size, this behavior is seen in Super Mario Odyssey. In other cases such as Metroid Dread, the game can run into race conditions that would lead to crashes, these were worked around by smashing a button during loading prior.
This commit introduces a thread which sleeps and wakes up occasionally to write LIFO entries into HID shared memory at the desired frequencies. This alleviates any issues as it fills up the LIFO instantly and correctly emulates HID Shared Memory behavior expected by the guest.
Co-authored-by: Narr the Reg <juangerman-13@hotmail.com>
It was determined that deadlocks inside `KThread::UpdatePriorityInheritance` would not only arise from the first level of locking with `waitingOn->waiterMutex` but also the second level of locking with `nextThread->waiterMutex` which has now also been fixed to fallback when facing contention.
PR #1758 introduced a bug where the game list would be entirely loaded every time the app was opened. This commit addresses that issue, which was caused by the `version` member of the cached game list being serialized to file (although incorrectly) but never actually read back when deserializing.
* Remove `package` from manifest and from activity prefixes, gradle `namespace` will be used instead
* Removed deprecated `android.support.PARENT_ACTIVITY` metadata
entries
* Make `MainActivity` and `SettingsActivity` launched in `singleTop` mode to avoid unnecessary activity restarts while navigating the app
Using `__attribute__((packed))` doesn't work in new NDKs when a struct contains 128-bit integer members, likely because of a ndk/compiler bug. We now enclose the requiring structs in `#pragma pack` directives to tightly pack them.
Since the blit engine itself samples from pixel corners and the helper shader from pixel centres teh src coordinates need to be adjusted to avoid the helper shader wrapping round on the final column.
We previously missed the hades pass for attribute conversion leading to crashes when games would attempt to use such an attribute. The hades pass for this isn't a proper fix however as it modifies the IR directly and will break if any of the previous stages in the pipeline change. Enable it to allow for games using them to at least have a chance at working. In the long term the pass will be reworked on the hades side to avoid modifying the IR in a way that can't be undone.
This vertex state must only be present for the last pipeline stage that touches vertices, if it is present for other stages it could result in incorrect behaviour like performing TFB in the fragment shader or flipping device coordinates twice.
As the code was before, if we had a shader that was disabled and enabled again after without being invalidated the pipeline stage would stay disabled and break rendering.
We previously only supported non-indexed quads. Support for this is implemented by converting the index buffer at record time and pushing the result into the megabuffer, which is then used as the index buffer in the final draw command.
The `Allocate` method allocates the given amount of space in a megabuffer chunk, returning a descriptor of the allocated region. This is useful for situations where you want to write directly to the megabuffer, avoiding the need for an intermediary buffer.
Entirely rewrites the engine and interconnect code to take advantage of the subpixel and OOB blit support offered by the blit helper shader. The interconnect code is also cleaned up significantly with the 'context' naming being dropped due to potential conflicts with the 'context' from context lock
It is desirable for us to use a shader for blits to allow easily emulating out of bounds blits and blits between different swizzled colour formats. The helper shader infrastructure is designed to be generic so it can be reused by any other helper shaders that we may need in the future.
These sometimes spuriously occur in games during transitions, to avoid crashing during them just use the null texture if they occur and log an error log
The constant destruction and creation of `BufferView`s in cbuf-heavy games showed up as a large chunk of the profiler. Fix this by taking advantage of the fact that constant buffer `BufferView`s are never deleted and always kept around in the cache to just return a pointer to them in the cache.
Currently we heavily thrash the heap each draw, with malloc/free taking up about 10% of GPFIFOs execution time. Using a linear allocator for the main offenders of buffer usage callbacks and index/vertex state helps to reduce this to about 4%
Certain titles can have a display frames out of order due to not waiting on the copy from the final RT to the swapchain image to occur. Although `PresentFrame` does wait on the syncpoint, that isn't enough to ensure the source texture is up-to-date due to us signalling syncpoints early.
By waiting on the swapchain texture after the copy is submitted, we now implicitly wait on the source texture's cycle to be signalled thus waiting on the frame to be done which fixes the issue.
After the introduction of workahead a system to hold a single large megabuffer per submission was implemented, this worked fine for most cases however when many submissions were flight at the same time memory usage would increase dramatically due to the amount of megabuffers needed. Since only one megabuffer was allowed per execution, it forced the buffer to be fairly large in order to accomodate the upper-bound, even further increasing memory usage.
This commit implements a system to fix the memory usage issue described above by allowing multiple megabuffers to be allocated per execution, as well as reuse across executions. Allocations now go through a global allocator object which chooses which chunk to allocate into on a per-allocation scale, if all are in use by the GPU another chunk will be allocated, that can then be reused for future allocations too. This reduces Hollow Knight megabuffer memory usage by a factor 4 and SMO by even more.
Accesses to unset entries is now clearly defined as returning a 0'd out value, the prior behavior would be to optimize sets for border segments to use L2 atomicity when the specific segment had no L1 entries set. This would lead to any future lookups of offsets within the same L2 segment but a different L1 entry to incorrectly return an inaccurate value as the only prior guarantee was that lookups after setting a segment would return the same value as was set but lacked the guarantee for unset segments to also consistently return unset values.
This could lead to issues in practical usages such as the `BufferManager` lookups returning the existence of a `Buffer` at a location falsely even though the segment was never set to the value, this was problematic as raw pointers were utilized and bound checks would lead to a segmentation fault.
This commit fixes this issue by introducing this guarantee and refactoring the class accordingly, it also deletes the `Set` method for setting a single entry as the meaning is ambiguous and it's functionality was more akin to the past guarantee and no longer makes sense.
Co-authored-by: PixelyIon <pixelyion@protonmail.com>
We would always write all L1 entries that correspond to an L2 entry, even if setting an input range ended before that. This would effectively reduce the atomicity of the segment table to that of the L2 range and lead to breaking API guarantees by returning entirely wrong segment values for a lookup covering a region that was overwritten.
It was determined that `RangeTable` was too ambiguous of a name as it could be interpreted to be holding ranges rather than looking them up, to avoid confusion the terminology has been changed to `range` to `segment`. As "segment table" is more clear in describing that it is a table comprised of descriptors regarding segments and it avoids any overlaps with terminology concerning "pages" which would be overly specific for this data structure or the ambiguous "ranges".
The PI CAS in `MutexUnlock` ends up loading `basePriority` rather than `priority` which could lead to an infinite CAS loop when `basePriority` doesn't equal to `priority` and the `highestPriorityThread`'s priority is lower than `basePriority`.
It was determined that `Texture::SynchronizeGuest`'s `TextureBufferCopy` had races that were exposed by the introduction of the cycle waiter thread, the synchronization did not take place under a locked context so the texture could be mutated at any point in addition to the destructor not being run during `FenceCycle::Wait` due to `shouldDestroy` being `false`.
This commit fixes the issue by making `SynchronizeGuest` entirely blocking as all usages of the function required blocking semantics regardless so it would be pointless to retain its async nature while solving any races that may arise from it being async.
Co-authored-by: Billy Laws <blaws05@gmail.com>
Since we don't call `SynchronizeHost` on source buffers which are GPU dirty, their mirrors will be out of date. The backing contents of this source buffer's region in the new buffer will be incorrect. By copying from the backing directly, we can ensure that no writes are lost and that if the newly created buffer needs to turn GPU dirty during recreation no copies need to be done since the backing is as up to date as the mirror at a minimum.
The code is much simpler to reason about when reading the code as it doesn't require evaluating all the potential edge cases of trap handlers in different states. It should be noted that this should not change behavior in any meaningful way, at most it can prevent a minor race where the protection could be upgraded after being downgraded by the signal handler leading to a redundant trap.
Two issues exist with locking of `KThread::waiterMutex`:
* It was not always locked when accessing waiter members such as `waitThread`, `waitKey` and `waitTag` which would lead to a race that could end up in a deadlock or most notably a segfault inside `UpdatePriorityInheritance`
* There could be a deadlock from `UpdatePriorityInheritance` locking `waiterMutex` of a thread and waiting to get the owner's `waiterMutex` while on another thread `MutexUnlock` holds the owner's `waiterMutex` and waits on locking the `waiterMutex` held by `UpdatePriorityInheritance`
This commit fixes both issues by adding appropriate locking to all locations where waiter members are accessed in addition to adding a fallback mechanism inside `UpdatePriorityInheritance` that unlocks `waiterMutex` on contention to avoid a deadlock.
The condition for exiting the CAS loops is incorrect in several places which leads to additional loops, while this doesn't make the behavior incorrect it does lead to redundant iterations.
Co-authored-by: Billy Laws <blaws05@gmail.com>
A substantial amount of time would be spent on creation/destruction of `VkDescriptorSet` which scales on titles doing a substantial amount of draws with bindings, this leads to poor performance on those titles as the frametime is dragged down by performing these tasks while they repeatedly create descriptor sets of the same layouts.
This commit fixes it by pooling descriptor sets per-layout in a dynamically resizable pool and keeping them around rather than destroying them after usage which leads to the vast majority of cases not requiring a new descriptor set to even be created. It leads to significantly improved performance where it would otherwise be spent on redundant destruction/recreation or push descriptor updates which took a substantial amount of time themselves.
Additionally, the `BaseDescriptorSizes` were not kept up to date with all of the descriptor types, it led to no crashes on Adreno/Mali as they were purely used for size calculations on either driver but has been corrected to avoid any future issues.
A substantial amount of time is spent destroying dependencies for any threads waiting or polling `FenceCycle`s, this is not optimal as it blocks them from moving onto other tasks while destruction is a fundamentally async task and can be delayed.
This commit solves this by introducing a thread that is dedicated to waiting on every `FenceCycle` then signalling and destroying all dependencies which entirely fixes the issue of destruction blocking on more important threads.
Buffer lookups are a fairly expensive operation that we currently spend `O(log n)` on the simplest and most frequent case of which is a direct match, this is a very frequent operation where that may be insufficient. This commit optimizes that case to `O(1)` by utilizing a `RangeTable` at the cost of slightly higher insertion/deletion costs for setting ranges of values but these are minimal in frequency compared to lookups.
A data structure that can represent the same value for a range of addresses (pages) is required for fast lookup in certain cases. This commit implements a near optimal data structure for mass insertion and O(1) lookup of range-based data, this is achieved using the host MMU and implementing multiple levels of atomicity for the ranges.
It should be noted that the table is limited to two levels but can be extended to a variable amount of ranges in the future, it was determined that additional levels of ranges can be beneficial for performance depending on the specific use-case.
Adreno drivers have certain errata which leads to Vulkan Push Descriptors to be broken on them in certain cases which leads to a descriptor set update being swallowed. This has been worked around by disabling push descriptors on Adreno drivers, this may lead to reduced performance on certain titles which frequently bind new descriptors.
Any semaphore releases are implicit synchronization events that can be utilized by the guest to pick up that the GPU has executed till a certain point and therefore we must submit all prior work accordingly.
DMA copies utilized `SubmitWithFlush` instead of `Submit`, this is not required and incurs significant additional synchronization penalties which will no longer be required.
We want to avoid blocking on surface creation unless necessary, this commit doesn't wait on the creation of the surface as it default initializes the value which'll generally be `Identity` or the transformation of the previous surface if it was lost.
Co-authored-by: Billy Laws <blaws05@gmail.com>
The V-Sync `KEvent` would be used by the presentation thread prior to construction leading to dereferencing an invalid value, this has been fixed by changing the order of construction to move the construction of the presentation thread after the V-Sync event.
The `TrapRegions` function performed a page-out on any regions that were trapped as read-only, this wasn't optimal as it would tie them both into the same operation while Buffers/Textures require to protect then synchronize and page-out. The trap was being moved to after the synchronize to get around this limitation but that can cause a potential race due to certain writes being done after the synchronization but prior to the trap which would be lost. This commit fixes these issues by splitting paging out into `PageOutRegions` which can be called after `TrapRegions` by any API users.
Co-authored-by: Billy Laws <blaws05@gmail.com>
`NCE::TrapRegions` was a bit too overloaded as a method as it implicitly trapped which was unnecessary in all current usage cases, this has now been made more explicit by consolidating the functionality into `NCE::CreateTrap` which handles just creation of the trap and nothing past that, `RetrapRegions` has been renamed to `TrapRegions` and handles all trapping now.
Co-authored-by: Billy Laws <blaws05@gmail.com>
Similar to `Buffer`s, `Texture`s suffered from unoptimal behavior due to using atomics for `DirtyState` and had certain bugs with replacement of the variable at times where it shouldn't be possible which have now been fixed by moving to using a mutex instead of atomics. This commit also updates the API to more closely match what is expected of it now and removes any functions that weren't utilized such as `SynchronizeGuestWithBuffer`.
Having a single variable denoting the exact state of a buffer and the operations that could be performed on it was found to be too restrictive, it's now been expanded into an additional `BackingImmutability` variable but due to these two. We can no longer use atomics without significant additional complexity so all accesses to the state are now mediated through `stateMutex`, a mutex specifically designed for tracking the state.
While designing the system around `stateMutex` it was determined to be more efficient than atomics as it would enforce blocking far less than it would generally have been compared to if the regular atomic fallback of locking the main resource lock which is locked for significantly longer generally.
Co-authored-by: PixelyIon <pixelyion@protonmail.com>
As a performance sensitive part of code, the NCE Trapping API benefits from having tracing and it helps us better determine where guest code is spending its time for more targeted optimizations.
The lifetime of the `this` pointer in the trap callbacks could be invalid as the lifetime of the underlying `Buffer`/`Texture` object wasn't guaranteed, this commit fixes that by passing a `weak_ptr` of the objects into the callbacks which is locked during the callbacks and ensures that a destroyed object isn't accessed.
Co-authored-by: Billy Laws <blaws05@gmail.com>
The `CommandExecutor`'s `MegaBuffer` was not being updated with the latest `FenceCycle` on being flushed in `SubmitWIthFlush`, this led to the megabuffer being overwritten prior to its GPU-side usage being complete. This commit fixes that by replacing the cycle to the latest cycle and prevents any races that occurred prior.
`FindOrCreate` ended up being monolithic function with poor readability, this commit addresses those concerns by refactoring the function to split it up into multiple member functions of `BufferManager`, while some of these member functions may only have a single call-site they are important to logically categorize tasks into individual functions. The end result is far neater logic which is far more readable and slightly better optimized by virtue of being abstracted better.
In certain cases the move constructor may not suffice and the move assignment operator is required, this commit implements that and moves to using a pointer for storing the `resource` member rather than a reference as its semantics matched what we desired more and allowed for assignment of the `resource`.
It was determined that `FindOrCreate` has several issues which this commit fixes:
* It wouldn't correctly handle locking of the newly created `Buffer` as the constructor would setup traps prior to being able to lock it which could lead to UB
* It wouldn't propagate the `usedByContext`/`everHadInlineUpdate` flags correctly
* It wouldn't correctly set the `dirtyState` of the buffer according to that of its source buffers
The condition for `setDirty` in the dirty state CAS was inverted from what it should've been resulting in synchronizing incorrectly, this commit fixes the condition to correct synchronization.
The formats of the textures involved in a texture were checked for equality, this broke certain copies as the presentation engine would invoke copies between textures of different yet compatible formats.
Co-authored-by: PixelyIon <pixelyion@protonmail.com>
`ContextLock` had unoptimal semantics in the form of direct access to the `isFirst` member which wasn't clearly defined, it's now been broken up into function calls `IsFirstUsage` and `OwnsLock` with explicit move semantics and a function for releasing the lock.
Co-authored-by: PixelyIon <pixelyion@protonmail.com>
The position at which we call submit is a significant factor in performance and we did so at the end of PBs (PushBuffers), this isn't optimal as there could be multiple PBs queued up that would benefit from being in the same submission. We now delay the submission of the workload till we run out of PBs.
A buffer that's attached to a context could be coalesced into a larger buffer which isn't attached, this would break as it wouldn't keep the buffer alive till the end of the associated context. To fix this if any source buffers are attached then the resulting coalesced buffer is also attached now.
The CAS condition for KThread PI was inverted which lead to entirely incorrect behavior for CAS conditions which while it might work in the vast majority of cases would lead to significantly inaccurate behavior.
The lock callback would `continue` which would end up skipping over the current item as it applied to the inner loop rather than the outer loop as intended. This has now been fixed by using `break` and a check instead.
The buffer's non-blocking behavior could lead to an invalid state where the dirty state doesn't adequately represent the buffer's true state, the check has now been moved inside the CAS loop as its behavior changes depending on the dirty state. In addition, `SynchronizeGuest` returns a boolean denoting if the synchronization was successful now to make code flows depending on non-blocking synchronization cleaner.
`SynchronizeGuest` could only set the dirty state to `Clean` which was redundant since calls to it from inside the write trap handler would set it to `CpuDirty` directly after, this fixes that by doing it inside the function when necessary.
The trap callbacks did not wait on the `Texture` to complete synchronization to the guest, this resulted in races where the contents written to the texture would be overwritten by the synced content. This commit fixes that by waiting on the fences at the end of the trap callback.
The lifetime of `TextureView` objects wasn't correctly managed as they weren't being attached the the `FenceCycle` in `AttachTexture`, this led to them getting deleted and causing all sorts of UB.
The flush callbacks inside `CommandExecutor` weren't being called prior to submission as they should've been, this fixes that by calling them. It additionally removes the requirement to manually flush Maxwell3D at the end of `ChannelGpfifo` pushbuffers as it's a flush callback and will automatically be called by `Submit`.
Co-authored-by: Billy Laws <blaws05@gmail.com>
Any work that was done in a `ChannelGpfifo` pushbuffer needs to be submitted at the end of it, if it isn't done then the work might incorrectly be not done till the next submission. This commit fixes it by calling `CommandExecutor::Submit` at the end of a pushbuffer, submitting any buffers that would've been left over.
Co-authored-by: Billy Laws <blaws05@gmail.com>
Certain submissions might not utilize megabuffering but reserve a `MegaBuffer` regardless, this is not optimal since it can inflate the allocations and waste memory. This commit addresses the issue by eliding the allocation given the current submission doesn't utilize them.
If a `FenceCycle` isn't attached then `PollFence` returned `false` while it should return if the buffer has any concurrent GPU usages in flight, this has now been fixed by returning `true` in those cases.
Certain resources can be attached to an empty `Submit` with no nodes, this can cause it to become a false dependency and not be removed till the next non-empty submission. This has now been fixed by doing a reset regardless of if any nodes exist.
The GPU inline copy callback was broken for `Buffer::Write` as it wasn't always called when it needed to be and didn't handle attaching of the buffer to the executor which would cause it to be unlocked. This commit addresses both of these issues, it introduces a `AttachLockedBuffer` method to attach an already locked buffer to the executor.
The FPS is implicitly bound to the refresh rate due to the timestamp being that of the presentation time, this leads to a misleading FPS figure for disabled frame throttling. It has now been fixed by using the frame submission time rather than the presentation time when frame throttling is disabled and to make this more apparent the color of the OSD FPS has been changed.
All `Packed` formats have their components stored in the opposite ordering to the label, this was not followed for `IsAdrenoAliasCompatible` prior and the ordering has now been flipped.
A deadlock was caused by holding `trapMutex` while waiting on the lock of a resource inside a callback while another thread holding the resource's mutex waits on `trapMutex`. This has been fixed by no longer allowing blocking locks inside the callbacks and introducing a separate callback for locking the resource which is done after unlocking the `trapMutex` which can then be locked by any contending threads.
The `end` pointer for `interval` was incorrectly calculated as `interval.data() + interval.size_bytes()` which would be incorrect when the interval span type is not `u8` as the pointer derived from `interval.data()` would be a pointer to the span type rather than a byte pointer and be subject to arithmetic of that object's size rather than in terms of a byte.
We generally don't need to lock the `Texture`/`Buffer` in the trap handler, this is particularly problematic now as we hold the lock for the duration of a submission of any workloads. This leads to a large amount of contention for the lock and stalling in the signal handler when the resource may be `Clean` and can simply be switched over to `CpuDirty` without locking and utilizing atomics which is what this commit addresses.
We utilized a `FenceCycle` to keep track of if the buffer was mutable or not and introduced another cycle to track GPU-side requirements only on fulfillment of which could the buffer be utilized on the host but due to the recent change in the behavior this system ended up being unoptimal.
This commit replaces the cycle with a boolean tracking if there are any usages of the resource on the GPU within the current context that may prevent it from being mutated on the CPU. The fence of the context is simply attached to the buffer based off this which was allowed as the new behavior of buffer fences matches all the requirements for this.
An atomic transactional loop was performed on the backing `std::shared_ptr` inside `BufferView`/`TextureView`'s `lock`/`LockWithTag`/`try_lock` functions, these locks utilized `std::atomic_load` for atomically loading the value from the `shared_ptr` recursively till it was the same value pre/post-locking.
This commit abstracts the locking functionality of `TextureView`/`BufferDelegate` into `LockableSharedPtr` to avoid code duplication and removes the usage of `std::atomic_load` in either case as it is not necessary due to the implicit memory barrier provided by locking a mutex.
`PresentationEngine` and `GraphicBufferProducer` methods that utilized textures for the surface utilized the `Texture` type rather than the `TextureView` type, this was never correct but at the time of authoring this code `TextureView` was not finalized and in a major flux which is why it was not utilized and `Texture` was utilized instead. Now that is is far more stable, it has been replaced with `TextureView`.
We want to block on the host thread during presentation while the host surface isn't present to implicitly pause the game, this can end up being fairly costly as it involves locking the `PresentationEngine` mutex which can lead to a lot of contention with the presentation thread. This fixes the issue by polling if there is a surface and only if there isn't then doing the wait as it isn't mandatory to wait always, we'll eventually run into the guest thread stalling.
Newer versions of the Deko3D homebrew were crashing due to this check and it was discovered that the check was incorrect and rather than comparing the `NvSurface` what had to be compared was the `GraphicBuffer` associated with the slot directly.
Co-authored-by: lynxnb <niccolo.betto@gmail.com>
The copyright headers for external project such as yuzu/Ryujinx were inconsistent in ordering, Skyline should always be the first item in the list. In addition, they didn't always link to the project's GitHub which has also been fixed.
Multiple threads concurrently accessing the `TextureManager`/`BufferManager` (Referred to as "resource managers") has a potential deadlock with a resource being locked while acquiring the resource manager lock while the thread owning it tries to acquire a lock on the resource resulting in a deadlock.
This has been fixed with locking of resource manager now being externally handled which ensures it can be locked prior to locking any resources, `CommandExecutor` provides accessors for retrieving the resource manager which automatically handles locking aside doing so on attachment of resources.
GPU resources have been designed with locking by fences in mind, fences were treated as implicit locks on a GPU, design paradigms such as `GraphicsContext` simply unlocking the texture mutex after attaching it which would set the fence cycle were considered fine prior but are unoptimal as it enforces that a `FenceCycle` effectively ensures exclusivity. This conflates the function of a mutex which is mutual exclusion and that of the fence which is to track GPU-side completion and led to tying if it was acceptable to use a GPU resource to GPU completion rather than simply if it was not currently being used by the CPU which is the function of the mutex.
This rework fixes this with the groundwork that has been laid with previous commits, as `Context` semantics are utilized to move back to using mutexes for locking of resources and tracking the usage on the GPU in a cleaner way rather than arbitrary fence comparisons. This also leads to cleaning up a lot of methods that involved usage of fences that no longer require it and therefore can be entirely removed, further cleaning up the codebase. It also opens the door for future improvements such as the removal of `hostImmutableCycle` and replacing them with better solutions, the implementation of which is broken at the moment regardless.
While moving to `Context`-based locking the question of multiple GPU workloads being in-flight while using overlapping resources came up which brought a fundamental limitation of `FenceCycle` to light which was that only one resource could be concurrently attached to a cycle and it could not adequately represent multi-cycle dependencies. `FenceCycle` chaining was designed to fix this inadequacy and allows for several different GPU workloads to be in-flight concurrently while utilizing the same resources as long as they can ensure GPU-GPU synchronization.
If we want to allow submitting multiple pieces of work to the GPU at once while still requiring CPU synchronization, we'll need to track all past fence cycles associated with a resource alongside the current one. To solve this the concept of chaining fences has been introduced, fences from past usages can be chained to the latest fence which'll then recursively forward operations to chained fences.
This change also ends up mandating a move away from `FenceCycleDependency` as it would prevent fences from concurrently locking the same resources which is required for chaining to work as two fences being chained fundamentally means they're locking the same resources. The `AtomicForwardList` is therefore used as the new container.
An implementation of a singly-linked list with atomic access to allow for lock-free access semantics, it eliminates the requirement for a mutex which can introduce additional consideration for synchronization.
Resources on the GPU can be fairly convoluted and involve overlaps which can lead to the same GPU resources being utilized with different views, we previously utilized fences to lock resources to prevent concurrent access but this was overly harsh as it would block usage of resources till GPU completion of the commands associated with a resource.
Fences have now been replaced with locks but locks run into the issue of being per-view and therefore to add a common object for tracking usage the concept of "tags" was introduced to track a single context so locks can be skipped if they're from the same context. This is important to prevent a deadlock when locking a resource which has been already locked from the current context with a different view.
We do not want to allow saving of user data on unsigned builds as they don't have a stable signature and will not properly handle reinstallation. This can lead to a situation where the user has to resort to complex techniques to completely uninstall the package such as ADB or calling into PM directly.
We currently present all frames synchronously on the thread that calls into SurfaceFlinger functions, this is unoptimal as it doesn't match guest behavior which can lead to delaying the guest from working on the next frame. This commit queuing up frames to non-blocking and handles all waiting then presenting the frame on a dedicated thread.
We utilize `pthread_setname_np` to set the thread names but didn't check for any errors which resulted in the `Skyline-Choreographer` and `ChannelCmdFifo` not having proper names as they exceeded the 16 character limit on thread names for the pthread function. This has now been fixed by changing the names and introducing error checking to invocations of this function.
All our normal alignment functions are designed to only handle power of 2 (`POT`) multiples as we only align or check alignment to `POT` multiples but there are cases where this is not possible and we deal with `NPOT` multiples which is why this function is required.
We waited on the host GPU after `Execute` but this isn't optimal as it causes a major stall on the CPU which can lead to several adverse effects such as downclocking by the governor and losing the opportunity to work in parallel with the GPU.
This has now been fixed by splitting `Execute`'s functionality into two functions: `Submit` and `SubmitWithFlush` which both execute all nodes and submit the resulting command buffer to the GPU but flushing will wait on the GPU to complete while the non-flush variant will not wait and work ahead of the GPU.
We need move-assignment semantics to viably utilize these objects as class members, they cannot be replaced without move-assign (or copy-assign but that is undesirable here). This commit fixes that by introducing a move assignment operator to them while making the `slot` a pointer which has the necessary nullability semantics.
This change lets items get the updated position of their view holder in the adapter. Fixes an issue where the position of items was not updated after being removed from a `SelectableGenericAdapter`.
This preference launches `GpuDriverActivity` for managing custom gpu drivers. When the device has an incompatible GPU, the preference will be disabled and greyed out.
The activity adds the following functionalities:
* Lists installed drivers
* Allows the user to install new drivers, or remove installed ones
* Allows the user to select the driver that will be used by the emulator
At some point we will call Submit within draws or constant buffer updates, to avoid any infinite recursion mark draw/cbuf pending as false before performing any operation
The previous name was chosen as an afterthought and didn't clearly indicate what the purpose of the class is. We needed a separate, simple class without delegates members (like PreferenceSettings), so that its fields can be easily accessed via JNI to get settings values from native code.
The `Settings` class now has a pure virtual `Update` method, and uses inheritance over template specialization for platform-specific behavior override.
A `Setting` delegate class has been introduced, holding the raw value of the setting and adding support for registering callbacks to that setting. Callbacks will then be called when the value of that setting changes.
As a result of this, raw setting values have been made accessible through pointer dereference semantics.
SharedPreferences will be partially swapped out in the future to support per-game settings. In the meantime, make it clear from which class settings are coming from.
Settings are now shared to the native side by passing an instance of the Kotlin's `Settings` class. This way the C++ `Settings` class doesn't need to parse the SharedPreferences xml anymore.
Mali GPU drivers utilize the `ppoll()` syscall inside `waitForFences` which isn't correctly restarted after a signal, which we can receive at any time on a guest thread. This commit fixes that by recursively calling the function on failure till it succeeds or returns an unexpected error.
Co-authored-by: PixelyIon <pixelyion@protonmail.com>
Co-authored-by: Billy Laws <blaws05@gmail.com>
These applets are used by applications to display a custom error message to the user. Both the error message and the detailed error message are printed to the error log.
Co-authored-by: lynxnb <niccolo.betto@gmail.com>
This conforms to the C++ 'Allocator' named requirement allowing it to be used with any STL type and allows drastically reducing allocation times in cases which are suited for linear allocation.
Certain non-indexed quad draws would mistakenly take the indexed quad path because of the assumption that they would not have a bound index buffer. This resulted in a crash for most games using quads due to a faulty exception `Indexed quad conversion is not supported`, when in fact they were not using indexed quads.
Co-authored-by: PixelyIon <pixelyion@protonmail.com>
Co-authored-by: Billy Laws <blaws05@gmail.com>
This commit implements several key optimisations in megabuffering that are all inherently interlinked.
- Megabuffering is moved from per-buffer to per-view copies, this makes megabuffering possible for small views into larger underlying buffers which is often the case with even the simplest of games,
- Megabuffering is no longer the default option, it is only enabled for buffer views that have had inline GPU writes applied to them in the past as that is the only case where they are beneficial. In any other case the cost of copying, even with a 128KiB limit can be significant.
- With both of these changes, there is now possibility for overlapping views where one uses megabuffering and one does not. In order to allow GPU inline writes to work consistently in such cases a system of 'host immutability' has been implemented, when a buffer is marked as host immutable for a given cycle, all writes to the buffer from that point to the point the cycle is signalled will be performed on the GPU, ensuring that the backing contents are correctly sequenced
Has the same guarantees of pointer stabilty while also being significantly faster in cases where a buffer has thousands of views. This is the case in RE4 and this change leads to an almost 1000% performance improvement in that game.
Uses an API found through RE since none of the AOSP APIs work, additionaly the code for setting RR was consolidated to a single function that can be ran after all display updates.
We currently have a global `MegaBuffer` instance that is shared across all channels, this is very problematic as `MegaBuffer` fundamentally works like a state machine with allocations (especially resetting/freeing) and is thread-specific. Therefore, we now have a pool of several `MegaBuffer`s which is allocated from by the `CommandExecutor` and kept channel specific as a result which also limits its usage to a single thread, this allows for individually resetting or freeing any allocations.
There was a lot of redundant code in the `CommandScheduler` when the same functionality could be achieved with much shorter and cleaner code which this commit fixes. This includes no changes to the user-facing API and does not require any changes on the user side as a result.
Some games remap rendertargets or map them late which would lead to weird graphical bugs or crashes. Drop the caching since VMM lookup is fairly cheap anyway.
The `VkBufferImageCopy` offset calculations were wrong inside `CopyIntoStagingBuffer` as it multiplied the mip level's linear size by `levelCount` rather than `layerCount`. This led to substantial UB in games which called this function as it led to an overflow and resulted in writing to other areas of the buffer which caused major issues such as vertex/index buffer corruption and corresponding graphical glitches alongside likely being the cause of some crashes.
BC7 CPU decoding had the red and blue channels swapped around as it outputted a BGRA image after decoding while we expected an RGBA image to be produced. This should fix the colors of certain textures in titles such as Cuphead or Sonic Forces.
The syncpoint maximum value represents the maximum possible syncpt value at a given time, however due to PBs being submitted before max was incremented, for a brief moment of time this is not the case which could lead to crashes or other such behaviour if a game waits on the fence at the right moment.
We used a `FileProvider` for log sharing prior, this is no longer necessary since it comes under the `DocumentsProvider` now which can be utilized to share the log document directly.
Any documents with the same name existing in a directory that is copied to would cause an exception due to existing already, this fixes that by handling conflict resolution in those cases and automatically determining a file name that would avoid a conflict.
Previously a broken state value was returned from GetState that caused crashes in games using newer SDKs and NFP, correctly handle state now by updating it after initialisation.
We can't render to a 3D texture through a 3D view, we instead have to create a 2D array view into it and render to that. The texture manager previously didn't support having a different view type/layer count between a guest texture view and the underlying storage texture that is required to support this so that was also implemented by reading the view layer count from the dimensions depth instead if the underlying texture is 3D (and the view type is 2D array). Additionally move away from our own view type enum to Vulkan, inline with other guest texture member types.
Sampler anisotropy was made a required feature in an earlier commit due to its widespread availability but this was determined to be incorrect as certain Mali GPUs that can otherwise run 2D games in Skyline do not have this feature, while they are still not officially supported as this was the only roadblock to support them, it has now been made an optional feature.
`android:hasFragileUserData` was added in an earlier commit but then removed due to it not functioning because of signature checks. Now that signatures are consistent across builds, it has been readded and should now allow carrying data across CI and developer builds.
With the Skyline document provider, easy access to the internal directory is required which may be hard to navigate to through the system file manager. This adds an option in settings to directly open up the directory in the system file manager.
The URIs (Document ID + Root) of the Skyline `DocumentsProvider` was unoptimal as it wasn't relative to a base directory. This is required for opening a root without knowledge of the full path in advance, it is therefore cleaner to provide a uniform `ROOT_ID` in a companion class.
On Android 12 and above, files from an application's external storage directory cannot be accessed by the user. The only proper SAF-compliant way to solve this is to create a `DocumentProvider` which proxies access to internal storage accordingly.
Certain GPU vendors such as ARM's Mali do not have support for BCn textures whatsoever while other vendors such as AMD only have partial support (BC1-BC3). Most titles on the guest utilize BC textures and to address this on host GPUs without support for BCn, we need to decompress the texture on the CPU. This commit implements a CPU BCn texture decoder based off Swiftshader's BC decoder, it also adds the necessary infrastructure to have different formats for the `GuestTexture` and `Texture` objects.
The iterations of the inner loop for sector deswizzling was miscalculated as `SectorWidth * SectorHeight` while the result was correct at `32`, it should be determined by the amount of sector lines within a GOB i.e.: `(GobWidth / SectorWidth) * GobHeight`.
Support for mipmapped textures was not implemented which is fairly crucial to proper rendering of games as the only level that would load is the first level (highest resolution), that might result in a lot more memory bandwidth being utilized. Mipmapping also has associated benefits regarding aliasing as it has a minor anti-aliasing effect on distant textures.
This commit entirely implements mipmapping support but it does not extend to full support for views into specific mipmap levels due to the texture manager implemention being incomplete.
Maxwell DMA requires swizzled copies to/from textures and earlier it had to construct an arbitrary `GuestTexture` to do so but with the introduction of the cleaner API, this has become redundant which this commit cleans up and replaces with direct calls to the API with all the necessary values.
The API for texture swizzling is now more concrete and abstracted out from `GuestTexture`, this allows for neater usage in certain areas such as MaxwellDMA while having a `GuestTexture` wrapper as well allowing for neater usage in those cases.
The code itself has also been cleaned up slightly with all usage of `u32`s being upgraded to `size_t` as this is simply more efficient due to the compiler not needing to emulate wraparound behavior for integer types smaller than the processor word size.
The Fermi 2D engine implements both image blit and resolve operations, supporting subpixel sampling with both linear and point filtering.
Resolve operations are performed by sampling from the center of each pixel in order to resolve the final image from the MSAA samples
MSAA images are stored in memory like regular images but each pixels dimensions are scaled: e.g for 2x2 MSAA
```
112233
112233
445566
445566
```
These would be sampled with both duDx and duDy as 2 (integer part), resolving to the following:
```
123
456
```
Blit operations are performed by sampling from the corner of each pixel, scaling the image as one would expect.
This implementation isn't fully complete as Vulkan blit doesn't support some combinations which Fermi does, most notably between colour and depth stencil. These will be implemented properly at a later date, likely after the texture manager rework.
Out of Bounds Blit, used by some OpenGL games is also missing since supporting it requires texture aliasing, this will also be supported after the texture manager rework.
Co-authored-by: Billy Laws <blaws05@gmail.com>
Certain writes during swizzling went out of bounds due to incorrect `blockExtentY` calculation, the previous commit to fix this ended up breaking it further. This commit returns to the original commit's calculations with the proper addendum of a check for exact alignment with a GOB which is the case that was broken earlier.
The `GuestTexture::GetLayerStride` function was not always being utilized to retrieve the layer stride inside `Texture`, it would instead directly access the `guestTexture::layerStride` member. This is problematic as it may not be initialized and return `0` which would lead to a broken image copy.
Most engines have the capability to release a semaphore payload (or reduce in the case of GPFIFO) when a method is called or action is complete. Semaphores are used by games for both timing how long things take on GPU and waiting on resources so missing them can cause deadlocks or other related issues.
Textures can have more than one layer which we currently don't handle, all layers past the initial one will be filled with random data or 0s, leading to incorrect rendering. This has now been implemented now which fixes any titles which utilize array textures, such as "Super Mario Odyssey" or "Hatsune Miku: Project DIVA MegaMix".
The Maxwell3D RT layer count wasn't being set correctly as it has the same register as the depth values and is toggled between the two based on another register value.
The Maxwell GPU supports 3D textures which are tiled with the block-linear layout which didn't handle swizzling 3D textures correctly till now. This commit addresses that by implementing proper swizzling for 3D textures. Titles such as Cluster Truck and Super Mario Odyssey utilize 3D textures alongside a vast majority of other titles.
As per VMA docs: 'Allocation size returned in this variable may be greater than the size requested for the resource e.g. as VkBufferCreateInfo::size. Whole size of the allocation is accessible for operations on memory e.g. using a pointer after mapping with vmaMapMemory(), but operations on the resource e.g. using vkCmdCopyBuffer must be limited to the size of the resource.'
There were two issues here:
- If a skyline span was passed as a param then the 'T &object' version would be called, filling the span itself with random values rather than its contents
- Random numbers were repeated every call since independent_bits_engine copied generator state and thus it was never actually updated
This calculation for the amount of lines on the Y axis relative to the start of the last block was wrong and would instead determine the amount of lines to the last Y-axis GOB which wasn't accurate when padding was considered, this resulted in titles like Celeste having broken texture decoding (on a 1922x1082 texture) for the last ROB as most pixels would be masked out.
Certain titles such as BOTW trigger behavior to reuse an attachment within the same subpass, this caused an exception inside `RenderPassNode::AddAttachment` as it cannot find corresponding subpass for attachment. To fix this issue, we now assume that when it cannot find a subpass for an existing attachment, it is attached to the latest subpass and return the attachment.
Certain textures may be unaligned with a GOB's height of 8 lines, we already handle the case of being unaligned with a GOB's width of 64-bytes. This case occurs on titles such as SMO when going in-game.
The function now returns from a segmentation fault when a debugger is present, this allows the entire context to be intact which can allow the debugger to correctly pick up variables from all stack frames while it could not extrapolate most variables when trapped inside the signal handler without the values of all registers.
In the Maxwell 3D engine, instanced draws are implemented by repeating the exact same draw in sequence with special flag set in vertexBeginGl. This flag allows either incrementing the instance counter or resetting it, since we need to supply an instance count to the host API we defer all draws until state changes occur. If there are no state changes between draws we can skip them and count the occurences to get the number of instances to draw.
Implements register state that corresponds to the size of a single point sprite in Maxwell 3D, this is emitted by the shader compiler in the preamble but needs to be only applied if the input topology is a point primitive and it is invalid to set the point size in any other case.
Earlier texture locking design required the lock to be retained but since the introduction of `AttachTexture`, this no longer needs to be done. This being done caused deadlocks when the depth texture is sampled by the fragment shader while being bound as an RT since it would attempt to lock the texture again.
A basic `bcat:u` implementation to prevent titles such as "Kirby and the Forgotten Land" dependent on BCAT support from crashing due to the lack of an implementation.
This is a widely supported feature that games may require conditionally but due to it being supported on effectively all target devices, it was made mandatory. This is used by titles such as ARMS.
Improves the readability of the log and replaces the previously uninformative prefix of `operator()` due to being in a lambda with `Controller support`.
Maxwell3D has a register for linking the TIC/TSC index in bindless texture handles, this is used by games to implement bindless combined texture-sampler handles.
Implements `GraphicsEnvironment::ReadCbufValue` & `GraphicsEnvironment::ReadTextureType` with a framework of heterogeneous lookups for caching and callbacks for querying constant buffer or TIC values with validation checks for successive draws to ensure unique IR is generated.
The `descriptorSetWrites` being filled is now optional and the case of it being empty is handled correctly, this is done by certain titles such as ARMS and is entirely valid behavior. It should be noted that not doing this leads to errors in the guest due to invalid GPU state while working on the host GPU.
SVC `SignalToAddress` had a bug with the behavior of `SignalAndModifyBasedOnWaitingThreadCountIfEqual` which was entirely incorrect and led to deadlocks in titles such as ARMS that were dependent on it. This commit corrects the behavior and refactors both SVCs and moves their arbitration/waiting to inside the corresponding `KProcess` function rather than the SVC to avoid redundancies and improve code readability.
Filtering of validation logs is now extended beyond BCn formats and now covers other format which have their feature set misreported by the driver, this significantly drives down the amount of logs depending on the title.
Implements an algorithm to determine formats that can be aliased as views without needing `VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT`, this avoids spamming warning logs on view creation when the aliased formats will function in practice.
There was an oversight with exclusive subpasses which could lead to RPs with more than one subpass could be created even though one pass was exclusive, this oversight was not finishing the render pass at the end of `AddSubpass`. This could lead to a future subpass adding to the end of that RP even though it was intended to exclusively have a single subpass.
This case occurs in titles such as Celeste (in-game) and breaks rendering on GPUs that may require exclusive subpasses for proper functionality.
The Khronos Validation Layer can often generate warning/error logs due to our intentional breakage from Vulkan specification, these can occur several times a frame resulting in the logs being spammed and making it difficult to extract useful information out of logs. The scope of these logs has now been reduced with more general filtering and the introduction of specialized filtering to handle complex cases such as BCn hacks with `libadrenotools` on Adreno devices.
Descriptor set updates were broken on the non-push-descriptor path due to lifetime issues with VkDescriptorSetLayout's usage during the execution phase which entirely broke rendering on AMD/Mali GPUs due to them not supporting `VK_KHR_push_descriptor`.
This commit addresses that by moving the allocation of a descriptor set to outside the lambda and into the recording phase, it also simplifies the semantics and resources passed into the lambda by removing redundancies.
The Vulkan render pass cache was fundamentally broken since it was designed around the Render Pass Compatibility clause due to being designed for framebuffer compatibility initially. As this scope was extended to a general render pass cache, the amount of data in the key was not extended to include everything it should have. This commit introduces the missing pieces in the RP cache and simplifies the underlying code in the process.
The backing for shader data would implicitly be zero-initialized due to a `resize` on every shader parse, this was entirely unnecessary as we would overwrite the entire range regardless.
We avoid this by using statically allocated storage and a span over it containing the shader bytecode which avoids any unnecessary clear semantics without resorting to more complex solutions such as a custom allocator.
Implements a cache for storing `VkFramebuffer` objects with a special path on devices with `VK_KHR_imageless_framebuffer` to allow for more cache hits due to an abstract image rather than a specific one.
Caching framebuffers is a fairly crucial optimization due to the cost of creating framebuffers on TBDRs since it involves calculating tiling memory allocations and in the case of Adreno's proprietary driver involves several kernel calls for mapping and allocating the corresponding framebuffer memory.
There are a lot of cases of `VkImageView` being recreated arbitrarily due to it being tied to the ephemeral object `TextureView` rather than `Texture`, this commit flips that by storing all `VkImageView`s inside `Texture` with `TextureView` simply holding a copy of the handle to them. Additionally, this change results in stable `VkImageView` handles and helps in paving the path for framebuffer caching when `VK_KHR_imageless_framebuffer` is unavailable.
As we desire more accurate profiling data in certain circumstances, making the app explicitly profilable will allow for this, it will also remove the (annoying) prompt to do this in the Android Studio profiler.
Implements a cache for storing `VkRenderPass` objects which are often reused, they are not extremely expensive to create generally but this is a required step to build up to a framebuffer cache which is an extremely expensive object to create on TBDRs generally since it involves calculating tiling memory allocations and in the case of Adreno's proprietary driver involves several kernel calls for mapping and allocating the corresponding memory.
We run into a lot of successive subpasses with the exact same framebuffer configuration which we now exploit to avoid the creation of a new subpass due to the overhead involved with this. This provides significant performance boosts in certain cases due to the magnitude of difference in the amount of subpasses being created while providing next to no benefit in other cases.
The check for the fence cycle being the same as the current cycle was incorrectly inverted to be the opposite of what it should have been, leading to bugs.
The responsibility for synchronizing a texture and locking it is now on the `PresentationEngine` rather than the API-user as this'll allow more fine grained locking and delay waiting until necessary.
As we require a relaxed version of the Vulkan render pass compatibility clause for caching multi-subpass render passes, we now utilize a quirk to determine if this is supported which it is on Nvidia/Adreno while AMD/Mali where it isn't supported we force single-subpass render passes.
We found out that certain vendors such as Nvidia had a limitation on the global priority of a queue and requesting `VK_QUEUE_GLOBAL_PRIORITY_HIGH_EXT` would result in `VK_ERROR_NOT_PERMITTED_EXT`. A quirk has been introduced to supply the maximum supported global priority which is currently set on a per-vendor basis to avoid future crashes.
Implements a cache for storing `VkPipeline` objects which are fairly expensive to create and doing so on a per-frame basis was rather wasteful and consumed a significant part of frametime. It should be noted that this is **not** compliant with the Vulkan specification and **will** break unless the driver supports a relaxed version of the Vulkan specification's Render Pass Compatibility clause.
We can use inline push descriptors for writing to descriptor rather than allocating a descriptor set for a one time write and freeing it as this is rather inefficient while an inline push descriptor generally ends up being a direct `memcpy` on the driver side designed for this use-case.
We want Skyline to have the most favorable GPU scheduling possible due to low latency and high throughput requirements, we request high priority scheduling due to this reason.
This implements all Maxwell3D registers and HLE Vulkan state for Tessellation including invalidation of the TCS (Tessellation Control Shader) state during state changes.
Previously constant buffer updates would be handled on the CPU and only the end result would be synced to the GPU before execute. This caused issues as if the constant buffer contents was changed between each draw in a renderpass (e.g. text rendering) the draws themselves would only see the final resulting constant buffer.
We had earlier tried to fix this by using vkCmdUpdateBuffer however this caused significant performance loss due to an oversight in Adreno drivers. We could have worked around this simply by using vkCmdCopy buffer however there would still be a performance loss due to renderpasses being split up with copies inbetween.
To avoid this we introduce 'megabuffers', a brand new technique not done before in any other switch emulators. Rather than replaying the copies in sequence on the GPU, we take advantage of the fact that buffers are generally small in order to replay buffers on the GPU instead. Each write and subsequent usage of a buffer will cause a copy of the buffer with that write, and all prior applied to be pushed into the megabuffer, this way at the start of execute the megabuffer will hold all used states of the buffer simultaneously. Draws then reference these individual states in sequence to allow everything to work without any copies. In order to support this buffers have been moved to an immediate sync model, with synchronisation being done at usage-time rather than execute (in order to keep contents properly sequenced) and GPU-side writes now need to be explictly marked (since they prevent megabuffering). It should also be noted that a fallback path using cmdCopyBuffer exists for the cases where buffers are too large or GPU dirty.
As bindings weren't correctly handled due to the fact that `EmitSPIRV` would change the bindings, the shader module cache would not correctly function and have no cache hits in `find` and rather have them in `try_emplace` which negated any performance benefit of it. This has now been fixed by retaining the initial cache key for insertion into the cache while also storing the post-emit bindings and restoring them during a cache hit.
Implements caching of the compiled shader module (`VkShaderModule`) in an associative map based on the supplied IR, bindings and runtime state to avoid constant recompilation of shaders. This doesn't entirely address shader compilation as an issue since host shader compilation is tied to Vulkan pipeline objects rather than Vulkan shader modules, they need to be cached to prevent costly host shader recompilation.
This implements the first step of a full shader cache with caching any IR by treating the shared pointer as a handle and key for an associative map alongside hashing the Maxwell shader bytecode, it supports both single shader program and dual vertex program caching.
We desire the ability to hash and check equality of data across spans to use associative containers such as `std::unordered_map` with spans. The implemented functions provide an easy way to do that.
Mostly based off of yuzu's implementation, this will need to be extended in the future to open up a UI for configuring controllers according to the applications requirements.
As there was no check for the lack of a `GuestTexture`/`GuestBuffer`, it would lead to UB when a texture/buffer that had no guest such as the `zeroTexture` from `GraphicsContext` would be marked as dirty they would cause a call to `NCE::RetrapRegions` with a `nullptr` handle that would be dereferenced and cause a segmentation fault.
In certain situations such as constant buffer updates, we desire to use the guest buffer as a shadow buffer forwarding all writes directly to it while we update the host using inline buffer updates so they happen in-sequence. This requires special behavior as we cannot let any synchronization operations take place as they would break the shadow buffer, as a result, an external synchronization flag has been added to prevent this from happening.
It should be noted that this flag is not respected for buffer recreation which will lead to UB, this can and will break updates in certain cases and this change isn't complete without buffer manager support.
The offset of the view wasn't added to the `vkCmdUpdateBuffer`, this would cause the offset to be incorrect given the buffer was a view of a larger buffer that wasn't the start of it. This commit fixes that by adding the offset of the view to the buffer update.
We didn't call `MarkGpuDirty` on textures/buffers prior to GPU usage, this would cause them to not be R/W protected when they should be and provide outdated copies if there were any read accesses from the CPU (which are not possible at the moment since we assume all accesses are writes at the moment). This has now been fixed by calling it after synchronizing the resource.
The terminology "Non-Graphics pass" was deemed to be fairly inaccurate since it simply covered all Vulkan commands (not "passes") outside the render-pass scope, these may be graphical operations such as blits and therefore it is more accurate to use the new terminology of "Outside-RenderPass command" due to the lack of such an implication while being consistent with the Vulkan specification.
Previously constant buffer updates would be handled on the CPU and only the end result would be synced to the GPU before execute. This caused issues as if the constant buffer contents was changed between each draw in a renderpass (e.g. text rendering) the draws themselves would only see the final resulting constant buffer. Fix this by updating cbufs on the GPU/CPU seperately, only ever syncing them back at the start or after a guest side CPU write, at the moment only a single word is updated at a time however this can be optimised in the future to batch all consecutive updates into one large one.
We require certain buffers to only be on the host while being accessible through the same abstractions as a guest buffer as they must be interchangeable in usage.
We needed to block stack frame lookups past JNI code as Java doesn't follow the ARMv8 frame pointer ABI which leads to invalid pointer dereferences. Any JNI function that throws or handles exceptions must do this now or it may lead to a `SIGSEGV`.
Some games may pass empty TICs as inputs to shaders while not actually using them within the shader. Create an empty texture and pass this in instead when we hit this case, the nullDescriptor feature could be used but it's not supported by all devices so we chose to do it this way instead.
Skyline's `exception` class now stores a list of all stack frames during the invocation of the exception. These can later be parsed by the exception handler to generate a human-readable stack trace. To assist with more complete stack traces, `-fno-omit-frame-pointer` is now passed on debug builds which forces the inclusion of frames on function calls.
NCE is implicitly depended on by the `GPU` class due to the NCE Memory Trapping API so the destruction of it must take place after the destruction of the `GPU` class. Additionally, to prevent bugs the NCE destructor must set `staticNce` to `nullptr` as the signal handler will potentially access a destroyed instance of NCE otherwise.
Without this sRGB textures would be interpreted as RGB leading to colours being slighly off. The sRGB flag isn't stored as part of format word so we reuse the _pad_ field of it to store the flag for the switch case.
We don't want to actually exit the process as it'll automatically be restarted gracefully due to a timeout after being unable to exit within a fixed duration so we just want to infinite sleep during termination. This should fix issues where exiting any game would cause the app to force close after some time as exception signal handling would fail in the background, the app should stay open now and automatically restart itself when another game is loaded in.
A lot of logs are incomplete due to being unable to flush inside the signal handler, now we flush after any exceptions so that there is a guarantee of any exceptions being logged as this is crucial for proper debugging.
B5G6R5 isn't generally supported by the swapchain and the format is used for R5G6B5 with swapped R/B channels to avoid aliasing so we reverse that by using R5G6B5 as the underlying Vulkan format for the swapchain which should be automatically handled by the driver for any copies from B5G6R5 textures and the data representation should be the same as B5G6R5 with swapped R/B channels so not reporting the correct texture::Format should be fine.
The DMA engine is used to perform DMA buffer/texture copies directly on the GPU. It can deswizzle arbritary regions of input textures, perform component remapping and swizzle into output textures.
This impl only supports 1D buffer copies, 2D ones will come later.
If we have a Nx1x1 image then determining the type from dimensions will result in a 1D image being created thus preventing us from creating a 2D view. By using the image view type we can avoid this for textures from TICs since we know in advance how they will be used
This enforces that the depth RT outlives the draw, without this the depth RT could be freed while in active use by command executor leading to UAFs and crashes.
This was erroneously included while migrating from older code where stack creation was entirely handled with host constructs such as `mmap` directly to using `KPrivateMemory` to manage it, we would create a guard page with `mprotect` that the guest was unaware about and would cause a segfault when a guest accessed the extents of the stack as reported to the guest.
A partial implementation of the `GetThreadContext3` SVC, we cannot return the whole thread context as the kernel only stores the registers we need according to the ARMv8 ABI convention and so far usages of this SVC do not require the unavailable registers but all future usage must be monitored and potentially require extending the amount of saved registers.
The vibration device had to be set manually prior which led to it generally not being set at all even though a user might want vibration, this commit fixes that by making controller #0 use the built-in vibrator by default.
Any Skyline files that should have been user-accessible were moved from `/data/data/skyline.emu/files` to `/sdcard/Android/data/skyline.emu/files` as the former directory is entirely private and cannot be accessed without either adb or root. This made retrieving certain data such as saves or loading custom driver shared objects extremely hard to do while this can be trivially done now.
In some games such as SMO thousands of constant buffers are bound per frame which was causing an unreasonable number of lookups in both vmm and the buffer manager. Work around this by introducing a simple hashmap based cache, eviction is currently unsupported but not really necessary yet due to the small size of the buffers in the cache.
We cannot ignore accesses from the host to a region protected by the NCE Memory Trapping API, there's often access to regions which have overlap with a protected region unintentionally and those accesses need to be handled correctly rather than leading to a crash. This is done by implementing an additional signal handler `NCE::HostSignalHandler` to lookup any potential traps on a `SIGSEGV` and handle them correctly or when there isn't a corresponding trap raise a `SIGTRAP` when debugger is connected or delegate to `signal::ExceptionalSignalHandler` when it isn't.
To cut down memory usage we now page out memory that is RW trapped via the NCE memory trapping API, the callbacks are supposed to page in the memory. This behavior is backed up by Texture/Buffer syncing which would read the host copies of data and write it to the guest, by paging the corresponding data on the guest we're avoiding redundant memory usage.
The `FileDescriptor` class is a RAII wrapper over FDs which handles their lifetimes alongside other C++ semantics such as moving and copying. It has been used in `skyline::kernel::MemoryManager` to handle the lifetime of the ashmem FD correctly, it wasn't being destroyed earlier which can result in leaking FDs across runs.
Initially this commit was only intended to update LLVM but due to a compilation error on latest LLVM libcxx due to the C++ stdlib header `<algorithm>` being a transitive dependency that is no longer transitively included on the latest LLVM libcxx (as of https://reviews.llvm.org/D119667), this required changes in Skyline and Oboe which were done in https://github.com/google/oboe/pull/1521 and the submodule has been updated to include those changes.
These are mostly used in 3D games like SMO, support is still quite basic and synchronising block linear 3D texture will crash in most cases due to them being unimplemented.
Some games crash due to requiring an `audren` version greater than 7. The `audren` version can be increased without any issues as `audren` is stubbed and therefore the reported version doesn't matter.
Older Adreno proprietary drivers (5xx and below) will segfault while destroying the renderpass and associated objects if more than 64 subpasses are within a renderpass due to internal driver implementation details. This commit introduces checks to automatically break up a renderpass when that limit is hit.
We have support for overlapping buffers which allows us to merge a lot of smaller buffers located on a single page into a single larger buffer which allows for better performance. It additionally ensures that all host buffers match the alignment guarantees of the guest and adequately fulfill host alignment requirements.
This commit encapsulates a complex sequence of cascading changes in the process of supporting overlaps for buffers:
* We determined that it is impossible to resolve overlaps with multiple intervals per buffer within the constraints of each overlap being a contiguous view, support for multiple intervals was therefore dropped. The older buffer manager code was entirely reworked to be simpler due to only handling one interval per buffer with code now being based off `IntervalMap` but tailored specifically for buffers.
* During overlap resolution, the problem of how existing views into the buffer being recreated would be updated, it had to be replaced with a larger buffer that could contain all overlaps and all existing views would need to be repointed to it. This was addressed by a buffer owning all views to itself, we could automatically recalculate the offset of all views and update the buffers with it.
* We still needed to update usage of existing views which was done by handling all access (such as inside a recorded draw) to buffer view properties via `BufferView::RegisterUsage` which dispatches a callback with the view and the corresponding backing buffer. This callback can be stored and called during overlap resolution with the new buffer.
* We had issues with lifetime of the buffer with the handle-like semantics of `BufferView` introduced in the last buffer-related commit, if we updated the view to be owned by a new buffer we'd need to extend the lifetime of the new buffer not the older one and the only way to do this was a proxy owner object `BufferDelegate` which holds a shared pointer to the real `Buffer` which in-turn holds a pointer to all `BufferDelegate` objects to update on repointing. A `BufferView` is effectively just a wrapper around `std::shared_ptr<BufferDelegate>` with more favorable semantics but generally just forwarding calls.
It should be additionally noted that to support usage of `RegisterUsage` the code around buffers in `GraphicsContext` was refactored to defer truly binding till the recording phase.
Due to an oversight, we weren't clearing the list of buffers that needed to be synced after every execution which led to them building up. Due to the relatively cheap synchronization of buffers and only doing so on faults this wasn't caught until now, it does depress the framerate significantly over time due to the size of the list growing to be in the range of 100k buffer views depending on the title.
The Kepler compute engine is used to run compute jobs encapsulated in to QMDs on the GPU, this commit doesn't implement compute itself but adds the register and QMD structs that will be needed for it in the future.
We wanted views to extend the lifetime of the underlying buffers and at the same time preserve all views until the destruction of the buffer to prevent recreation which might be costly in the future when we need `VkBufferView`s of the buffer but also require a centralized list of all views for recreation of the buffer. It also removes the inconsistency between `BufferView*` being returned in `GetXView` in `GraphicsContext`.
Alised descriptor sets are incorrectly interpreted by the shader compiler causing it to bugger up LLVM function argument types and crash
Co-authored-by: PixelyIon <pixelyion@protonmail.com>
This controls the depth range used by the shader, hades already has support for the necessary patching so we only need to pass the current mode over to it and it'll do the necessary work.
Using `eB5G6R5UnormPack16` (with a swizzle for `R5G6B5Unorm`) removes the need for `VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT` when those formats are aliased which happens in Sonic Mania among other titles.
Adreno GPUs have significant performance penalties from usage of `VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT` which require disabling UBWC and on Turnip, forces linear tiling. As a result, it's been made an optional quirk which doesn't supply the flag in `VkImageCreateInfo` and logs a warning if a view with a different Vulkan format from the original image is created.
We often need to alias the underlying data as multiple Vulkan formats which requires the `eMutableFormat` bit to be set in `VkImageCreateInfo`, without doing this there'll be validation layer errors and potentially GPU bugs.
As we no longer set the layout to general inside the Texture constructor, yet, we need it to be set prior to the image being used as an attachment. We need to transition the layout to `eGeneral` after creation of the texture object.
Any `RecyclerView`s with an app bar in a `CoordinatorLayout` would end up going off-screen due to the layout behavior implementing an offset by using a transform which would not correctly handle focusing on off-screen objects. This has now been fixed by manually adjusting height to be clipped to what is visible on the screen.
We collapse the app bar when the focus is on the app list which only occurs while using a controller, this is required as the app bar will never be collapsed otherwise. It also removes the older code to work around the limitation on `View.FOCUS_DOWN` by collapsing only when the end of the list was reached.
Removes card elevation as it visually conflicts with the scrim, this also makes the scrim a bit darker to emphasize the text and slightly reduces the border radius.
The entire layout is now selectable for grid items rather than just the card, this greatly increases the visibility of the selection when not in touch mode as the contrast of a darken effect on the icon can be minimal depending on how dark the icon already is.
The `InputStream` would not be closed after reading the key file in `KeyReader#import`, it's now wrapped with `use{ }` which handles closing the stream after usage.
Setting the refresh rate via the Display API's`preferredDisplayModeId` is an outdated method to do it on Android 11 and above, we now use `Surface#setFrameRate` alongside it to suggest a refresh rate for the display.
We incorrectly determined an Adreno driver bug to require padding between binding slots but the real issue was not supporting consecutive binding writes for `VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER` and was fixed by the padding slot unintentionally requiring individual writes. The quirk has now been corrected to explicitly specify this as the bug and the solution is more apt.
Any lookups done using `GetAlignedRecursiveRange` incorrectly added intervals in the exclusive interval entry lookups as the condition for adding them was the reverse of what it should've been due to a last minute refactor, it led to graphical glitches and crashes. This has been fixed and the lookups should return the correct results.
On certain devices, accesses to a protected memory region can return `si_code` as non-`SEGV_ACCERR` values, this leads to a crash as we only pass access violations to the trap handler and would lead to not doing so on those devices which would then result in going to the crash handler.
A large amount of Texture/Buffer views would expire before reuse could occur in `Texture::GetView`/`Buffer::GetView`. These can lead to a substantial memory allocation given enough time and they are now deleted during the lookup while iterating on all entries.
It should be noted that there are a lot of duplicate views that don't live long enough to be reused and the ultimate solution here is to make those views live long enough to be reused.
Similar to constant redundant synchronization for textures, there is a lot of redundant synchronization of buffers. Albeit, buffer synchronization is far cheaper than texture synchronization it still has associated costs which have now been reduced by only synchronizing on access.
There was a lot of redundant synchronization of textures to and from host constantly as we were not aware of guest memory access, this has now been averted by tracking any memory accesses to the texture memory using the NCE Memory Trapping API and synchronizing only when required.
An API for trapping accesses to guest memory and performing callbacks based on those accesses alongside managing protection of the memory. This is a fundamental building block for avoiding redundant synchronization of resources from the guest and host.
Note: All accesses are treated as write accesses at the moment, support for picking up read accesses will be implemented later
An interval map is a crucial piece of infrastructure required for memory faulting to track any regions that have an associated callback and their protection. Additionally, efficient page-aligned lookups with semantics optimal for memory faulting are also a requirement and the ability to associate multiple regions with a single callback/protection entry rather than doing so on a per-region basis as we deal with split-mapping resources.
This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access.
An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
`CreateMirror` is limited to creating a mirror of a single contiguous region which does not work when creating a contiguous mirror of multiple non-contiguous regions. To support this functionality, `CreateMirrors` which expects a list of page-aligned regions and maps them into a contiguous mirror.
We want to create arbitrary mirrors in the guest address space and to make this possible, we map the entire address space as a shared memory file. A mirror is mapped by using `mmap` with the offset into the guest address space.
Previously for methods with count > 1 the subchannel and engine would be looked up for each part of the method rather than only doing so at the start. Each call also needed to be looked up to see if it touched a macro or GPFIFO method. Fix this by doing checks outside of the main dispatch loop with templated helper lambdas to avoid needing to repeat lots of code. Maxwell3D is the only subchannel with a fast path for now but more can be added later if needed.
Almost every Maxwell format now directly corresponds to a Vulkan format. This allows formats to be passed through and the swizzle used directly from guest (with some extra swizzle handling for edge cases) thus saving the need to explicitly support each swizzle combination which is adds a lot of code bloat. The format header is additionally reordered with line breaks to separate formats by their bits-per-block.
We always submit pipeline divisor descriptions regardless of binding input rate being vertex rather than instance. This is invalid behavior and has been fixed by only submitting binding descriptors when the input rate is per-instance.
Adreno proprietary drivers suffer from a bug where `VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER` requires 2 descriptor slots rather than one, we add a padding slot to fix this issue. `QuirkManager` was introduced to handle per-vendor/per-device errata and allow enabling this on Adreno proprietary drivers specifically as to not affect the performance of other devices.
Quirk terminology was deemed to be inappropriate for describing the features/extensions of a device. It has been replaced with traits which is far more fitting but quirks will be used as a terminology for errata in devices.
The texture handle offset calculation involved an incorrect shift by descriptor size which was found to be unnecessary and would result in an invalid handle that had the wrong TIC/TSC index and caused broken rendering.
`nodes` and `syncTextures` were cleared after waiting on the `CommandExecutor` fence rather than before, this wasted execution time after the wait for something that could be performed prior to the wait.
We now attempt to enable `VK_KHR_uniform_buffer_standard_layout` when present as lax UBO layout significantly reduces complexity. If a device doesn't support this extension, we still assume that the device supports it implicitly as this has proven to be true across all major mobile GPU vendors regardless of the driver version but enabling this prevents validation layer errors.
We depend on past commands to have completed execution in a renderpass, a subpass dependency on all graphics stages from `VK_SUBPASS_EXTERNAL` to subpass #0 is used to enforce this. Nvidia and Adreno proprietary drivers implicitly do this but Turnip or Mali drivers require this or they execute out of order.
Blocklinear texture decoding was broken for padding blocks and would incorrectly decode them resulting in major texture corruption for any textures with their widths not aligned to 64 bytes. This has now been fixed with neater code which avoids redundant repetition of any code using lambdas and functions where necessary.
Stencil operations are configurable to be the same for both sides or have independent stencil state for both sides. It is controlled via the previously unimplemented `stencilTwoSideEnable`.
Fermi2D supports macros in addition to Maxwell3D, these both share code memory. To support this we rework the macro interpreter to support passing in a target engine and abstract the communications out into an interface that can be implemented by applicable engines.
```
GPFIFO <-> MME <-> Maxwell3D
^ ^---> Fermi2D
X------------> I2M
X------------> MaxwellComputeB
X--Flush-----> MaxwellDMA
```
Shader programs allocate instructions and blocks within an `ObjectPool`, there was a global pool prior that was never reaped aside from on destruction. This led to a leak where the pool would contain resources from shader programs that had been deleted, to avert this the pools are now tied to shader programs.
The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.
erase invalidated `it` leading to a potential segfault if the GPU was very far behind, bail out early to avoid that since there can only be one occurence at most in the buffer anyway.
Implements the entirety of Maxwell3D Depth/Stencil state for both faces including compare/write masks and reference value. Maxwell3D register `stencilTwoSideEnable` is ignored as its behavior is unknown and could mean the same behavior for both stencils or the back facing stencil being disabled as a result of this it is unimplemented.
We don't respect the host subresource layout in synchronizing linear textures from the guest to the host when mapped to memory directly, this leads to texture corruption and while the real fix would involve respecting the host subresource layout, this has been deferred for later as real world performance advantages/disadvantages associated with this change can be observed more carefully to determine if it's worth it.
Color RTs are disabled by setting their format as `None`, it was removed while transitioning to macros and resulted in a missing format exception. It has been readded as several applications depend on this behavior.
