Initial PR was made by Avan for Ryubing.
This allows for Trails in the Sky 1st Chapter (and by extension Trails through Daybreak, Trails through Daybreak II, Trails Beyond the Horizon, and the future Trails in the Sky 2nd Chapter – tested via demo – to be fully playable.
Initial PR caused issues on macOS devices, specifically crashing in certain games (such as Mario Kart 8 Deluxe and the aforementioned Trails games). It was reverted on the original Ryubing project (at the time of writing this description) for causing small rendering issues on The Legend of Zelda: Breath of the Wild.
The rendering issues do not appear on macOS.
This commit improves the original PR by eliminating said macOS crash.
Trails in the Sky 1st writes its exposure/brightness value through an R32G32Float 1x1 texture and later reads the raw bits back through an R32Uint 2x1 texture mapped to the same guest memory. The two texture formats/dimensions are fully incompatible as texture views, so they end
up as separate host textures, and nothing kept their contents coherent - the reader saw stale zeroed data, making the exposure calculation (and the rendered image) too dark. (This is the initial fix for the game)
TextureGroup already had a raw-copy dependency mechanism intended to
handle this class of alias (TextureDependency, TextureGroupHandle
raw-copy plumbing), gated by CanCreateRawCopyDependency's strict same-guest-memory checks. This commit fixes three bugs that kept that mechanism from working correctly (and thus crashing Trails and other games, such as Mario Kart 8 Deluxe):
(1) TextureGroup.InitializeOverlaps() and TextureGroup.RegisterIncompatibleOverlap() only forwarded overlaps to CreateCopyDependency() when compatibility was LayoutIncompatible or better, silently excluding the fully Incompatible case the raw-copy path exists for. Widened both guards to let Incompatible overlaps through so CanCreateRawCopyDependency actually gets a chance to run.
(2) Once Incompatible overlaps were allowed through, CreateCopyDependency(TextureGroup, ...) could still fall back to a regular, non-raw textureCopy for such pairs, since ViewLayoutCompatible/CopySizeMatches only check byte size and were never meant to reason about fully incompatible pairs (e.g. a depth format aliasing a color format with the same byte size). On Vulkan running through MoltenVK on macOS, that non-raw copy path requires a pixel-format-reinterpreting texture view, and MoltenVK/Metal refuses to create any view onto a depth-format texture ("not castable"), aborting emulation. textureCopy is now forced false for Incompatible pairs, leaving raw copy (which already excludes depth/stencil formats) as the only route for that severity level.
(3) TextureGroupHandle.Inherit() copied a handle's pending DeferredCopy to the new handle when a view was recreated, but did not copy DeferredCopyRaw alongside it. A handle that inherited a pending raw copy would then execute it through the regular (non-raw) CopyTo path once acknowledged, reinterpreting the source bytes with the wrong row layout and corrupting the image - visible as vertical flickering stripes whenever a view happened to be recreated with a raw copy still pending. DeferredCopyRaw is now carried over together with DeferredCopy.
(As mentioned earlier, vertical stripes aren't present on macOS, so (3) doesn't negatively have an effect on macOS in any way).
Trails in the Sky 1st uses texture descriptors with different formats and dimensions that map to the same region of guest GPU memory. For example, the game interprets the same underlying data as both an R32Uint 2x1 texture and an R32G32Float 1x1 texture. Both textures have an 8-byte logical payload, but their texel formats and dimensions are different.
Ryubing cannot represent these two descriptors as normally compatible texture views, so it creates separate host texture objects for them. Since these host textures represent the same guest memory, a modification made to one of them should become visible to the other before the other texture is read. However, the existing synchronization mechanism does not fully handle this situation.
The game uses this pattern for its exposure or brightness history. A compute shader first writes the updated exposure data to the R32G32Float 1x1 host texture. A later operation then reads the raw bits of the same data through the R32Uint 2x1 host texture. The original Ryubing implementation did not synchronize the contents of these two separate host textures. As a result, the reading texture received stale zero values, which caused the shader to calculate an incorrectly low exposure value and made the rendered image appear too dark.
TextureGroup already has a guest-memory synchronization mechanism for incompatible overlaps. When _flushIncompatibleOverlaps is enabled, the most recent texture contents are written back to guest memory, after which the other texture reloads the data. However, this mechanism was primarily enabled through IsFormatHostIncompatible, which checks whether an individual texture format can be correctly supported by the host GPU.
In this case, both R32Uint and R32G32Float are individually supported by the host GPU. Therefore, IsFormatHostIncompatible does not enable this synchronization path. The issue is not that either format is unsupported. The issue is that two individually supported aliases, represented by separate host textures, do not maintain coherent contents.
In theory, the latest contents could be flushed to guest memory immediately after every GPU write to a host texture, allowing other textures that map to the same guest memory to reload the updated data. However, doing so could frequently trigger expensive operations such as GPU waits, data readbacks, guest texture layout conversions, and memory-tracking updates. This would be particularly costly for textures that are updated every frame or during every compute dispatch.
For this reason, the fix does not require an immediate flush after every write. Instead, it extends Ryubing's existing TextureDependency and TextureGroupHandle mechanisms. When two fully incompatible textures map to exactly the same guest memory and satisfy a strict set of raw-copy requirements, the program creates a raw-copy dependency between them.
When the source texture is modified, the target texture is only marked as requiring a raw copy. The latest logical raw bytes are copied from the source only when the target texture is actually about to be used. This keeps the contents of the separate host texture aliases coherent while avoiding unnecessary round trips through guest memory.
General memory improvements to decrease GC pressure and frequency.
Pool big arrays and objects that are created and deleted often.
Skip data copies when they aren't needed.
Inline flag checks to skip unneeded allocations.
From my testing the performance is about the same, but the GC frequency is much lower and collection is faster causing less and smaller spikes.
* refactor(perf): pass MemoryOwner<byte> around as itself rather than IMemoryOwner<byte>
* fix(perf): get span via MemoryOwner<byte>.Span property instead of through Memory property
* fix: incorrect comment change
* GPU: Migrate buffers on GPU project, pre-emptively flush device local mappings
Essentially retreading #4540, but it's on the GPU project now instead of the backend. This allows us to have a lot more control + knowledge of where the buffer backing has been changed and allows us to pre-emptively flush pages to host memory for quicker readback. It will allow us to do other stuff in the future, but we'll get there when we get there.
Performance greatly improved in Hyrule Warriors: Age of Calamity. Performance notably improved in TOTK (average). Performance for BOTW restored to how it was before #4911, perhaps a bit better.
- Rewrites a bunch of buffer migration stuff. Might want to tighten up how dispose stuff works.
- Fixed an issue where the copy for texture pre-flush would happen _after_ the syncpoint.
TODO: remove a page from pre-flush if it isn't flushed after a certain number of copies.
* Add copy deactivation
* Fix dependent virtual buffers
* Remove logging
* Fix format issues (maybe)
* Vulkan: Remove backing swap
* Add explicit memory access types for most buffers
* Fix typo
* Add device local force expiry, change buffer inheritance behaviour
* General cleanup, OGL fix
* BufferPreFlush comments
* BufferBackingState comments
* Add an extra precaution to BufferMigration
This is very unlikely, but it's important to cover loose ends like this.
* Address some feedback
* Docs
* rebase
* add methods Ryyjinx.Common EmbeddedResources and SteamUtils
* GAL changes - change SetData() methods and ThreadedTexture commands to use IMemoryOwner<byte> instead of SpanOrArray<byte>
* Ryujinx.Graphics.Texture: change texture conversion methods to return IMemoryOwner<byte> and allocate from ByteMemoryPool
* Ryujinx.Graphics.OpenGL: update ITexture and Texture-like types with SetData() methods to take IMemoryOwner<byte> instead of SpanOrArray<byte>
* Ryujinx.Graphics.Vulkan: update ITexture and Texture-like types with SetData() methods to take IMemoryOwner<byte> instead of SpanOrArray<byte>
* Ryujinx.Graphics.Gpu: update ITexture and Texture-like types with SetData() methods to take IMemoryOwner<byte> instead of SpanOrArray<byte>
* Remove now-unused SpanOrArray<T>
* post-rebase cleanup
* PixelConverter: remove unsafe modifier on safe methods, and remove one unnecessary cast
* use ByteMemoryPool.Rent() in GetWritableRegion() impls
* fix formatting, rename `ReadRentedMemory()` to `ReadFileToRentedMemory()``
* Texture.ConvertToHostCompatibleFormat(): dispose of `result` in Astc decode branch
* Add host tracked memory manager mode
* Skipping flush is no longer needed
* Formatting + revert unrelated change
* LightningJit: Ensure that dest register is saved for load ops that do partial updates
* avoid allocations when doing address space lookup
Add missing improvement
* IsRmwMemory -> IsPartialRegisterUpdateMemory
* Ensure we iterate all private allocations in range
* PR feedback and potential fixes
* Simplified bridges a lot
* Skip calling SignalMappingChanged if Guest is true
* Late map bridge too
* Force address masking for prefetch instructions
* Reprotection for bridges
* Move partition list validation to separate debug method
* Move host tracked related classes to HostTracked folder
* New HostTracked namespace
* Move host tracked modes to the end of enum to avoid PPTC invalidation
---------
Co-authored-by: riperiperi <rhy3756547@hotmail.com>
* GPU: Don't create tracking handles for buffer textures
Buffer texture memory is handled by the buffer cache - the texture shouldn't create any tracking handles as they aren't used. This change simply makes them create and iterate 0 tracking handles, while keeping the rest of the texture group around.
This prevents a possible issue where many buffer textures are created as views of overlapping buffer ranges, and virtual regions have many dependant textures that don't actually contribute anything to handle state.
Should improve performance in Mortal Kombat 1, possibly certain UE4 games when FIFO raises to 100%.
* Fix interval tree bug
* Don't check view compatibility for buffer textures
* dotnet format style --severity info
Some changes were manually reverted.
* dotnet format analyzers --serverity info
Some changes have been minimally adapted.
* Restore a few unused methods and variables
* Silence dotnet format IDE0060 warnings
* Silence dotnet format IDE0052 warnings
* Address dotnet format CA1816 warnings
* Address or silence dotnet format CA1069 warnings
* Address or silence dotnet format CA2211 warnings
* Address remaining dotnet format analyzer warnings
* Address review comments
* Address most dotnet format whitespace warnings
* Apply dotnet format whitespace formatting
A few of them have been manually reverted and the corresponding warning was silenced
* Format if-blocks correctly
* Run dotnet format whitespace after rebase
* Run dotnet format style after rebase
* Another rebase, another dotnet format run
* Run dotnet format style after rebase
* Run dotnet format after rebase and remove unused usings
- analyzers
- style
- whitespace
* Disable 'prefer switch expression' rule
* Add comments to disabled warnings
* Remove a few unused parameters
* Replace MmeShadowScratch with Array256<uint>
* Simplify properties and array initialization, Use const when possible, Remove trailing commas
* Start working on disabled warnings
* Fix and silence a few dotnet-format warnings again
* Run dotnet format after rebase
* Address IDE0251 warnings
* Silence IDE0060 in .editorconfig
* Revert "Simplify properties and array initialization, Use const when possible, Remove trailing commas"
This reverts commit 9462e4136c0a2100dc28b20cf9542e06790aa67e.
* dotnet format whitespace after rebase
* First pass of dotnet format
* Add unsafe dotnet format changes
* Fix typos
* Add trailing commas
* Disable formatting for FormatTable
* Address review feedback
* GPU: Remove swizzle undefined matching and rework depth aliasing
@gdkchan pointed out that UI textures in TOTK seemed to be setting their texture swizzle incorrectly (texture was RGB but was sampling A, swizzle for A was wrong), so I determined that SwizzleComponentMatches was the problem and set on eliminating it. This PR combines existing work to select the most recently modified texture (now used when selecting which aliased texture to use) with some additional changes to remove the swizzle check and support aliased view creation.
The original observation (#1538) was that we wanted to match depth textures for the purposes of aliasing with color textures, but they often had different swizzle from what was sampled (as it's generally the identity swizzle once rendered). At the time, I decided to allow swizzles to match if only the defined components matched, which fixed the issue in all known cases but could easily be broken by a game _expecting_ a given swizzle, such as a 1/0 value on a component.
This error case could also occur in textures that don't even depth alias, such as R11G11B10, as the rule was created to generally apply to all cases.
The solution is now to fail this exact match test, and allow the search for an R32 texture to create a swizzled view of a D32 texture (and other such cases). This allows the creation of a view that mismatches the requested format, which wasn't present before and was the reason for the swizzle matching approach.
The exact match and view creation rules now follow the same rules over what textures to select when there are multiple options (such as a "perfect" match and an "aliased" match at the same time). It now selects the most recently modified texture, which is done with a new sequence number in the GpuContext (because we don't have enough of these).
Reportedly fixes UI having weird coloured backgrounds in TOTK. This also fixes an issue in MK8D where returning from a race resulted in the character selection cubemaps being broken. May work around issues introduced by the "short texture cache" PR due to modification ordering, though they won't be truly fixed.
Should allow (#4365) to avoid copies in more cases. Need to test that.
I tested a bunch of games #1538 originally affected and they seem to be fine. This change affects all games so it would be good to get some wide testing on it.
* Address feedback 1, fix an issue
* Workaround: Do not allow copies for format alias.
These should be removed when D32<->R32 copy dependencies become legal
* GPU: Remove CPU region handle containers.
Another one for the "I don't know why I didn't do this earlier" pile.
This removes the "Cpu" prefixed region handle classes, which each mirror a region handle type from Ryujinx.Memory.
Originally, not all projects had a reference to Ryujinx.Memory, so these classes were introduced to bridge the gap. Someone else crossed that bridge since, so these classes don't have much of a purpose anymore.
This PR replaces all uses of CpuRegionHandle etc to their direct Ryujinx.Memory versions.
RegionHandle methods (specifically QueryModified) are about the hottest path there is in the entire emulator, so there is a nice boost from doing this.
* Add docs
* WIP texture pre-flush
Improve performance of TextureView GetData to buffer
Fix copy/sync ordering
Fix minor bug
Make this actually work
WIP host mapping stuff
* Fix usage flags
* message
* Cleanup 1
* Fix rebase
* Fix
* Improve pre-flush rules
* Fix pre-flush
* A lot of cleanup
* Use the host memory bits
* Select the correct memory type
* Cleanup TextureGroupHandle
* Missing comment
* Remove debugging logs
* Revert BufferHandle _value access modifier
* One interrupt action at a time.
* Support D32S8 to D24S8 conversion, safeguards
* Interrupt cannot happen in sync handle's lock
Waitable needs to be checked twice now, but this should stop it from deadlocking.
* Remove unused using
* Address some feedback
* Address feedback
* Address more feedback
* Address more feedback
* Improve sync rules
Should allow for faster sync in some cases.
Neo