* Eliminate CB0 accesses
Still some work to do, decouple from hle?
* Forgot the important part somehow
* Fix and improve alignment test
* Address Feedback
* Remove some complexity when checking storage buffer alignment
* Update Ryujinx.Graphics.Shader/Translation/Optimizations/GlobalToStorage.cs
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
* Implement HLE macro for DrawElementsIndirect
* Shader cache version bump
* Use GL_ARB_shader_draw_parameters extension on OpenGL
* Fix DrawIndexedIndirectCount on Vulkan when extension is not supported
* Implement DrawIndex
* Alignment
* Fix some validation errors
* Rename BaseIds to DrawParameters
* Fix incorrect index buffer and vertex buffer size in some cases
* Add HLE macros for DrawArraysInstanced and DrawElementsInstanced
* Perform a regular draw when indirect data is not modified
* Use non-indirect draw methods if indirect buffer was not GPU modified
* Only check if draw parameters match if the shader actually uses them
* Expose Macro HLE setting on GUI
* Reset FirstVertex and FirstInstance after draw
* Update shader cache version again since some people already tested this
* PR feedback
Co-authored-by: riperiperi <rhy3756547@hotmail.com>
* WIP Vulkan implementation
* No need to initialize attributes on the SPIR-V backend anymore
* Allow multithreading shaderc and vkCreateShaderModule
You'll only really see the benefit here with threaded-gal or parallel shader cache compile.
Fix shaderc multithreaded changes
Thread safety for shaderc Options constructor
Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes.
* Support multiple levels/layers for blit.
Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now.
* TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use
* New depth-stencil blit method for AMD
* Workaround for AMD driver bug
* Fix some tessellation related issues (still doesn't work?)
* Submit command buffer before Texture GetData. (UE4 fix)
* DrawTexture support
* Fix BGRA on OpenGL backend
* Fix rebase build break
* Support format aliasing on SetImage
* Fix uniform buffers being lost when bindings are out of order
* Fix storage buffers being lost when bindings are out of order
(also avoid allocations when changing bindings)
* Use current command buffer for unscaled copy (perf)
Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies.
* Update to .net6
* Update Silk.NET to version 2.10.1
Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before.
* Fix PrimitivesGenerated query, disable Transform Feedback queries for now
Lets Splatoon 2 work on nvidia. (mostly)
* Update counter queue to be similar to the OGL one
Fixes softlocks when games had to flush counters.
* Don't throw when ending conditional rendering for now
This should be re-enabled when conditional rendering is enabled on nvidia etc.
* Update findMSB/findLSB to match master's instruction enum
* Fix triangle overlay on SMO, Captain Toad, maybe others?
* Don't make Intel Mesa pay for Intel Windows bugs
* Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders)
* Update Spv.Generator
* Add alpha test emulation on shader (but no shader specialisation yet...)
* Fix R4G4B4A4Unorm texture format permutation
* Validation layers should be enabled for any log level other than None
* Add barriers around vkCmdCopyImage
Write->Read barrier for src image (we want to wait for a write to read it)
Write->Read barrier for dst image (we want to wait for the copy to complete before use)
* Be a bit more careful with texture access flags, since it can be used for anything
* Device local mapping for all buffers
May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?)
Also some performance things and fixes issues with opengl games loading textures weird.
* Cleanup, disable device local buffers for now.
* Add single queue support
Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment.
* Fix some validation errors around extended dynamic state
* Remove Intel bug workaround, it was fixed on the latest driver
* Use circular queue for checking consumption on command buffers
Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once.
* Use SupportBufferUpdater, add single layer flush
* Fix counter queue leak when game decides to use host conditional rendering
* Force device local storage for textures (fixes linux performance)
* Port #3019
* Insert barriers around vkCmdBlitImage (may fix some amd flicker)
* Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers
* Don't pause transform feedback for multi draw
* Fix draw outside of render pass and missing capability
* Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more)
* Better workaround for AMD vertex buffer size alignment issue
* More instructions + fixes on SPIR-V backend
* Allow custom aspect ratio on Vulkan
* Correct GTK UI status bar positions
* SPIR-V: Functions must always end with a return
* SPIR-V: Fix ImageQuerySizeLod
* SPIR-V: Set DepthReplacing execution mode when FragDepth is modified
* SPIR-V: Implement LoopContinue IR instruction
* SPIR-V: Geometry shader support
* SPIR-V: Use correct binding number on storage buffers array
* Reduce allocations for Spir-v serialization
Passes BinaryWriter instead of the stream to Write and WriteOperand
- Removes creation of BinaryWriter for each instruction
- Removes allocations for literal string
* Some optimizations to Spv.Generator
- Dictionary for lookups of type declarations, constants, extinst
- LiteralInteger internal data format -> ushort
- Deterministic HashCode implementation to avoid spirv result not being the same between runs
- Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost)
TODO: improve instruction allocation, structured program creator, ssa?
* Pool Spv.Generator resources, cache delegates, spv opts
- Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module.
- NewInstruction is called instead of new Instruction()
- Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually.
- Estimate code size when creating the output MemoryStream
- LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast.
Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call.
TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads.
* LocalDefMap for Ssa Rewriter
Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator.
* SPIR-V: Transform feedback support
* SPIR-V: Fragment shader interlock support (and image coherency)
* SPIR-V: Add early fragment tests support
* SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta
* Don't pass depth clip state right now (fix decals)
Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted.
* Multisampling support
* Multisampling: Use resolve if src samples count > dst samples count
* Multisampling: We can only resolve for unscaled copies
* SPIR-V: Only add FSI exec mode if used.
* SPIR-V: Use ConstantComposite for Texture Offset Vector
Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are.
* SPIR-V: Don't OpReturn if we already OpExit'ed
Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist)
* SPIR-V: Only use input attribute type for input attributes
Output vertex attributes should always be of type float.
* Multithreaded Pipeline Compilation
* Address some feedback
* Make this 32
* Update topology with GpuAccessorState
* Cleanup for merge (note: disables spir-v)
* Make more robust to shader compilation failure
- Don't freeze when GLSL compilation fails
- Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure.
* Fix Multisampling
* Only update fragment scale count if a vertex texture needs a scale.
Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage.
This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now.
* Use a bitmap to do granular tracking for buffer uploads.
This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful.
Avoids ending render passes to update buffer data (not all the time)
- 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI)
- Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI)
As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory)
Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here.
* Copy query results after RP ends, rather than ending to copy
We need to end the render pass to get the data (submit command buffer) anyways...
Reduces render passes created in games that use queries.
* Rework Query stuff a bit to avoid render pass end
Tries to reset returned queries in background when possible, rather than ending the render pass.
Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?)
* Remove unnecessary lines
Was for testing
* Fix validation error for query reset
Need to think of a better way to do this.
* SPIR-V: Fix SwizzleAdd and some validation errors
* SPIR-V: Implement attribute indexing and StoreAttribute
* SPIR-V: Fix TextureSize for MS and Buffer sampler types
* Fix relaunch issues
* SPIR-V: Implement LogicalExclusiveOr
* SPIR-V: Constant buffer indexing support
* Ignore unsupported attributes rather than throwing (matches current GLSL behaviour)
* SPIR-V: Implement tessellation support
* SPIR-V: Geometry shader passthrough support
* SPIR-V: Implement StoreShader8/16 and StoreStorage8/16
* SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug
* SPIR-V: Fix field index for scale count
* SPIR-V: Fix another case of wrong field index
* SPIRV/GLSL: More scaling related fixes
* SPIR-V: Fix ImageLoad CompositeExtract component type
* SPIR-V: Workaround for Intel FrontFacing bug
* Enable SPIR-V backend by default
* Allow null samplers (samplers are not required when only using texelFetch to access the texture)
* Fix some validation errors related to texel block view usage flag and invalid image barrier base level
* Use explicit subgroup size if we can (might fix some block flickering on AMD)
* Take componentMask and scissor into account when clearing framebuffer attachments
* Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA)
* Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight)
Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different
* Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc)
* Check if the subgroup size is supported before passing a explicit size
* Only enable ShaderFloat64 if the GPU supports it
* We don't need to recompile shaders if alpha test state changed but alpha test is disabled
* Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before)
* Fix pipeline state saving before it is updated.
This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache.
* Allow null samplers on OpenGL backend
* _unit0Sampler should be set only for binding 0
* Remove unused PipelineConverter format variable (was causing IOR)
* Raise textures limit to 64 on Vulkan
* No need to pack the shader binaries if shader cache is disabled
* Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL
* Do not clear unbound framebuffer color attachments
* Geometry shader passthrough emulation
* Consolidate UpdateDepthMode and GetDepthMode implementation
* Fix A1B5G5R5 texture format and support R4G4 on Vulkan
* Add barrier before use of some modified images
* Report 32 bit query result on AMD windows (smo issue)
* Add texture recompression support (disabled for now)
It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures
* Do not report R4G4 format as supported on Vulkan
It was causing mario head to become white on Super Mario 64 (???)
* Improvements to -1 to 1 depth mode.
- Transformation is only applied on the last stage in the vertex pipeline.
- Should fix some issues with geometry and tessellation (hopefully)
- Reading back FragCoord Z on fragment will transform back to -1 to 1.
* Geometry Shader index count from ThreadsPerInputPrimitive
Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL
* Remove gl_FragDepth scaling
This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade.
* Add Gl StencilOp enum values to Vulkan
* Update guest cache to v1.1 (due to specialization state changes)
This will explode your shader cache from earlier vulkan build, but it must be done. 😔
* Vulkan/SPIR-V support for viewport inverse
* Fix typo
* Don't create query pools for unsupported query types
* Return of the Vector Indexing Bug
One day, everyone will get this right.
* Check for transform feedback query support
Sometimes transform feedback is supported without the query type.
* Fix gl_FragCoord.z transformation
FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs.
Fixes Pokemon Sword/Shield, possibly others.
* Fix Avalonia Rebase
Vulkan is currently not available on Avalonia, but the build does work and you can use opengl.
* Fix headless build
* Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host
* Fix BCn 4/5 conversion, GetTextureTarget
BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect.
GetTextureTarget was not creating a view with the replacement format.
* Fix dependency
* Fix inverse viewport transform vector type on SPIR-V
* Do not require null descriptors support
* If MultiViewport is not supported, do not try to set more than one viewport/scissor
* Bounds check on bitmap add.
* Flush queries on attachment change rather than program change
Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending.
Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets.
* Add support for avalonia (#6)
* add avalonia support
* only lock around skia flush
* addressed review
* cleanup
* add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check
* fix getting window handle on linux. skip render is size is 0
* Combine non-buffer with buffer image descriptor sets
* Support multisample texture copy with automatic resolve on Vulkan
* Remove old CompileShader methods from the Vulkan backend
* Add minimal pipeline layouts that only contains used bindings
They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders
* Pre-compile helper shader as SPIR-V, and some fixes
* Remove pre-compiled shaderc binary for Windows as its no longer needed by default
* Workaround RADV crash
Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist".
* Use RobustBufferAccess on NVIDIA gpus
Avoids the SMO waterfall triangle on older NVIDIA gpus.
* Implement GPU selector and expose texture recompression on the UI and config
* Fix and enable background compute shader compilation
Also disables warnings from shader cache pipeline misses.
* Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added
* If S8D24 is not supported, use D32FS8
* Ensure all fences are destroyed on dispose
* Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks
* Add missing clear layer parameter after rebase
* Use selected gpu from config for avalonia (#7)
* use configured device
* address review
* Fix D32S8 copy workaround (AMD)
Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things.
* Use push descriptors for uniform buffer updates (disabled for now)
* Push descriptor support check, buffer redundancy checks
Should make push descriptors faster, needs more testing though.
* Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs
* Adjust bindings array sizes
* Force submit command buffers if memory in use by its resources is high
* Add workaround for AMD GCN cubemap view sins
`ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue.
This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry)
Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there.
* Add mobile, non-RX variants to the GCN regex.
Also make sure that the 3 digit ones only include numbers starting with 7 or 8.
* Increase image limit per stage from 8 to 16
Xenoblade Chronicles 2 was hiting the limit of 8
* Minor code cleanup
* Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer
* Add gpu selector to Avalonia (#8)
* Add gpu selector to avalonia settings
* show backend label on window
* some fixes
* address review
* Minor changes to the Avalonia UI
* Update graphics window UI and locales. (#9)
* Update xaml and update locales
* locale updates
Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish?
* Fix locales with more (?) correct translations.
* add separator to render widget
* fix spanish and portuguese
* Add new IdList, replaces buffer list that could not remove elements and had unbounded growth
* Don't crash the settings window if Vulkan is not supported
* Fix Actions menu not being clickable on GTK UI after relaunch
* Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer
* Fix IdList and make it not thread safe
* Revert useless OpenGL format table changes
* Fix headless project build
* List throws ArgumentOutOfRangeException
* SPIR-V: Fix tessellation
* Increase shader cache version due to tessellation fix
* Reduce number of Sync objects created (improves perf in some specific titles)
* Fix vulkan validation errors for NPOT compressed upload and GCN workaround.
* Add timestamp to the shader cache and force rebuild if host cache is outdated
* Prefer Mail box present mode for popups (#11)
* Prefer Mail box present mode
* fix debug
* switch present mode when vsync is toggled
* only disable vsync on the main window
* SPIR-V: Fix geometry shader input load with transform feedback
* BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0
* Fix Avalonia build
* Address initial PR feedback
* Only set transform feedback outputs on last vertex stage
* Address riperiperi PR feedback
* Remove outdated comment
* Remove unused constructor
* Only throw for negative results
* Throw for QueueSubmit and other errors
No point in delaying the inevitable
* Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler
* Fix some resolution scale issues
* No need for two UpdateScale calls
* Fix comments on SPIR-V generator project
* Try to fix shader local memory size
On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region?
* Remove RectangleF that is now unused
* Fix ImageGather with multiple offsets
Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct`
* Address PR feedback from jD in all projects except Avalonia
* Address most of jD PR feedback on Avalonia
* Remove unsafe
* Fix VulkanSkiaGpu
* move present mode request out of Create Swapchain method
* split more parts of create swapchain
* addressed reviews
* addressed review
* Address second batch of jD PR feedback
* Fix buffer <-> image copy row length and height alignment
AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes
* Better fix for NPOT alignment issue
* Use switch expressions on Vulkan EnumConversion
Thanks jD
* Fix Avalonia build
* Add Vulkan selection prompt on startup
* Grammar fixes on Vulkan prompt message
* Add missing Vulkan migration flag
Co-authored-by: riperiperi <rhy3756547@hotmail.com>
Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com>
Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
* Changes 1
* Changes 2
* Better ModifiedSequence handling
This should handle PreciseEvents properly, and simplifies a few things.
* Minor changes, remove debug log
* Handle stage.Info being null
Hopefully fixes Catherine crash
* Fix shader specialization fast texture lookup
* Fix some things.
* Address Feedback Part 1
* Make method static.
* New shader cache implementation
* Remove some debug code
* Take transform feedback varying count into account
* Create shader cache directory if it does not exist + fragment output map related fixes
* Remove debug code
* Only check texture descriptors if the constant buffer is bound
* Also check CPU VA on GetSpanMapped
* Remove more unused code and move cache related code
* XML docs + remove more unused methods
* Better codegen for TransformFeedbackDescriptor.AsSpan
* Support migration from old cache format, remove more unused code
Shader cache rebuild now also rewrites the shared toc and data files
* Fix migration error with BRX shaders
* Add a limit to the async translation queue
Avoid async translation threads not being able to keep up and the queue growing very large
* Re-create specialization state on recompile
This might be required if a new version of the shader translator requires more or less state, or if there is a bug related to the GPU state access
* Make shader cache more error resilient
* Add some missing XML docs and move GpuAccessor docs to the interface/use inheritdoc
* Address early PR feedback
* Fix rebase
* Remove IRenderer.CompileShader and IShader interface, replace with new ShaderSource struct passed to CreateProgram directly
* Handle some missing exceptions
* Make shader cache purge delete both old and new shader caches
* Register textures on new specialization state
* Translate and compile shaders in forward order (eliminates diffs due to different binding numbers)
* Limit in-flight shader compilation to the maximum number of compilation threads
* Replace ParallelDiskCacheLoader state changed event with a callback function
* Better handling for invalid constant buffer 1 data length
* Do not create the old cache directory structure if the old cache does not exist
* Constant buffer use should be per-stage. This change will invalidate existing new caches (file format version was incremented)
* Replace rectangle texture with just coordinate normalization
* Skip incompatible shaders that are missing texture information, instead of crashing
This is required if we, for example, support new texture instruction to the shader translator, and then they allow access to textures that were not accessed before. In this scenario, the old cache entry is no longer usable
* Fix coordinates normalization on cubemap textures
* Check if title ID is null before combining shader cache path
* More robust constant buffer address validation on spec state
* More robust constant buffer address validation on spec state (2)
* Regenerate shader cache with one stream, rather than one per shader.
* Only create shader cache directory during initialization
* Logging improvements
* Proper shader program disposal
* PR feedback, and add a comment on serialized structs
* XML docs for RegisterTexture
Co-authored-by: riperiperi <rhy3756547@hotmail.com>
* De-tile GOB when DMA copying from block linear to pitch kind memory regions
* XML docs + nits
* Remove using
* No flush for regular buffer copies
* Add back ulong casts, fix regression due to oversight
* Allow textures to have their data partially mapped
* Explicitly check for invalid memory ranges on the MultiRangeList
* Update GetWritableRegion to also support unmapped ranges
* Initial test for texture sync
* WIP new texture flushing setup
* Improve rules for incompatible overlaps
Fixes a lot of issues with Unreal Engine games. Still a few minor issues (some caused by dma fast path?) Needs docs and cleanup.
* Cleanup, improvements
Improve rules for fast DMA
* Small tweak to group together flushes of overlapping handles.
* Fixes, flush overlapping texture data for ASTC and BC4/5 compressed textures.
Fixes the new Life is Strange game.
* Flush overlaps before init data, fix 3d texture size/overlap stuff
* Fix 3D Textures, faster single layer flush
Note: nosy people can no longer merge this with Vulkan. (unless they are nosy enough to implement the new backend methods)
* Remove unused method
* Minor cleanup
* More cleanup
* Use the More Fun and Hopefully No Driver Bugs method for getting compressed tex too
This one's for metro
* Address feedback, ASTC+ETC to FormatClass
* Change offset to use Span slice rather than IntPtr Add
* Fix this too
Fixes a regression from #2663 where buffer flush would not happen after a resize. Specifically caused the world map in Yoshi's Crafted World to flash.
I have other planned changes to this class so this might change soon, but this regression could affect a lot so it couldn't wait.
This fixes a potential regression with the new range list changes, where the cost for creating new ones would be rather large due to creating a 1024 size array. Also reduces cost for range list inheritance by using the first existing range list as a base, rather than creating a new one then adding both lists to it.
The growth size for the RangeList is now identical to its initial size. Every 32 elements was probably a little too common - now it is 1024 for most things and 8 for the buffer modified range list.
The Unmapped and SyncMethod methods have been changed to ensure that they behave properly if the range list is set null. Cleaned up a few calls to use the null-conditional operator.
* Replace CacheResourceWrite with more general "precise" write
The goal of CacheResourceWrite was to notify GPU resources when they were modified directly, by looking up the modified address/size in a structure and calling a method on each resource. The downside of this is that each resource cache has to be queried individually, they all have to implement their own way to do this, and it can only signal to resources using the same PhysicalMemory instance.
This PR adds the ability to signal a write as "precise" on the tracking, which signals a special handler (if present) which can be used to avoid unnecessary flush actions, or maybe even more. For buffers, precise writes specifically do not flush, and instead punch a hole in the modified range list to indicate that the data on GPU has been replaced.
The downside is that precise actions must ignore the page protection bits and always signal - as they need to notify the target resource to ignore the sequence number optimization.
I had to reintroduce the sequence number increment after I2M, as removing it was causing issues in rabbids kingdom battle. However - all resources modified by I2M are notified directly to lower their sequence number, so the problem is likely that another unrelated resource is not being properly updated. Thankfully, doing this does not affect performance in the games I tested.
This should fix regressions from #2624. Test any games that were broken by that. (RF4, rabbids kingdom battle)
I've also added a sequence number increment to ThreedClass.IncrementSyncpoint, as it seems to fix buffer corruption in OpenGL homebrew. (this was a regression from removing sequence number increment from constant buffer update - another unrelated resource thing)
* Add tests.
* Add XML docs for GpuRegionHandle
* Skip UpdateProtection if only precise actions were called
This allows precise actions to skip reprotection costs.
* Fast path for Inline2Memory buffer write
This PR adds a method to PhysicalMemory that attempts to write all cached resources directly, so that memory tracking can be avoided. The goal of this is both to avoid flushing buffer data, and to avoid raising the sequence number when data is written, which causes buffer and texture handles to be re-checked.
This currently only targets buffers, with a side check on textures that falls back to a tracked write if any exist within the target range. It's not expected to write textures from here - this is just a mechanism to protect us if someone does decide to do that. It's possible to add a fast path for this in future (and for ShaderCache, once that starts using tracking)
The forced read before inline2memory begins has been skipped, as the data is fully written when the transfer is completed anyways. This allows us to flush on read in emergency situations, but still write the new data over the flushed data.
Improves performance on Xenoblade 2 and DE, which was flushing buffer data on the GPU thread when trying to write compute data. May improve performance in other games that write SSBOs from compute, and update data in the same/nearby pages often.
Super Smash Bros Ultimate should probably be tested to make sure the vertex explosions haven't returned, as I think that's what this AdvanceSequence was for.
* ForceDirty before write, to make sure data does not flush over the new write
* Array based RangeList that caches Address/EndAddress
In isolation, this was more than 2x faster than the RangeList that checks using the interface. In practice I'm seeing much better results than I expected. The array is used because checking it is slightly faster than using a list, which loses time to struct copies, but I still want that data locality.
A method has been added to the list to update the cached end address, as some users of the RangeList currently modify it dynamically.
Greatly improves performance in Super Mario Odyssey, Xenoblade and any other GPU limited games.
* Address Feedback
* Initial Implementation
About as fast as nvidia GL multithreading, can be improved with faster command queuing.
* Struct based command list
Speeds up a bit. Still a lot of time lost to resource copy.
* Do shader init while the render thread is active.
* Introduce circular span pool V1
Ideally should be able to use structs instead of references for storing these spans on commands. Will try that next.
* Refactor SpanRef some more
Use a struct to represent SpanRef, rather than a reference.
* Flush buffers on background thread
* Use a span for UpdateRenderScale.
Much faster than copying the array.
* Calculate command size using reflection
* WIP parallel shaders
* Some minor optimisation
* Only 2 max refs per command now.
The command with 3 refs is gone. 😌
* Don't cast on the GPU side
* Remove redundant casts, force sync on window present
* Fix Shader Cache
* Fix host shader save.
* Fixup to work with new renderer stuff
* Make command Run static, use array of delegates as lookup
Profile says this takes less time than the previous way.
* Bring up to date
* Add settings toggle. Fix Muiltithreading Off mode.
* Fix warning.
* Release tracking lock for flushes
* Fix Conditional Render fast path with threaded gal
* Make handle iteration safe when releasing the lock
This is mostly temporary.
* Attempt to set backend threading on driver
Only really works on nvidia before launching a game.
* Fix race condition with BufferModifiedRangeList, exceptions in tracking actions
* Update buffer set commands
* Some cleanup
* Only use stutter workaround when using opengl renderer non-threaded
* Add host-conditional reservation of counter events
There has always been the possibility that conditional rendering could use a query object just as it is disposed by the counter queue. This change makes it so that when the host decides to use host conditional rendering, the query object is reserved so that it cannot be deleted. Counter events can optionally start reserved, as the threaded implementation can reserve them before the backend creates them, and there would otherwise be a short amount of time where the counter queue could dispose the event before a call to reserve it could be made.
* Address Feedback
* Make counter flush tracked again.
Hopefully does not cause any issues this time.
* Wait for FlushTo on the main queue thread.
Currently assumes only one thread will want to FlushTo (in this case, the GPU thread)
* Add SDL2 headless integration
* Add HLE macro commands.
Co-authored-by: Mary <mary@mary.zone>
* Add support for HLE macros and accelerate MultiDrawElementsIndirectCount
* Add missing barrier
* Fix index buffer count
* Add support check for each macro hle before use
* Add missing xml doc
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
* Replace BGRA and scale uniforms with a uniform block
* Setting the data again on program change is no longer needed
* Optimize and resolve some warnings
* Avoid redundant support buffer updates
* Some optimizations to BindBuffers (now inlined)
* Unify render scale arrays
* Return mapped buffer pointer directly for flush, WriteableRegion for textures
A few changes here to generally improve performance, even for platforms not using the persistent buffer flush.
- Texture and buffer flush now return a ReadOnlySpan<byte>. It's guaranteed that this span is pinned in memory, but it will be overwritten on the next flush from that thread, so it is expected that the data is used before calling again.
- As a result, persistent mappings no longer copy to a new array - rather the persistent map is returned directly as a Span<>. A similar host array is used for the glGet flushes instead of allocating new arrays each time.
- Texture flushes now do their layout conversion into a WriteableRegion when the texture is not MultiRange, which allows the flush to happen directly into guest memory rather than into a temporary span, then copied over. This avoids another copy when doing layout conversion.
Overall, this saves 1 data copy for buffer flush, 1 copy for linear textures with matching source/target stride, and 2 copies for block textures or linear textures with mismatching strides.
* Fix tests
* Fix array pointer for Mesa/Intel path
* Address some feedback
* Update method for getting array pointer.
* 3D engine now uses DeviceState too, plus new state modification tracking
* Remove old methods code
* Remove GpuState and friends
* Optimize DeviceState, force inline some functions
* This change was not supposed to go in
* Proper channel initialization
* Optimize state read/write methods even more
* Fix debug build
* Do not dirty state if the write is redundant
* The YControl register should dirty either the viewport or front face state too, to update the host origin
* Avoid redundant vertex buffer updates
* Move state and get rid of the Ryujinx.Graphics.Gpu.State namespace
* Comments and nits
* Fix rebase
* PR feedback
* Move changed = false to improve codegen
* PR feedback
* Carry RyuJIT a bit more
* Use DeviceState for compute and i2m
* Migrate 2D class, more comments
* Migrate DMA copy engine
* Remove now unused code
* Replace GpuState by GpuAccessorState on GpuAcessor, since compute no longer has a GpuState
* More comments
* Add logging (disabled)
* Add back i2m on 3D engine
* Make GPU memory manager a member of GPU channel
* Move physical memory instance to the memory manager, and the caches to the physical memory
* PR feedback
* Ground work for separate GPU channels
* Rename TextureManager to TextureCache
* Decouple texture bindings management from the texture cache
* Rename BufferManager to BufferCache
* Decouple buffer bindings management from the buffer cache
* More comments and proper disposal
* PR feedback
* Force host state update on channel switch
* Typo
* PR feedback
* Missing using
This greatly speeds up games that constantly resize buffers, and removes stuttering on games that resize large buffers occasionally:
- Large improvement on Super Mario 3D All-Stars (#1663 needed for best performance)
- Improvement to Hyrule Warriors: AoC, and UE4 games. These games can still stutter due to texture creation/loading.
- Small improvement to other games, potential 1-frame stutters avoided.
`ForceSynchronizeMemory`, which was added with POWER, is no longer needed. Some tests have been added for the MultiRegionHandle.
* Refactoring of KMemoryManager class
* Replace some trivial uses of DRAM address with VA
* Get rid of GetDramAddressFromVa
* Abstracting more operations on derived page table class
* Run auto-format on KPageTableBase
* Managed to make TryConvertVaToPa private, few uses remains now
* Implement guest physical pages ref counting, remove manual freeing
* Make DoMmuOperation private and call new abstract methods only from the base class
* Pass pages count rather than size on Map/UnmapMemory
* Change memory managers to take host pointers
* Fix a guest memory leak and simplify KPageTable
* Expose new methods for host range query and mapping
* Some refactoring of MapPagesFromClientProcess to allow proper page ref counting and mapping without KPageLists
* Remove more uses of AddVaRangeToPageList, now only one remains (shared memory page checking)
* Add a SharedMemoryStorage class, will be useful for host mapping
* Sayonara AddVaRangeToPageList, you served us well
* Start to implement host memory mapping (WIP)
* Support memory tracking through host exception handling
* Fix some access violations from HLE service guest memory access and CPU
* Fix memory tracking
* Fix mapping list bugs, including a race and a error adding mapping ranges
* Simple page table for memory tracking
* Simple "volatile" region handle mode
* Update UBOs directly (experimental, rough)
* Fix the overlap check
* Only set non-modified buffers as volatile
* Fix some memory tracking issues
* Fix possible race in MapBufferFromClientProcess (block list updates were not locked)
* Write uniform update to memory immediately, only defer the buffer set.
* Fix some memory tracking issues
* Pass correct pages count on shared memory unmap
* Armeilleure Signal Handler v1 + Unix changes
Unix currently behaves like windows, rather than remapping physical
* Actually check if the host platform is unix
* Fix decommit on linux.
* Implement windows 10 placeholder shared memory, fix a buffer issue.
* Make PTC version something that will never match with master
* Remove testing variable for block count
* Add reference count for memory manager, fix dispose
Can still deadlock with OpenAL
* Add address validation, use page table for mapped check, add docs
Might clean up the page table traversing routines.
* Implement batched mapping/tracking.
* Move documentation, fix tests.
* Cleanup uniform buffer update stuff.
* Remove unnecessary assignment.
* Add unsafe host mapped memory switch
On by default. Would be good to turn this off for untrusted code (homebrew, exefs mods) and give the user the option to turn it on manually, though that requires some UI work.
* Remove C# exception handlers
They have issues due to current .NET limitations, so the meilleure one fully replaces them for now.
* Fix MapPhysicalMemory on the software MemoryManager.
* Null check for GetHostAddress, docs
* Add configuration for setting memory manager mode (not in UI yet)
* Add config to UI
* Fix type mismatch on Unix signal handler code emit
* Fix 6GB DRAM mode.
The size can be greater than `uint.MaxValue` when the DRAM is >4GB.
* Address some feedback.
* More detailed error if backing memory cannot be mapped.
* SetLastError on all OS functions for consistency
* Force pages dirty with UBO update instead of setting them directly.
Seems to be much faster across a few games. Need retesting.
* Rebase, configuration rework, fix mem tracking regression
* Fix race in FreePages
* Set memory managers null after decrementing ref count
* Remove readonly keyword, as this is now modified.
* Use a local variable for the signal handler rather than a register.
* Fix bug with buffer resize, and index/uniform buffer binding.
Should fix flickering in games.
* Add InvalidAccessHandler to MemoryTracking
Doesn't do anything yet
* Call invalid access handler on unmapped read/write.
Same rules as the regular memory manager.
* Make unsafe mapped memory its own MemoryManagerType
* Move FlushUboDirty into UpdateState.
* Buffer dirty cache, rather than ubo cache
Much cleaner, may be reusable for Inline2Memory updates.
* This doesn't return anything anymore.
* Add sigaction remove methods, correct a few function signatures.
* Return empty list of physical regions for size 0.
* Also on AddressSpaceManager
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
* Improve Buffer Textures and flush Image Stores
Fixes a number of issues with buffer textures:
- Reworked Buffer Textures to create their buffers in the TextureManager, then bind them with the BufferManager later.
- Fixes an issue where a buffer texture's buffer could be invalidated after it is bound, but before use.
- Fixed width unpacking for large buffer textures. The width is now 32-bit rather than 16.
- Force buffer textures to be rebound whenever any buffer is created, as using the handle id wasn't reliable, and the cost of binding isn't too high.
Fixes vertex explosions and flickering animations in UE4 games.
* Set ImageStore flag... for ImageStore.
* Check the offset and size.
* Initial implementation (3d tex mips broken)
This works rather well for most games, just need to fix 3d texture mips.
* Cleanup
* Address feedback
* Copy Dependencies and various other fixes
* Fix layer/level offset for copy from view<->view.
* Remove dirty flag from dependency
The dirty flag behaviour is not needed - DeferredCopy is all we need.
* Fix tracking mip slices.
* Propagate granularity (fix astral chain)
* Address Feedback pt 1
* Save slice sizes as part of SizeInfo
* Fix nits
* Fix disposing multiple dependencies causing a crash
This list is obviously modified when removing dependencies, so create a copy of it.
* Initial implementation of buffer flush (VERY WIP)
* Host shaders need to be rebuilt for the SSBO write flag.
* New approach with reserved regions and gl sync
* Fix a ton of buffer issues.
* Remove unused buffer unmapped behaviour
* Revert "Remove unused buffer unmapped behaviour"
This reverts commit f1700e52fb8760180ac5e0987a07d409d1e70ece.
* Delete modified ranges on unmap
Fixes potential crashes in Super Smash Bros, where a previously modified range could lie on either side of an unmap.
* Cache some more delegates.
* Dispose Sync on Close
* Also create host sync for GPFifo syncpoint increment.
* Copy buffer optimization, add docs
* Fix race condition with OpenGL Sync
* Enable read tracking on CommandBuffer, insert syncpoint on WaitForIdle
* Performance: Only flush individual pages of SSBO at a time
This avoids flushing large amounts of data when only a small amount is actually used.
* Signal Modified rather than flushing after clear
* Fix some docs and code style.
* Introduce a new test for tracking memory protection.
Sucessfully demonstrates that the bug causing write protection to be cleared by a read action has been fixed. (these tests fail on master)
* Address Comments
* Add host sync for SetReference
This ensures that any indirect draws will correctly flush any related buffer data written before them. Fixes some flashing and misplaced world geometry in MH rise.
* Make PageAlign static
* Re-enable read tracking, for reads.
* Support for resources on non-contiguous GPU memory regions
* Implement MultiRange physical addresses, only used with a single range for now
* Actually use non-contiguous ranges
* GetPhysicalRegions fixes
* Documentation and remove Address property from TextureInfo
* Finish implementing GetWritableRegion
* Fix typo
* Implement TreeMap from scratch.
Begin implementation of MemoryBlockManager
* Implement GetFreePosition using MemoryBlocks
* Implementation of Memory Management using a Tree.
Still some issues to work around, but promising thus far.
* Resolved invalid mapping issue.
Performance appears promising.
* Add tick metrics
* Use the logger instead
* Use debug loggin instead of info.
* Remove unnecessary code. Add descriptions of added functions.
* Improve memory allocation even further. As well as improve speed of position fetching.
* Add TreeDictionary to Ryujinx Commons
Removed Unnecessary Usigns
* Add a Performance Profiler + Improve ReserveFixed
* Begin transition to allocation in nvdrv
* Create singleton nvmemallocator
* Moved Allocation into Nv Related Files
As requested by gdkchan, any allocation of memory has been moved into the driver files.
Mapping remains in the GPU MemoryManager.
* Remove unnecessary usings
* Add missing descriptions
* Correct descriptions
* Fix formatting.
* Remove unnecessary whitespace
* Formatting / Convention Updates
* Changes / Fixes
Made syntax and convention changes as requested by gdkchan.
Fixed an issue where IsRegionUsed would return the wrong boolean.
Fixed an issue where GetFreePosition was asked for an address instead of a size.
* Undo commenting of Assert in shader cache
* Update Ryujinx.Common/Collections/TreeDictionary.cs
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
* Resolved many suggestions
* Implement Improved TreeDictionary
Based off of Pseudo code and custom implementations.
* Rename _set to _dictionary
* Remove unused code
* Remove unused code.
* Remove unnecessary MapLow function.
* Resolve data-structure based issues
* Make adjustments to memory management.
Deactive de-allocation for now, it causes more harm than good.
* Minor refactorings + Re-implement deallocation
Also cleaned up unnecessary code.
* Add Tests for TreeDictionary
* Update data structure to properly balance the tree
* Experimental Implementation:
1. Reduce Time to Next Node to O(1) Runtime
2. Reduce While Loop Ct To 2 (In Most Cases)
* Address issues w/ Deallocating Memory
* Final Build
+ Fully Implement Dictionary Interface for new Data Structure
+ Cover All Memory Allocation Edge Cases, particularly w/ Games that De-Allocate a lot.
* Minor Corrections
Give TreeDictionary its own count (do not depend on inner dictionary)
Properly remove adjacent allocations
* Add AsList
* Fix bug where internal dictionary wasn't being updated w/ new node for overwritten key.
* Address comments in review.
* Fix issue where block wouldn't break out (Fixes UE4 issues)
* Update descriptions
* Update descriptions
* Reduce Node visibility to protect TreeDictionary Integrity + Remove usage of struct.
* Update tests to use new TreeDictionary implementation.
* Remove usage of dictionary in TreeDictionary
* Refactoring / Renaming
* Remove unneeded memoryblock class.
* Add space for while
* Add space for if
* Formatting / descriptions
* Clarified some descriptions
* Reduce visibility of memory allocator
* Edit method names to make more sense as memory blocks are no longer in use.
* Make names consistent.
* Protect against npe when sucessorof is called against keys that don't exist. (Not in use by memory manager, this is for other prs that might use this data structure)
* Possible edge-case resolve
* Update Ryujinx.Common/Collections/TreeDictionary.cs
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
* Update Ryujinx.HLE/HOS/Services/Nv/NvMemoryAllocator.cs
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
* Reduce # of unnecessary duplicate variables / Reduce visibility of variables only internally used.
* Rename count to _count
* Update Description of Add method.
* Fix copypasta
* Address comments
* Address comments
* Remove whitespace
* Address comments, condense variables.
* Consolidate vars
* Fix whitespace.
* Nit
* Fix exception msg
* Fix arrayIndex check
* Fix arrayIndex check + indexer
* Remove whitespace from cast
Co-authored-by: gdkchan <gab.dark.100@gmail.com>
* IPC refactor part 2: Use ReplyAndReceive on HLE services and remove special handling from kernel
* Fix for applet transfer memory + some nits
* Keep handles if possible to avoid server handle table exhaustion
* Fix IPC ZeroFill bug
* am: Correctly implement CreateManagedDisplayLayer and implement CreateManagedDisplaySeparableLayer
CreateManagedDisplaySeparableLayer is requires since 10.x+ when appletResourceUserId != 0
* Make it exit properly
* Make ServiceNotImplementedException show the full message again
* Allow yielding execution to avoid starving other threads
* Only wait if active
* Merge IVirtualMemoryManager and IAddressSpaceManager
* Fix Ro loading data from the wrong process
Co-authored-by: Thog <me@thog.eu>
* Fix transform feedback errors caused by host pause/resume
* Fix TFB being used as something else issue with copies
* This is supposed to be StreamCopy
* WIP Range Tracking
- Texture invalidation seems to have large problems
- Buffer/Pool invalidation may have problems
- Mirror memory tracking puts an additional `add` in compiled code, we likely just want to make HLE access slower if this is the final solution.
- Native project is in the messiest possible location.
- [HACK] JIT memory access always uses native "fast" path
- [HACK] Trying some things with texture invalidation and views.
It works :)
Still a few hacks, messy things, slow things
More work in progress stuff (also move to memory project)
Quite a bit faster now.
- Unmapping GPU VA and CPU VA will now correctly update write tracking regions, and invalidate textures for the former.
- The Virtual range list is now non-overlapping like the physical one.
- Fixed some bugs where regions could leak.
- Introduced a weird bug that I still need to track down (consistent invalid buffer in MK8 ribbon road)
Move some stuff.
I think we'll eventually just put the dll and so for this in a nuget package.
Fix rebase.
[WIP] MultiRegionHandle variable size ranges
- Avoid reprotecting regions that change often (needs some tweaking)
- There's still a bug in buffers, somehow.
- Might want different api for minimum granularity
Fix rebase issue
Commit everything needed for software only tracking.
Remove native components.
Remove more native stuff.
Cleanup
Use a separate window for the background context, update opentk. (fixes linux)
Some experimental changes
Should get things working up to scratch - still need to try some things with flush/modification and res scale.
Include address with the region action.
Initial work to make range tracking work
Still a ton of bugs
Fix some issues with the new stuff.
* Fix texture flush instability
There's still some weird behaviour, but it's much improved without this. (textures with cpu modified data were flushing over it)
* Find the destination texture for Buffer->Texture full copy
Greatly improves performance for nvdec videos (with range tracking)
* Further improve texture tracking
* Disable Memory Tracking for view parents
This is a temporary approach to better match behaviour on master (where invalidations would be soaked up by views, rather than trigger twice)
The assumption is that when views are created to a texture, they will cover all of its data anyways. Of course, this can easily be improved in future.
* Introduce some tracking tests.
WIP
* Complete base tests.
* Add more tests for multiregion, fix existing test.
* Cleanup Part 1
* Remove unnecessary code from memory tracking
* Fix some inconsistencies with 3D texture rule.
* Add dispose tests.
* Use a background thread for the background context.
Rather than setting and unsetting a context as current, doing the work on a dedicated thread with signals seems to be a bit faster.
Also nerf the multithreading test a bit.
* Copy to texture with matching alignment
This extends the copy to work for some videos with unusual size, such as tutorial videos in SMO. It will only occur if the destination texture already exists at XCount size.
* Track reads for buffer copies. Synchronize new buffers before copying overlaps.
* Remove old texture flushing mechanisms.
Range tracking all the way, baby.
* Wake the background thread when disposing.
Avoids a deadlock when games are closed.
* Address Feedback 1
* Separate TextureCopy instance for background thread
Also `BackgroundContextWorker.InBackground` for a more sensible idenfifier for if we're in a background thread.
* Add missing XML docs.
* Address Feedback
* Maybe I should start drinking coffee.
* Some more feedback.
* Remove flush warning, Refocus window after making background context
* Initial implementation. Still pending better valid-overlap handling,
disposed pool, compressed format flush fix.
* Very messy backend resource cache.
* Oops
* Dispose -> Release
* Improve Release/Dispose.
* More rule refinement.
* View compatibility levels as an enum - you can always know if a view is only copy compatible.
* General cleanup.
Use locking on the resource cache, as it is likely to be used by other threads in future.
* Rename resource cache to resource pool.
* Address some of the smaller nits.
* Fix regression with MK8 lens flare
Texture flushes done the old way should trigger memory tracking.
* Use TextureCreateInfo as a key.
It now implements IEquatable and generates a hashcode based on width/height.
* Fix size change for compressed+non-compressed view combos.
Before, this could set either the compressed or non compressed texture with a size with the wrong size, depending on which texture had its size changed. This caused exceptions when flushing the texture.
Now it correctly takes the block size into account, assuming that these textures are only related because a pixel in the non-compressed texture represents a block in the compressed one.
* Implement JD's suggestion for HashCode Combine
Co-authored-by: jduncanator <1518948+jduncanator@users.noreply.github.com>
* Address feedback
* Address feedback.
Co-authored-by: jduncanator <1518948+jduncanator@users.noreply.github.com>
