Add `Copy` to some structs

[CrazyboyQCD]

Godbolt Link
As you can see in the asm output, even set opt-level = 3, if we don't add Copy to structs with all fields Copy derived, in clone() it generates more mov and large struct can't trigger memcpy, I suggest to add Copy to them to alleviate binary size bloated and performance problem.
List of structs that could add Copy (Simple search may miss):

• Dav1dWarpedMotionParams
• Dav1dSequenceHeader
• Rav1dSequenceHeader
• Dav1dSegmentationData
• Rav1dSegmentationData
• Dav1dSegmentationDataSet
• Rav1dSegmentationDataSet
• Dav1dLoopfilterModeRefDeltas
• Rav1dLoopfilterModeRefDeltas
• Rav1dFilmGrainData
• Dav1dFilmGrainData
• Dav1dFrameHeaderFilmGrain
• Rav1dFrameHeaderFilmGrain
• Dav1dFrameHeaderOperatingPoint
• Dav1dFrameHeaderSuperRes
• Rav1dFrameHeaderSuperRes
• Dav1dFrameHeaderTiling
• Rav1dFrameHeaderTiling
• Dav1dFrameHeaderQuant
• Rav1dFrameHeaderQuant
• Dav1dFrameHeaderSegmentation
• Rav1dFrameHeaderSegmentation
• Dav1dFrameHeaderDeltaQ
• Rav1dFrameHeaderDeltaQ
• Dav1dFrameHeaderDeltaLF
• Rav1dFrameHeaderDeltaLF
• Dav1dFrameHeaderDelta
• Rav1dFrameHeaderDelta
• Dav1dFrameHeaderLoopFilter
• Rav1dFrameHeaderLoopFilter
• Dav1dFrameHeaderCdef
• Rav1dFrameHeaderCdef
• Dav1dFrameHeaderRestoration
• Rav1dFrameHeaderRestoration
• Dav1dFrameHeader
• Rav1dFrameSize
• Rav1dFrameSkipMode
• Rav1dFrameHeader
• Rav1dPictureParameters
• AlignedPixelChunk
• CdfContext
• CdfMvContext
• CdfMvComponent
• CdfCoefContext
• CdfModeContext
• CdfModeInterContext
• Rav1dTileGroupHeader
• Av1BlockIntra
• Av1BlockInter1d
• Av1BlockInter2d
• Av1BlockInterNd
• Av1BlockInter

[kkysen]

I didn't realize this happens. Very good catch! We'll definitely look into this. It's a shame rustc optimizes non-Copy types so much worse, though, even when they're otherwise identical.

[CrazyboyQCD]

Opened an issue.

[kkysen]

rust-lang/rust#128081 (comment) was fixed, at least in significant part, by rust-lang/rust#128299, so we should check if this is still an issue for our types.

[kkysen]

rust-lang/rust#128081 (comment) was fixed, at least in significant part, by rust-lang/rust#128299, so we should check if this is still an issue for our types.

This should land in 1.83 on 2024-11-28, so we should just wait until then and upgrade our toolchain.

[CrazyboyQCD]

rust-lang/rust#132356 disabled the optimization, should wait to be fixed.

[tdelmas]

Did it landed in 1.83? Did it fixed the performance?

[tgross35]

1.88 (current beta) has them enabled again. There are still a couple missed cases but the worst ones seem significantly improved.