[x265] [PATCH 00/18] AArch64: Enable building with -flax-vector-conversions=none

[chen]

Hi Hari,




Thank for the explain, agree with you, we can keep your version.




Regards,

Chen

At 2024-08-15 18:58:09, "Hari Limaye" <hari.limaye at arm.com> wrote:
>Hi Chen,
>
>Thank you for reviewing the patches.
>
>>What's reason rename these parameters? in the AArch64 ABI, only first 8 parameters passthrough SIMD registers, compiler will take care these inout parameters.
>
>The reason for separating and renaming these parameters is because the ACLE type that this function "returns" via the output parameters is either int16x8_t or int32x4_t depending on whether HIGH_BIT_DEPTH is set. Explicitly separating the parameters here allows us to use the correct ACLE types for all variables and operations, enabling the use of the compiler flag -flax-vector-conversions=none (which is the primary purpose of this patch series). It is for code sanitisation, rather than optimisation.
>
>>if you want to improve these interpolate functions, how about also improve algorithm?
>
>Thanks for the suggestion here - these patches are intended purely to clean up the code (with only a couple of minor peephole optimisations here and there), to enable future optimisation patches. We have a large set of patches providing significant uplift for interpolation functions that will be pushed after this series.
>
>Many thanks,
>Hari
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20240816/079b23ac/attachment.htm>