[x265] [PATCH 00/12] AArch64: Optimise low bitdepth ipfilter primitives
chen
chenm003 at 163.com
Mon Sep 2 03:49:38 UTC 2024
Hi Hari,
Thank for the patches, I have no comment on this initialize version.
In future version, we may improve by algorithm.
For example, width=12 may split into 2 of witdh=8, it got cache benefits.
Regards,
Chen
At 2024-08-31 03:18:48, "Hari Limaye" <hari.limaye at arm.com> wrote:
>This patch series optimises the existing Neon intrinsics implementations of the ipfilter primitives, and removes the assembly implementations in favour of these new implementations.
>
>Relative performance observed for the new Neon intrinsics implementations, compared to the existing assembly implementations, is in the respective commit messages.
>
>Many thanks,
>Hari
>
>Hari Limaye (12):
> Test: Remove check for unused coeffIdx in ipfilter tests
> Move ipfilter primitives into X265_NS
> AArch64: Move ipfilter primitives into X265_NS
> AArch64: Support all block sizes in p2s Neon
> AArch64: Optimise low bitdepth interp_horiz_pp_neon
> AArch64: Optimise low bitdepth interp_horiz_ps_neon
> AArch64: Optimise low bitdepth interp_vert_ss_neon
> AArch64: Optimise low bitdepth interp_vert_pp_neon
> AArch64: Optimise low bitdepth interp_vert_ps_neon
> AArch64: Optimise low bitdepth interp_vert_sp_neon
> AArch64: Define all low bitdepth Neon ipfilter primitives
> AArch64: Remove Assembly ipfilter primitives
>
> source/common/CMakeLists.txt | 4 +-
> source/common/aarch64/asm-primitives.cpp | 186 --
> source/common/aarch64/filter-prim.cpp | 2877 ++++++++++++++++++----
> source/common/aarch64/fun-decls.h | 15 -
> source/common/aarch64/ipfilter-common.S | 1436 -----------
> source/common/aarch64/ipfilter-sve2.S | 1282 ----------
> source/common/aarch64/ipfilter.S | 1054 --------
> source/common/aarch64/mem-neon.h | 193 ++
> source/common/ipfilter.cpp | 8 +-
> source/test/ipfilterharness.cpp | 24 +-
> 10 files changed, 2580 insertions(+), 4499 deletions(-)
> delete mode 100644 source/common/aarch64/ipfilter-common.S
> delete mode 100644 source/common/aarch64/ipfilter-sve2.S
> delete mode 100644 source/common/aarch64/ipfilter.S
>
>--
>2.42.1
>
>_______________________________________________
>x265-devel mailing list
>x265-devel at videolan.org
>https://mailman.videolan.org/listinfo/x265-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20240902/1c16aff3/attachment.htm>
More information about the x265-devel
mailing list