[FFmpeg-devel] [PATCH 2/5] lavc/aarch64: Add neon implementation for sse4

[Martin Storsjö]

On Fri, 15 Jul 2022, Hubert Mazur wrote:

> Provide neon implementation for sse4 function.
>
> Performance comparison tests are shown below.
> - sse_2_c: 74.0
> - sse_2_neon: 24.0
>
> Benchmarks and tests are run with checkasm tool on AWS Graviton 3.
>
> Signed-off-by: Hubert Mazur <hum at semihalf.com>
> ---
> libavcodec/aarch64/me_cmp_init_aarch64.c |  3 ++
> libavcodec/aarch64/me_cmp_neon.S         | 65 ++++++++++++++++++++++++
> 2 files changed, 68 insertions(+)
>
> diff --git a/libavcodec/aarch64/me_cmp_init_aarch64.c b/libavcodec/aarch64/me_cmp_init_aarch64.c
> index 3ff5767bd0..72a2062e7e 100644
> --- a/libavcodec/aarch64/me_cmp_init_aarch64.c
> +++ b/libavcodec/aarch64/me_cmp_init_aarch64.c
> @@ -32,6 +32,8 @@ int ff_pix_abs16_x2_neon(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
> 
> int sse16_neon(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
>                   ptrdiff_t stride, int h);
> +int sse4_neon(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
> +                  ptrdiff_t stride, int h);
> 
> av_cold void ff_me_cmp_init_aarch64(MECmpContext *c, AVCodecContext *avctx)
> {
> @@ -44,5 +46,6 @@ av_cold void ff_me_cmp_init_aarch64(MECmpContext *c, AVCodecContext *avctx)
>
>         c->sad[0] = ff_pix_abs16_neon;
>         c->sse[0] = sse16_neon;
> +        c->sse[2] = sse4_neon;
>     }
> }
> diff --git a/libavcodec/aarch64/me_cmp_neon.S b/libavcodec/aarch64/me_cmp_neon.S
> index 88cd335443..bacf151314 100644
> --- a/libavcodec/aarch64/me_cmp_neon.S
> +++ b/libavcodec/aarch64/me_cmp_neon.S
> @@ -360,3 +360,68 @@ function sse16_neon, export=1
>         ret
> 
> endfunc
> +
> +function sse4_neon, export=1
> +        // x0 - unused
> +        // x1 - pix1
> +        // x2 - pix2
> +        // x3 - stride
> +        // w4 - h
> +
> +        movi            d18, #0
> +        movi            d17, #0
> +        cmp             w4, #4
> +        b.le            2f
> +
> +// make 4 iterations at once
> +1:
> +
> +        // res = abs(pix1[0] - pix2[0])
> +        // res * res
> +
> +        ld1             {v0.4b}, [x1], x3

This fails to assemble for me with essentially all tools I have (old 
binutils, moderately recent binutils, current llvm, MS armasm64.exe):

src/libavcodec/aarch64/me_cmp_neon.S:374: Error: operand mismatch -- `ld1 
{v0.4b},[x1],x3'
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:    did you mean this?
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.8b}, [x1], x3
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:    other valid variant(s):
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.16b}, [x1], x3
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.4h}, [x1], x3
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.8h}, [x1], x3
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.2s}, [x1], x3
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.4s}, [x1], x3
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.1d}, [x1], x3
src/libavcodec/aarch64/me_cmp_neon.S:374: Info:         ld1 {v0.2d}, [x1], x3


I'll follow up with an actual review of the patches later. I'm sorry I 
have a bit longer review latency than usual at the moment, as I'm on 
vacation.

// Martin