[FFmpeg-devel] [PATCH 3/5] aarch64: me_cmp: Interleave some of the loads in ff_pix_abs16_xy2_neon

[Swinney, Jonathan]

This looks good to me. As you can see it makes very little difference on Graviton 3, which is why I had not already done it. :)

-- 

Jonathan Swinney

On 7/13/22, 3:49 PM, "ffmpeg-devel on behalf of Martin Storsjö" <ffmpeg-devel-bounces at ffmpeg.org on behalf of martin at martin.st> wrote:

    CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.



    Before:       Cortex A53    A72    A73   Graviton 3
    pix_abs_0_3_neon:  183.7  112.7   97.5   41.2
    After:
    pix_abs_0_3_neon:  175.7  109.2   92.0   41.2
    ---
     libavcodec/aarch64/me_cmp_neon.S | 22 ++++++++++++++--------
     1 file changed, 14 insertions(+), 8 deletions(-)

    diff --git a/libavcodec/aarch64/me_cmp_neon.S b/libavcodec/aarch64/me_cmp_neon.S
    index 31db3793d9..0ae23d8922 100644
    --- a/libavcodec/aarch64/me_cmp_neon.S
    +++ b/libavcodec/aarch64/me_cmp_neon.S
    @@ -101,26 +101,32 @@ function ff_pix_abs16_xy2_neon, export=1
             ld1             {v6.16b}, [x5], x3          // load pix3
             ld1             {v16.16b}, [x1], x3         // load pix1

    -        ldur            q19, [x5, #1]               // load pix3+1
    -        ld1             {v18.16b}, [x5], x3         // load pix3
    -        ld1             {v17.16b}, [x1], x3         // load pix1
    -
    -        ldur            q22, [x5, #1]               // load pix3+1
    -        ld1             {v21.16b}, [x5], x3         // load pix3
    -        ld1             {v20.16b}, [x1], x3         // load pix1
    -
             // These blocks compute the average: avg(pix2[n], pix2[n+1], pix3[n], pix3[n+1])
             uaddl           v30.8h, v4.8b, v5.8b        // pix3 + pix3+1 0..7
             uaddl2          v31.8h, v4.16b, v5.16b      // pix3 + pix3+1 8..15
    +
    +        ldur            q19, [x5, #1]               // load pix3+1
    +
             add             v23.8h, v2.8h, v30.8h       // add up 0..7, using pix2 + pix2+1 values from previous iteration
             add             v24.8h, v3.8h, v31.8h       // add up 8..15, using pix2 + pix2+1 values from previous iteration
    +
    +        ld1             {v18.16b}, [x5], x3         // load pix3
    +        ld1             {v17.16b}, [x1], x3         // load pix1
    +
             rshrn           v23.8b, v23.8h, #2          // shift right 2 0..7 (rounding shift right)
             rshrn2          v23.16b, v24.8h, #2         // shift right 2 8..15

             uaddl           v2.8h, v6.8b, v7.8b         // pix3 + pix3+1 0..7
             uaddl2          v3.8h, v6.16b, v7.16b       // pix3 + pix3+1 8..15
    +
    +        ldur            q22, [x5, #1]               // load pix3+1
    +
             add             v26.8h, v30.8h, v2.8h       // add up 0..7, using pix2 + pix2+1 values from pix3 above
             add             v27.8h, v31.8h, v3.8h       // add up 8..15, using pix2 + pix2+1 values from pix3 above
    +
    +        ld1             {v21.16b}, [x5], x3         // load pix3
    +        ld1             {v20.16b}, [x1], x3         // load pix1
    +
             rshrn           v26.8b, v26.8h, #2          // shift right 2 0..7 (rounding shift right)
             rshrn2          v26.16b, v27.8h, #2         // shift right 2 8..15

    --
    2.25.1

    _______________________________________________
    ffmpeg-devel mailing list
    ffmpeg-devel at ffmpeg.org
    https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

    To unsubscribe, visit link above, or email
    ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".