[FFmpeg-cvslog] lavc/aarch64: Add pix_abs16_x2 neon implementation

[Hubert Mazur]

ffmpeg | branch: master | Hubert Mazur <hum at semihalf.com> | Tue Jul 12 11:15:33 2022 +0200| [01e190dc9950f4c35a4d5d795736460577807e3f] | committer: Martin Storsjö

lavc/aarch64: Add pix_abs16_x2 neon implementation

Provide neon implementation for pix_abs16_x2 function.

Performance tests of implementation are below.
 - pix_abs_0_1_c: 283.5
 - pix_abs_0_1_neon: 39.0

Benchmarks and tests run with checkasm tool on AWS Graviton 3.

Signed-off-by: Hubert Mazur <hum at semihalf.com>
Signed-off-by: Martin Storsjö <martin at martin.st>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=01e190dc9950f4c35a4d5d795736460577807e3f
---

 libavcodec/aarch64/me_cmp_init_aarch64.c |  3 ++
 libavcodec/aarch64/me_cmp_neon.S         | 75 ++++++++++++++++++++++++++++++++
 2 files changed, 78 insertions(+)

diff --git a/libavcodec/aarch64/me_cmp_init_aarch64.c b/libavcodec/aarch64/me_cmp_init_aarch64.c
index bec9148a1a..136b008eb7 100644
--- a/libavcodec/aarch64/me_cmp_init_aarch64.c
+++ b/libavcodec/aarch64/me_cmp_init_aarch64.c
@@ -27,6 +27,8 @@ int ff_pix_abs16_neon(MpegEncContext *s, uint8_t *blk1, uint8_t *blk2,
                       ptrdiff_t stride, int h);
 int ff_pix_abs16_xy2_neon(MpegEncContext *s, uint8_t *blk1, uint8_t *blk2,
                       ptrdiff_t stride, int h);
+int ff_pix_abs16_x2_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)
 {
@@ -34,6 +36,7 @@ av_cold void ff_me_cmp_init_aarch64(MECmpContext *c, AVCodecContext *avctx)
 
     if (have_neon(cpu_flags)) {
         c->pix_abs[0][0] = ff_pix_abs16_neon;
+        c->pix_abs[0][1] = ff_pix_abs16_x2_neon;
         c->pix_abs[0][3] = ff_pix_abs16_xy2_neon;
 
         c->sad[0] = ff_pix_abs16_neon;
diff --git a/libavcodec/aarch64/me_cmp_neon.S b/libavcodec/aarch64/me_cmp_neon.S
index a7937bd8be..e49d049fc2 100644
--- a/libavcodec/aarch64/me_cmp_neon.S
+++ b/libavcodec/aarch64/me_cmp_neon.S
@@ -203,3 +203,78 @@ function ff_pix_abs16_xy2_neon, export=1
         fmov            w0, s0                      // copy result to general purpose register
         ret
 endfunc
+
+function ff_pix_abs16_x2_neon, export=1
+        // x0           unused
+        // x1           uint8_t *pix1
+        // x2           uint8_t *pix2
+        // x3           ptrdiff_t stride
+        // w4           int h
+
+        cmp             w4, #4
+        // initialize buffers
+        movi            d20, #0
+        add             x5, x2, #1 // pix2 + 1
+        b.lt            2f
+
+// make 4 iterations at once
+1:
+
+        // abs(pix1[0] - avg2(pix2[0], pix2[1]))
+        // avg2(a,b) = (((a) + (b) + 1) >> 1)
+        // abs(x) = (x < 0 ? -x : x)
+
+        ld1             {v1.16b}, [x2], x3
+        ld1             {v2.16b}, [x5], x3
+        urhadd          v30.16b, v1.16b, v2.16b
+        ld1             {v0.16b}, [x1], x3
+        uabdl           v16.8h, v0.8b, v30.8b
+        ld1             {v4.16b}, [x2], x3
+        uabdl2          v17.8h, v0.16b, v30.16b
+        ld1             {v5.16b}, [x5], x3
+        urhadd          v29.16b, v4.16b, v5.16b
+        ld1             {v3.16b}, [x1], x3
+        uabal           v16.8h, v3.8b, v29.8b
+        ld1             {v7.16b}, [x2], x3
+        uabal2          v17.8h, v3.16b, v29.16b
+        ld1             {v22.16b}, [x5], x3
+        urhadd          v28.16b, v7.16b, v22.16b
+        ld1             {v6.16b}, [x1], x3
+        uabal           v16.8h, v6.8b, v28.8b
+        ld1             {v24.16b}, [x2], x3
+        uabal2          v17.8h, v6.16b, v28.16b
+        ld1             {v25.16b}, [x5], x3
+        urhadd          v27.16b, v24.16b, v25.16b
+        ld1             {v23.16b}, [x1], x3
+        uabal           v16.8h, v23.8b, v27.8b
+        uabal2          v17.8h, v23.16b, v27.16b
+
+        sub             w4, w4, #4
+
+        add             v16.8h, v16.8h, v17.8h
+        uaddlv          s16, v16.8h
+        cmp             w4, #4
+        add             d20, d20, d16
+
+        b.ge            1b
+        cbz             w4, 3f
+
+// iterate by one
+2:
+        ld1             {v1.16b}, [x2], x3
+        ld1             {v2.16b}, [x5], x3
+        urhadd          v29.16b, v1.16b, v2.16b
+        ld1             {v0.16b}, [x1], x3
+        uabd            v28.16b, v0.16b, v29.16b
+
+        uaddlv          h28, v28.16b
+        subs            w4, w4, #1
+
+        add             d20, d20, d28
+        b.ne            2b
+
+3:
+        fmov            w0, s20
+
+        ret
+endfunc