lavc/aarch64: new optimization for 8-bit hevc_epel_v

checkasm bench:
put_hevc_epel_v4_8_c: 79.9
put_hevc_epel_v4_8_neon: 25.7
put_hevc_epel_v6_8_c: 151.4
put_hevc_epel_v6_8_neon: 46.4
put_hevc_epel_v8_8_c: 250.9
put_hevc_epel_v8_8_neon: 41.7
put_hevc_epel_v12_8_c: 542.7
put_hevc_epel_v12_8_neon: 108.7
put_hevc_epel_v16_8_c: 939.4
put_hevc_epel_v16_8_neon: 169.2
put_hevc_epel_v24_8_c: 2104.9
put_hevc_epel_v24_8_neon: 307.9
put_hevc_epel_v32_8_c: 3713.9
put_hevc_epel_v32_8_neon: 524.2
put_hevc_epel_v48_8_c: 8175.2
put_hevc_epel_v48_8_neon: 1197.2
put_hevc_epel_v64_8_c: 16049.4
put_hevc_epel_v64_8_neon: 2094.9

Co-Authored-By: J. Dekker <jdek@itanimul.li>
Signed-off-by: Martin Storsjö <martin@martin.st>

2 files changed, 228 insertions, 0 deletions

diff --git a/libavcodec/aarch64/hevcdsp_epel_neon.S b/libavcodec/aarch64/hevcdsp_epel_neon.S
index 0705213eed..96ec58cdbc 100644
--- a/libavcodec/aarch64/hevcdsp_epel_neon.S
+++ b/libavcodec/aarch64/hevcdsp_epel_neon.S
@@ -243,6 +243,229 @@ function ff_hevc_put_hevc_pel_pixels64_8_neon, export=1
         ret
 endfunc
 
+
+function ff_hevc_put_hevc_epel_v4_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #(MAX_PB_SIZE * 2)
+        ldr             s16, [x1]
+        ldr             s17, [x1, x2]
+        add             x1, x1, x2, lsl #1
+        ld1             {v18.s}[0], [x1], x2
+.macro calc src0, src1, src2, src3
+        ld1             {\src3\().s}[0], [x1], x2
+        movi            v4.8h, #0
+        calc_epelb      v4, \src0, \src1, \src2, \src3
+        subs            w3, w3, #1
+        st1             {v4.4h}, [x0], x10
+.endm
+1:      calc_all4
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v6_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #(MAX_PB_SIZE * 2 - 8)
+        ldr             d16, [x1]
+        ldr             d17, [x1, x2]
+        add             x1, x1, x2, lsl #1
+        ld1             {v18.8b}, [x1], x2
+.macro calc src0, src1, src2, src3
+        ld1             {\src3\().8b}, [x1], x2
+        movi            v4.8h, #0
+        calc_epelb      v4, \src0, \src1, \src2, \src3
+        st1             {v4.d}[0], [x0], #8
+        subs            w3, w3, #1
+        st1             {v4.s}[2], [x0], x10
+.endm
+1:      calc_all4
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v8_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #(MAX_PB_SIZE * 2)
+        ldr             d16, [x1]
+        ldr             d17, [x1, x2]
+        add             x1, x1, x2, lsl #1
+        ld1             {v18.8b}, [x1], x2
+.macro calc src0, src1, src2, src3
+        ld1             {\src3\().8b}, [x1], x2
+        movi            v4.8h, #0
+        calc_epelb      v4, \src0, \src1, \src2, \src3
+        subs            w3, w3, #1
+        st1             {v4.8h}, [x0], x10
+.endm
+1:      calc_all4
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v12_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #(MAX_PB_SIZE * 2)
+        ldr             q16, [x1]
+        ldr             q17, [x1, x2]
+        add             x1, x1, x2, lsl #1
+        ld1             {v18.16b}, [x1], x2
+.macro calc src0, src1, src2, src3
+        ld1             {\src3\().16b}, [x1], x2
+        movi            v4.8h, #0
+        movi            v5.8h, #0
+        calc_epelb      v4, \src0, \src1, \src2, \src3
+        calc_epelb2     v5, \src0, \src1, \src2, \src3
+        str             q4, [x0]
+        subs            w3, w3, #1
+        str             d5, [x0, #16]
+        add             x0, x0, x10
+.endm
+1:      calc_all4
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v16_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #(MAX_PB_SIZE * 2)
+        ldr             q16, [x1]
+        ldr             q17, [x1, x2]
+        add             x1, x1, x2, lsl #1
+        ld1             {v18.16b}, [x1], x2
+.macro calc src0, src1, src2, src3
+        ld1             {\src3\().16b}, [x1], x2
+        movi            v4.8h, #0
+        movi            v5.8h, #0
+        calc_epelb      v4, \src0, \src1, \src2, \src3
+        calc_epelb2     v5, \src0, \src1, \src2, \src3
+        subs            w3, w3, #1
+        st1             {v4.8h, v5.8h}, [x0], x10
+.endm
+1:      calc_all4
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v24_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #(MAX_PB_SIZE * 2)
+        ld1             {v16.8b, v17.8b, v18.8b}, [x1], x2
+        ld1             {v19.8b, v20.8b, v21.8b}, [x1], x2
+        ld1             {v22.8b, v23.8b, v24.8b}, [x1], x2
+.macro calc src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10, src11
+        ld1             {\src9\().8b, \src10\().8b, \src11\().8b}, [x1], x2
+        movi            v4.8h, #0
+        movi            v5.8h, #0
+        movi            v6.8h, #0
+        calc_epelb      v4, \src0, \src3, \src6, \src9
+        calc_epelb      v5, \src1, \src4, \src7, \src10
+        calc_epelb      v6, \src2, \src5, \src8, \src11
+        subs            w3, w3, #1
+        st1             {v4.8h-v6.8h}, [x0], x10
+.endm
+1:      calc_all12
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v32_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #(MAX_PB_SIZE * 2)
+        ld1             {v16.16b, v17.16b}, [x1], x2
+        ld1             {v18.16b, v19.16b}, [x1], x2
+        ld1             {v20.16b, v21.16b}, [x1], x2
+.macro calc src0, src1, src2, src3, src4, src5, src6, src7
+        ld1             {\src6\().16b, \src7\().16b}, [x1], x2
+        movi            v4.8h, #0
+        movi            v5.8h, #0
+        movi            v6.8h, #0
+        movi            v7.8h, #0
+        calc_epelb      v4, \src0, \src2, \src4, \src6
+        calc_epelb2     v5, \src0, \src2, \src4, \src6
+        calc_epelb      v6, \src1, \src3, \src5, \src7
+        calc_epelb2     v7, \src1, \src3, \src5, \src7
+        subs            w3, w3, #1
+        st1             {v4.8h-v7.8h}, [x0], x10
+.endm
+1:      calc_all8
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v48_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             x1, x1, x2
+        mov             x10, #64
+        ld1             {v16.16b, v17.16b, v18.16b}, [x1], x2
+        ld1             {v19.16b, v20.16b, v21.16b}, [x1], x2
+        ld1             {v22.16b, v23.16b, v24.16b}, [x1], x2
+.macro calc src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10, src11
+        ld1             {\src9\().16b, \src10\().16b, \src11\().16b}, [x1], x2
+        movi            v4.8h, #0
+        movi            v5.8h, #0
+        movi            v6.8h, #0
+        movi            v7.8h, #0
+        movi            v28.8h, #0
+        movi            v29.8h, #0
+        calc_epelb      v4,  \src0, \src3, \src6, \src9
+        calc_epelb2     v5,  \src0, \src3, \src6, \src9
+        calc_epelb      v6,  \src1, \src4, \src7, \src10
+        calc_epelb2     v7,  \src1, \src4, \src7, \src10
+        calc_epelb      v28, \src2, \src5, \src8, \src11
+        calc_epelb2     v29, \src2, \src5, \src8, \src11
+        st1             {v4.8h-v7.8h}, [x0], #64
+        subs            w3, w3, #1
+        st1             {v28.8h-v29.8h}, [x0], x10
+.endm
+1:      calc_all12
+.purgem calc
+2:      ret
+endfunc
+
+function ff_hevc_put_hevc_epel_v64_8_neon, export=1
+        load_epel_filterb x5, x4
+        sub             sp, sp, #32
+        st1             {v8.8b-v11.8b}, [sp]
+        sub             x1, x1, x2
+        ld1             {v16.16b, v17.16b, v18.16b, v19.16b}, [x1], x2
+        ld1             {v20.16b, v21.16b, v22.16b, v23.16b}, [x1], x2
+        ld1             {v24.16b, v25.16b, v26.16b, v27.16b}, [x1], x2
+.macro calc src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10, src11, src12, src13, src14, src15
+        ld1             {\src12\().16b-\src15\().16b}, [x1], x2
+        movi            v4.8h, #0
+        movi            v5.8h, #0
+        movi            v6.8h, #0
+        movi            v7.8h, #0
+        movi            v8.8h, #0
+        movi            v9.8h, #0
+        movi            v10.8h, #0
+        movi            v11.8h, #0
+        calc_epelb      v4,  \src0, \src4, \src8,  \src12
+        calc_epelb2     v5,  \src0, \src4, \src8,  \src12
+        calc_epelb      v6,  \src1, \src5, \src9,  \src13
+        calc_epelb2     v7,  \src1, \src5, \src9,  \src13
+        calc_epelb      v8,  \src2, \src6, \src10, \src14
+        calc_epelb2     v9,  \src2, \src6, \src10, \src14
+        calc_epelb      v10, \src3, \src7, \src11, \src15
+        calc_epelb2     v11, \src3, \src7, \src11, \src15
+        st1             {v4.8h-v7.8h}, [x0], #64
+        subs            w3, w3, #1
+        st1             {v8.8h-v11.8h}, [x0], #64
+.endm
+1:      calc_all16
+.purgem calc
+2:      ld1             {v8.8b-v11.8b}, [sp]
+        add             sp, sp, #32
+        ret
+endfunc
+
 function ff_hevc_put_hevc_epel_uni_v4_8_neon, export=1
         load_epel_filterb x6, x5
         sub             x2, x2, x3
diff --git a/libavcodec/aarch64/hevcdsp_init_aarch64.c b/libavcodec/aarch64/hevcdsp_init_aarch64.c
index c203d65d34..42aa76ddde 100644
--- a/libavcodec/aarch64/hevcdsp_init_aarch64.c
+++ b/libavcodec/aarch64/hevcdsp_init_aarch64.c
@@ -156,6 +156,10 @@ NEON8_FNPROTO(pel_pixels, (int16_t *dst,
         const uint8_t *src, ptrdiff_t srcstride,
         int height, intptr_t mx, intptr_t my, int width),);
 
+NEON8_FNPROTO(epel_v, (int16_t *dst,
+        const uint8_t *src, ptrdiff_t srcstride,
+        int height, intptr_t mx, intptr_t my, int width),);
+
 NEON8_FNPROTO(pel_uni_pixels, (uint8_t *_dst, ptrdiff_t _dststride,
         const uint8_t *_src, ptrdiff_t _srcstride,
         int height, intptr_t mx, intptr_t my, int width),);
@@ -305,6 +309,7 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->put_hevc_qpel_bi[9][0][1]   = ff_hevc_put_hevc_qpel_bi_h16_8_neon;
 
         NEON8_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels,);
+        NEON8_FNASSIGN(c->put_hevc_epel, 1, 0, epel_v,);
         NEON8_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels,);
         NEON8_FNASSIGN(c->put_hevc_epel_uni, 0, 0, pel_uni_pixels,);
         NEON8_FNASSIGN(c->put_hevc_epel_uni, 1, 0, epel_uni_v,);