Quelle  h264pred_neon.S

/*
 * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/aarch64/asm.S"

.macro ldcol.8  rd,  rs,  rt,  n=8,  hi=0
.if \n >= 8 || \hi == 0
        ld1             {\rd\().b}[0],  [\rs], \rt
        ld1             {\rd\().b}[1],  [\rs], \rt
        ld1             {\rd\().b}[2],  [\rs], \rt
        ld1             {\rd\().b}[3],  [\rs], \rt
.endif
.if \n >= 8 || \hi == 1
        ld1             {\rd\().b}[4],  [\rs], \rt
        ld1             {\rd\().b}[5],  [\rs], \rt
        ld1             {\rd\().b}[6],  [\rs], \rt
        ld1             {\rd\().b}[7],  [\rs], \rt
.endif
.if \n == 16
        ld1             {\rd\().b}[8],  [\rs], \rt
        ld1             {\rd\().b}[9],  [\rs], \rt
        ld1             {\rd\().b}[10], [\rs], \rt
        ld1             {\rd\().b}[11], [\rs], \rt
        ld1             {\rd\().b}[12], [\rs], \rt
        ld1             {\rd\().b}[13], [\rs], \rt
        ld1             {\rd\().b}[14], [\rs], \rt
        ld1             {\rd\().b}[15], [\rs], \rt
.endif
.endm

function ff_pred16x16_128_dc_neon, export=1
        movi            v0.16b,  #128
        b               .L_pred16x16_dc_end
endfunc

function ff_pred16x16_top_dc_neon, export=1
        sub             x2,  x0,  x1
        ld1             {v0.16b},  [x2]
        uaddlv          h0,  v0.16b
        rshrn           v0.8b,  v0.8h,  #4
        dup             v0.16b, v0.b[0]
        b               .L_pred16x16_dc_end
endfunc

function ff_pred16x16_left_dc_neon, export=1
        sub             x2,  x0,  #1
        ldcol.8         v0,  x2,  x1, 16
        uaddlv          h0,  v0.16b
        rshrn           v0.8b,  v0.8h,  #4
        dup             v0.16b, v0.b[0]
        b               .L_pred16x16_dc_end
endfunc

function ff_pred16x16_dc_neon, export=1
        sub             x2,  x0,  x1
        sub             x3,  x0,  #1
        ld1             {v0.16b}, [x2]
        ldcol.8         v1,  x3,  x1, 16
        uaddlv          h0,  v0.16b
        uaddlv          h1,  v1.16b
        add             v0.4h,  v0.4h,  v1.4h
        rshrn           v0.8b,  v0.8h,  #5
        dup             v0.16b, v0.b[0]
.L_pred16x16_dc_end:
        mov             w3,  #8
6:      st1             {v0.16b}, [x0], x1
        subs            w3,  w3,  #1
        st1             {v0.16b}, [x0], x1
        b.ne            6b
        ret
endfunc

function ff_pred16x16_hor_neon, export=1
        sub             x2,  x0,  #1
        mov             w3,  #16
1:      ld1r            {v0.16b}, [x2], x1
        subs            w3,  w3,  #1
        st1             {v0.16b}, [x0], x1
        b.ne            1b
        ret
endfunc

function ff_pred16x16_vert_neon, export=1
        sub             x2,  x0,  x1
        add             x1,  x1,  x1
        ld1             {v0.16b}, [x2], x1
        mov             w3,  #8
1:      subs            w3,  w3,  #1
        st1             {v0.16b}, [x0], x1
        st1             {v0.16b}, [x2], x1
        b.ne            1b
        ret
endfunc

function ff_pred16x16_plane_neon, export=1
        sub             x3,  x0,  x1
        movrel          x4,  p16weight
        add             x2,  x3,  #8
        sub             x3,  x3,  #1
        ld1             {v0.8b},  [x3]
        ld1             {v2.8b},  [x2], x1
        ldcol.8         v1,  x3,  x1
        add             x3,  x3,  x1
        ldcol.8         v3,  x3,  x1
        rev64           v0.8b,  v0.8b
        rev64           v1.8b,  v1.8b
        uaddl           v7.8h,  v2.8b,  v3.8b
        usubl           v2.8h,  v2.8b,  v0.8b
        usubl           v3.8h,  v3.8b,  v1.8b
        ld1             {v0.8h},     [x4]
        mul             v2.8h,  v2.8h,  v0.8h
        mul             v3.8h,  v3.8h,  v0.8h
        addp            v2.8h,  v2.8h,  v3.8h
        addp            v2.8h,  v2.8h,  v2.8h
        addp            v2.4h,  v2.4h,  v2.4h
        sshll           v3.4s,  v2.4h,  #2
        saddw           v2.4s,  v3.4s,  v2.4h
        rshrn           v4.4h,  v2.4s,  #6
        trn2            v5.4h,  v4.4h,  v4.4h
        add             v2.4h,  v4.4h,  v5.4h
        shl             v3.4h,  v2.4h,  #3
        ext             v7.16b, v7.16b, v7.16b, #14
        sub             v3.4h,  v3.4h,  v2.4h   // 7 * (b + c)
        add             v7.4h,  v7.4h,  v0.4h
        shl             v2.4h,  v7.4h,  #4
        sub             v2.4h,  v2.4h,  v3.4h
        shl             v3.4h,  v4.4h,  #4
        ext             v0.16b, v0.16b, v0.16b, #14
        sub             v6.4h,  v5.4h,  v3.4h
        mov             v0.h[0],  wzr
        mul             v0.8h,  v0.8h,  v4.h[0]
        dup             v1.8h,  v2.h[0]
        dup             v2.8h,  v4.h[0]
        dup             v3.8h,  v6.h[0]
        shl             v2.8h,  v2.8h,  #3
        add             v1.8h,  v1.8h,  v0.8h
        add             v3.8h,  v3.8h,  v2.8h
        mov             w3,  #16
1:
        sqshrun         v0.8b,  v1.8h,  #5
        add             v1.8h,  v1.8h,  v2.8h
        sqshrun2        v0.16b, v1.8h,  #5
        add             v1.8h,  v1.8h,  v3.8h
        subs            w3,  w3,  #1
        st1             {v0.16b}, [x0], x1
        b.ne            1b
        ret
endfunc

const   p16weight, align=4
        .short          1,2,3,4,5,6,7,8
endconst
const   p8weight, align=4
        .short          1,2,3,4,1,2,3,4
endconst

function ff_pred8x8_hor_neon, export=1
        sub             x2,  x0,  #1
        mov             w3,  #8
1:      ld1r            {v0.8b},  [x2], x1
        subs            w3,  w3,  #1
        st1             {v0.8b},  [x0], x1
        b.ne            1b
        ret
endfunc

function ff_pred8x8_vert_neon, export=1
        sub             x2,  x0,  x1
        lsl             x1,  x1,  #1
        ld1             {v0.8b},  [x2], x1
        mov             w3,  #4
1:      subs            w3,  w3,  #1
        st1             {v0.8b},  [x0], x1
        st1             {v0.8b},  [x2], x1
        b.ne            1b
        ret
endfunc

function ff_pred8x8_plane_neon, export=1
        sub             x3,  x0,  x1
        movrel          x4,  p8weight
        movrel          x5,  p16weight
        add             x2,  x3,  #4
        sub             x3,  x3,  #1
        ld1             {v0.s}[0],  [x3]
        ld1             {v2.s}[0],  [x2], x1
        ldcol.8         v0,  x3,  x1,  4,  hi=1
        add             x3,  x3,  x1
        ldcol.8         v3,  x3,  x1,  4
        uaddl           v7.8h,  v2.8b,  v3.8b
        rev32           v0.8b,  v0.8b
        trn1            v2.2s,  v2.2s,  v3.2s
        usubl           v2.8h,  v2.8b,  v0.8b
        ld1             {v6.8h},  [x4]
        mul             v2.8h,  v2.8h,  v6.8h
        ld1             {v0.8h},  [x5]
        saddlp          v2.4s,  v2.8h
        addp            v2.4s,  v2.4s,  v2.4s
        shl             v3.4s,  v2.4s,  #4
        add             v2.4s,  v3.4s,  v2.4s
        rshrn           v5.4h,  v2.4s,  #5
        addp            v2.4h,  v5.4h,  v5.4h
        shl             v3.4h,  v2.4h,  #1
        add             v3.4h,  v3.4h,  v2.4h
        rev64           v7.4h,  v7.4h
        add             v7.4h,  v7.4h,  v0.4h
        shl             v2.4h,  v7.4h,  #4
        sub             v2.4h,  v2.4h,  v3.4h
        ext             v0.16b, v0.16b, v0.16b, #14
        mov             v0.h[0],  wzr
        mul             v0.8h,  v0.8h,  v5.h[0]
        dup             v1.8h,  v2.h[0]
        dup             v2.8h,  v5.h[1]
        add             v1.8h,  v1.8h,  v0.8h
        mov             w3,  #8
1:
        sqshrun         v0.8b,  v1.8h,  #5
        subs            w3,  w3,  #1
        add             v1.8h,  v1.8h,  v2.8h
        st1             {v0.8b},  [x0], x1
        b.ne            1b
        ret
endfunc

function ff_pred8x8_128_dc_neon, export=1
        movi            v0.8b,  #128
        movi            v1.8b,  #128
        b               .L_pred8x8_dc_end
endfunc

function ff_pred8x8_top_dc_neon, export=1
        sub             x2,  x0,  x1
        ld1             {v0.8b},  [x2]
        uaddlp          v0.4h,  v0.8b
        addp            v0.4h,  v0.4h,  v0.4h
        zip1            v0.8h,  v0.8h,  v0.8h
        rshrn           v2.8b,  v0.8h,  #2
        zip1            v0.8b,  v2.8b,  v2.8b
        zip1            v1.8b,  v2.8b,  v2.8b
        b               .L_pred8x8_dc_end
endfunc

function ff_pred8x8_left_dc_neon, export=1
        sub             x2,  x0,  #1
        ldcol.8         v0,  x2,  x1
        uaddlp          v0.4h,  v0.8b
        addp            v0.4h,  v0.4h,  v0.4h
        rshrn           v2.8b,  v0.8h,  #2
        dup             v1.8b,  v2.b[1]
        dup             v0.8b,  v2.b[0]
        b               .L_pred8x8_dc_end
endfunc

function ff_pred8x8_dc_neon, export=1
        sub             x2,  x0,  x1
        sub             x3,  x0,  #1
        ld1             {v0.8b}, [x2]
        ldcol.8         v1,  x3,  x1
        uaddlp          v0.4h,  v0.8b
        uaddlp          v1.4h,  v1.8b
        trn1            v2.2s,  v0.2s,  v1.2s
        trn2            v3.2s,  v0.2s,  v1.2s
        addp            v4.4h,  v2.4h,  v3.4h
        addp            v5.4h,  v4.4h,  v4.4h
        rshrn           v6.8b,  v5.8h,  #3
        rshrn           v7.8b,  v4.8h,  #2
        dup             v0.8b,  v6.b[0]
        dup             v2.8b,  v7.b[2]
        dup             v1.8b,  v7.b[3]
        dup             v3.8b,  v6.b[1]
        zip1            v0.2s,  v0.2s,  v2.2s
        zip1            v1.2s,  v1.2s,  v3.2s
.L_pred8x8_dc_end:
        mov             w3,  #4
        add             x2,  x0,  x1,  lsl #2
6:      subs            w3,  w3,  #1
        st1             {v0.8b},  [x0], x1
        st1             {v1.8b},  [x2], x1
        b.ne            6b
        ret
endfunc

function ff_pred8x8_l0t_dc_neon, export=1
        sub             x2,  x0,  x1
        sub             x3,  x0,  #1
        ld1             {v0.8b},  [x2]
        ldcol.8         v1,  x3,  x1,  4
        zip1            v0.4s,  v0.4s,  v1.4s
        uaddlp          v0.8h,  v0.16b
        addp            v0.8h,  v0.8h,  v0.8h
        addp            v1.4h,  v0.4h,  v0.4h
        rshrn           v2.8b,  v0.8h,  #2
        rshrn           v3.8b,  v1.8h,  #3
        dup             v4.8b,  v3.b[0]
        dup             v6.8b,  v2.b[2]
        dup             v5.8b,  v2.b[0]
        zip1            v0.2s,  v4.2s,  v6.2s
        zip1            v1.2s,  v5.2s,  v6.2s
        b               .L_pred8x8_dc_end
endfunc

function ff_pred8x8_l00_dc_neon, export=1
        sub             x2,  x0,  #1
        ldcol.8         v0,  x2,  x1,  4
        uaddlp          v0.4h,  v0.8b
        addp            v0.4h,  v0.4h,  v0.4h
        rshrn           v0.8b,  v0.8h,  #2
        movi            v1.8b,  #128
        dup             v0.8b,  v0.b[0]
        b               .L_pred8x8_dc_end
endfunc

function ff_pred8x8_0lt_dc_neon, export=1
        add             x3,  x0,  x1,  lsl #2
        sub             x2,  x0,  x1
        sub             x3,  x3,  #1
        ld1             {v0.8b},  [x2]
        ldcol.8         v1,  x3,  x1,  4,  hi=1
        zip1            v0.4s,  v0.4s,  v1.4s
        uaddlp          v0.8h,  v0.16b
        addp            v0.8h,  v0.8h,  v0.8h
        addp            v1.4h,  v0.4h,  v0.4h
        rshrn           v2.8b,  v0.8h,  #2
        rshrn           v3.8b,  v1.8h,  #3
        dup             v4.8b,  v2.b[0]
        dup             v5.8b,  v2.b[3]
        dup             v6.8b,  v2.b[2]
        dup             v7.8b,  v3.b[1]
        zip1            v0.2s,  v4.2s,  v6.2s
        zip1            v1.2s,  v5.2s,  v7.2s
        b               .L_pred8x8_dc_end
endfunc

function ff_pred8x8_0l0_dc_neon, export=1
        add             x2,  x0,  x1,  lsl #2
        sub             x2,  x2,  #1
        ldcol.8         v1,  x2,  x1,  4
        uaddlp          v2.4h,  v1.8b
        addp            v2.4h,  v2.4h,  v2.4h
        rshrn           v1.8b,  v2.8h,  #2
        movi            v0.8b,  #128
        dup             v1.8b,  v1.b[0]
        b               .L_pred8x8_dc_end
endfunc

.macro ldcol.16  rd,  rs,  rt,  n=4,  hi=0
.if \n >= 4 && \hi == 0
        ld1             {\rd\().h}[0],  [\rs], \rt
        ld1             {\rd\().h}[1],  [\rs], \rt
        ld1             {\rd\().h}[2],  [\rs], \rt
        ld1             {\rd\().h}[3],  [\rs], \rt
.endif
.if \n == 8 || \hi == 1
        ld1             {\rd\().h}[4],  [\rs], \rt
        ld1             {\rd\().h}[5],  [\rs], \rt
        ld1             {\rd\().h}[6],  [\rs], \rt
        ld1             {\rd\().h}[7],  [\rs], \rt
.endif
.endm

// slower than C
/*
function ff_pred16x16_128_dc_neon_10, export=1
        movi            v0.8h, #2, lsl #8 // 512, 1 << (bit_depth - 1)

        b               .L_pred16x16_dc_10_end
endfunc
*/

function ff_pred16x16_top_dc_neon_10, export=1
        sub             x2,  x0,  x1

        ld1             {v0.8h, v1.8h}, [x2]

        add             v0.8h, v0.8h, v1.8h
        addv            h0, v0.8h

        urshr           v0.4h,  v0.4h,  #4
        dup             v0.8h, v0.h[0]
        b               .L_pred16x16_dc_10_end
endfunc

// slower than C
/*
function ff_pred16x16_left_dc_neon_10, export=1
        sub             x2,  x0,  #2 // access to the "left" column
        ldcol.16        v0,  x2,  x1,  8
        ldcol.16        v1,  x2,  x1,  8 // load "left" column

        add             v0.8h, v0.8h, v1.8h
        addv            h0,  v0.8h

        urshr           v0.4h,  v0.4h,  #4
        dup             v0.8h, v0.h[0]
        b               .L_pred16x16_dc_10_end
endfunc
*/

function ff_pred16x16_dc_neon_10, export=1
        sub             x2,  x0,  x1 // access to the "top" row
        sub             x3,  x0,  #2 // access to the "left" column

        ld1             {v0.8h, v1.8h}, [x2]
        ldcol.16        v2,  x3,  x1,  8
        ldcol.16        v3,  x3,  x1,  8 // load pixels in "top" row and "left" col

        add             v0.8h, v0.8h, v1.8h
        add             v2.8h, v2.8h, v3.8h
        add             v0.8h, v0.8h, v2.8h
        addv            h0, v0.8h

        urshr           v0.4h,  v0.4h,  #5
        dup             v0.8h,  v0.h[0]
.L_pred16x16_dc_10_end:
        mov             v1.16b,  v0.16b
        mov             w3,  #8
6:      st1             {v0.8h, v1.8h}, [x0], x1
        subs            w3,  w3,  #1
        st1             {v0.8h, v1.8h}, [x0], x1
        b.ne            6b
        ret
endfunc

function ff_pred16x16_hor_neon_10, export=1
        sub             x2,  x0,  #2
        add             x3,  x0,  #16

        mov             w4,  #16
1:      ld1r            {v0.8h},  [x2],  x1
        subs            w4,  w4,  #1
        st1             {v0.8h},  [x0],  x1
        st1             {v0.8h},  [x3],  x1
        b.ne            1b
        ret
endfunc

function ff_pred16x16_vert_neon_10, export=1
        sub             x2,  x0,  x1
        add             x1,  x1,  x1

        ld1             {v0.8h, v1.8h},  [x2],  x1

        mov             w3,  #8
1:      subs            w3,  w3,  #1
        st1             {v0.8h, v1.8h},  [x0],  x1
        st1             {v0.8h, v1.8h},  [x2],  x1

        b.ne            1b
        ret
endfunc

function ff_pred16x16_plane_neon_10, export=1
        sub             x3,  x0,  x1
        movrel          x4,  p16weight
        add             x2,  x3,  #16
        sub             x3,  x3,  #2
        ld1             {v0.8h},  [x3]
        ld1             {v2.8h},  [x2], x1
        ldcol.16        v1,  x3,  x1, 8
        add             x3,  x3,  x1
        ldcol.16        v3,  x3,  x1, 8

        rev64           v16.8h,  v0.8h
        rev64           v17.8h,  v1.8h
        ext             v0.16b, v16.16b, v16.16b, #8
        ext             v1.16b, v17.16b, v17.16b, #8

        add             v7.8h,  v2.8h,  v3.8h
        sub             v2.8h,  v2.8h,  v0.8h
        sub             v3.8h,  v3.8h,  v1.8h
        ld1             {v0.8h},     [x4]
        mul             v2.8h,  v2.8h,  v0.8h
        mul             v3.8h,  v3.8h,  v0.8h
        addp            v2.8h,  v2.8h,  v3.8h
        addp            v2.8h,  v2.8h,  v2.8h
        addp            v2.4h,  v2.4h,  v2.4h
        sshll           v3.4s,  v2.4h,  #2
        saddw           v2.4s,  v3.4s,  v2.4h
        rshrn           v4.4h,  v2.4s,  #6
        trn2            v5.4h,  v4.4h,  v4.4h
        add             v2.4h,  v4.4h,  v5.4h
        shl             v3.4h,  v2.4h,  #3
        ext             v7.16b, v7.16b, v7.16b, #14
        sub             v3.4h,  v3.4h,  v2.4h   // 7 * (b + c)
        add             v7.4h,  v7.4h,  v0.4h
        shl             v2.4h,  v7.4h,  #4
        ssubl           v2.4s,  v2.4h,  v3.4h
        ext             v0.16b, v0.16b, v0.16b, #14
        sxtl            v6.4s,  v5.4h          // c

        mov             v0.h[0],  wzr
        mul             v0.8h,  v0.8h,  v4.h[0]
        dup             v16.4s, v2.s[0]
        dup             v17.4s, v2.s[0]
        dup             v2.8h,  v4.h[0]        // b
        dup             v3.4s,  v6.s[0]        // c
        sshll           v2.4s,  v2.4h,  #3     // b * 8
        saddw           v16.4s, v16.4s, v0.4h
        saddw2          v17.4s, v17.4s, v0.8h
        sub             v3.4s,  v3.4s,  v2.4s

        mov             w3,      #16
        mvni            v4.8h,   #0xFC, lsl #8 // 1023 for clipping
1:
        sqshrun         v0.4h,  v16.4s, #5
        sqshrun2        v0.8h,  v17.4s, #5
        add             v16.4s, v16.4s, v2.4s
        add             v17.4s, v17.4s, v2.4s
        sqshrun         v1.4h,  v16.4s, #5
        sqshrun2        v1.8h,  v17.4s, #5
        add             v16.4s, v16.4s, v3.4s
        add             v17.4s, v17.4s, v3.4s

        subs            w3,  w3,  #1

        smin            v0.8h,  v0.8h,  v4.8h
        smin            v1.8h,  v1.8h,  v4.8h

        st1             {v0.8h, v1.8h}, [x0], x1
        b.ne            1b
        ret
endfunc

function ff_pred8x8_hor_neon_10, export=1
        sub             x2,  x0,  #2
        mov             w3,  #8

1:      ld1r            {v0.8h},  [x2], x1
        subs            w3,  w3,  #1
        st1             {v0.8h},  [x0], x1
        b.ne            1b
        ret
endfunc

function ff_pred8x8_vert_neon_10, export=1
        sub             x2,  x0,  x1
        lsl             x1,  x1,  #1

        ld1             {v0.8h},  [x2], x1
        mov             w3,  #4
1:      subs            w3,  w3,  #1
        st1             {v0.8h},  [x0], x1
        st1             {v0.8h},  [x2], x1
        b.ne            1b
        ret
endfunc

function ff_pred8x8_plane_neon_10, export=1
        sub             x3,  x0,  x1
        movrel          x4,  p8weight
        movrel          x5,  p16weight
        add             x2,  x3,  #8
        sub             x3,  x3,  #2
        ld1             {v0.d}[0],  [x3]
        ld1             {v2.d}[0],  [x2], x1
        ldcol.16        v0,  x3,  x1,  hi=1
        add             x3,  x3,  x1
        ldcol.16        v3,  x3,  x1,  4
        add             v7.8h,  v2.8h,  v3.8h
        rev64           v0.8h,  v0.8h
        trn1            v2.2d,  v2.2d,  v3.2d
        sub             v2.8h,  v2.8h,  v0.8h
        ld1             {v6.8h},  [x4]
        mul             v2.8h,  v2.8h,  v6.8h
        ld1             {v0.8h},  [x5]
        saddlp          v2.4s,  v2.8h
        addp            v2.4s,  v2.4s,  v2.4s
        shl             v3.4s,  v2.4s,  #4
        add             v2.4s,  v3.4s,  v2.4s
        rshrn           v5.4h,  v2.4s,  #5
        addp            v2.4h,  v5.4h,  v5.4h
        shl             v3.4h,  v2.4h,  #1
        add             v3.4h,  v3.4h,  v2.4h
        rev64           v7.4h,  v7.4h
        add             v7.4h,  v7.4h,  v0.4h
        shl             v2.4h,  v7.4h,  #4
        ssubl           v2.4s,  v2.4h,  v3.4h
        ext             v0.16b, v0.16b, v0.16b, #14
        mov             v0.h[0],  wzr
        dup             v1.4s,  v2.s[0]
        dup             v2.4s,  v2.s[0]
        dup             v3.8h,  v5.h[1]
        smlal           v1.4s,  v0.4h,  v5.h[0]
        smlal2          v2.4s,  v0.8h,  v5.h[0]
        mov             w3,  #8
        mvni            v4.8h,  #0xFC,  lsl #8 // 1023 for clipping
1:
        sqshrun         v0.4h,  v1.4s,  #5
        sqshrun2        v0.8h,  v2.4s,  #5

        saddw           v1.4s,  v1.4s,  v3.4h
        saddw           v2.4s,  v2.4s,  v3.4h

        subs            w3,  w3,  #1

        smin            v0.8h,  v0.8h,  v4.8h

        st1             {v0.8h},  [x0],  x1
        b.ne            1b
        ret
endfunc

function ff_pred8x8_128_dc_neon_10, export=1
        movi            v0.8h,  #2, lsl #8      // 512, 1 << (bit_depth - 1)
        movi            v1.8h,  #2, lsl #8
        b               .L_pred8x8_dc_10_end
endfunc

function ff_pred8x8_top_dc_neon_10, export=1
        sub             x2,  x0,  x1
        ld1             {v0.8h},  [x2]

        addp            v0.8h,  v0.8h,  v0.8h
        addp            v0.4h,  v0.4h,  v0.4h
        zip1            v0.4h,  v0.4h,  v0.4h
        urshr           v2.4h,  v0.4h,  #2
        zip1            v0.8h,  v2.8h,  v2.8h
        zip1            v1.8h,  v2.8h,  v2.8h
        b               .L_pred8x8_dc_10_end
endfunc

function ff_pred8x8_left_dc_neon_10, export=1
        sub             x2,  x0,  #2
        ldcol.16        v0,  x2,  x1,  8

        addp            v0.8h,  v0.8h,  v0.8h
        addp            v0.4h,  v0.4h,  v0.4h
        urshr           v2.4h,  v0.4h,  #2
        dup             v1.8h,  v2.h[1]
        dup             v0.8h,  v2.h[0]
        b               .L_pred8x8_dc_10_end
endfunc

function ff_pred8x8_dc_neon_10, export=1
        sub             x2,  x0,  x1
        sub             x3,  x0,  #2

        ld1             {v0.8h}, [x2]
        ldcol.16        v1,  x3,  x1, 8

        addp            v0.8h,  v0.8h, v0.8h
        addp            v1.8h,  v1.8h, v1.8h
        trn1            v2.2s,  v0.2s,  v1.2s
        trn2            v3.2s,  v0.2s,  v1.2s
        addp            v4.4h,  v2.4h,  v3.4h
        addp            v5.4h,  v4.4h,  v4.4h
        urshr           v6.4h,  v5.4h,  #3
        urshr           v7.4h,  v4.4h,  #2
        dup             v0.8h,  v6.h[0]
        dup             v2.8h,  v7.h[2]
        dup             v1.8h,  v7.h[3]
        dup             v3.8h,  v6.h[1]
        zip1            v0.2d,  v0.2d,  v2.2d
        zip1            v1.2d,  v1.2d,  v3.2d
.L_pred8x8_dc_10_end:
        mov             w3,  #4
        add             x2,  x0,  x1,  lsl #2

6:      st1             {v0.8h},  [x0], x1
        subs            w3,  w3,  #1
        st1             {v1.8h},  [x2], x1
        b.ne            6b
        ret
endfunc

function ff_pred8x8_l0t_dc_neon_10, export=1
        sub             x2,  x0,  x1
        sub             x3,  x0,  #2

        ld1             {v0.8h},  [x2]
        ldcol.16        v1,  x3,  x1, 4

        addp            v0.8h,  v0.8h,  v0.8h
        addp            v1.4h,  v1.4h,  v1.4h
        addp            v0.4h,  v0.4h,  v0.4h
        addp            v1.4h,  v1.4h,  v1.4h
        add             v1.4h,  v1.4h,  v0.4h

        urshr           v2.4h,  v0.4h,  #2
        urshr           v3.4h,  v1.4h,  #3      // the pred4x4 part

        dup             v4.4h,  v3.h[0]
        dup             v5.4h,  v2.h[0]
        dup             v6.4h,  v2.h[1]

        zip1            v0.2d,  v4.2d,  v6.2d
        zip1            v1.2d,  v5.2d,  v6.2d
        b               .L_pred8x8_dc_10_end
endfunc

function ff_pred8x8_l00_dc_neon_10, export=1
        sub             x2,  x0,  #2

        ldcol.16        v0,  x2,  x1,  4

        addp            v0.4h,  v0.4h,  v0.4h
        addp            v0.4h,  v0.4h,  v0.4h
        urshr           v0.4h,  v0.4h,  #2

        movi            v1.8h,  #2, lsl #8      // 512
        dup             v0.8h,  v0.h[0]
        b               .L_pred8x8_dc_10_end
endfunc

function ff_pred8x8_0lt_dc_neon_10, export=1
        add             x3,  x0,  x1,  lsl #2
        sub             x2,  x0,  x1
        sub             x3,  x3,  #2

        ld1             {v0.8h},  [x2]
        ldcol.16        v1,  x3,  x1,  hi=1

        addp            v0.8h,  v0.8h,  v0.8h
        addp            v1.8h,  v1.8h,  v1.8h
        addp            v0.4h,  v0.4h,  v0.4h
        addp            v1.4h,  v1.4h,  v1.4h
        zip1            v0.2s,  v0.2s,  v1.2s
        add             v1.4h,  v0.4h,  v1.4h

        urshr           v2.4h,  v0.4h,  #2
        urshr           v3.4h,  v1.4h,  #3

        dup             v4.4h,  v2.h[0]
        dup             v5.4h,  v2.h[3]
        dup             v6.4h,  v2.h[1]
        dup             v7.4h,  v3.h[1]

        zip1            v0.2d,  v4.2d,  v6.2d
        zip1            v1.2d,  v5.2d,  v7.2d
        b               .L_pred8x8_dc_10_end
endfunc

function ff_pred8x8_0l0_dc_neon_10, export=1
        add             x2,  x0,  x1,  lsl #2
        sub             x2,  x2,  #2

        ldcol.16        v1,  x2,  x1,  4

        addp            v2.8h,  v1.8h,  v1.8h
        addp            v2.4h,  v2.4h,  v2.4h
        urshr           v1.4h,  v2.4h,  #2

        movi            v0.8h,  #2,  lsl #8     // 512
        dup             v1.8h,  v1.h[0]
        b               .L_pred8x8_dc_10_end
endfunc