mpegvideodsp.c 6.02 KB
Newer Older
1
/*
2
 * This file is part of FFmpeg.
3
 *
4
 * FFmpeg is free software; you can redistribute it and/or
5 6 7 8
 * 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.
 *
9
 * FFmpeg is distributed in the hope that it will be useful,
10 11 12 13 14
 * 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
15
 * License along with FFmpeg; if not, write to the Free Software
16 17 18 19 20 21 22 23
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "config.h"
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/x86/cpu.h"
#include "libavcodec/mpegvideodsp.h"
24
#include "libavcodec/videodsp.h"
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45

#if HAVE_INLINE_ASM

static void gmc_mmx(uint8_t *dst, uint8_t *src,
                    int stride, int h, int ox, int oy,
                    int dxx, int dxy, int dyx, int dyy,
                    int shift, int r, int width, int height)
{
    const int w    = 8;
    const int ix   = ox  >> (16 + shift);
    const int iy   = oy  >> (16 + shift);
    const int oxs  = ox  >> 4;
    const int oys  = oy  >> 4;
    const int dxxs = dxx >> 4;
    const int dxys = dxy >> 4;
    const int dyxs = dyx >> 4;
    const int dyys = dyy >> 4;
    const uint16_t r4[4]   = { r, r, r, r };
    const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys };
    const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys };
    const uint64_t shift2  = 2 * shift;
46 47 48
#define MAX_STRIDE 4096U
#define MAX_H 8U
    uint8_t edge_buf[(MAX_H + 1) * MAX_STRIDE];
49 50 51 52 53 54
    int x, y;

    const int dxw = (dxx - (1 << (16 + shift))) * (w - 1);
    const int dyh = (dyy - (1 << (16 + shift))) * (h - 1);
    const int dxh = dxy * (h - 1);
    const int dyw = dyx * (w - 1);
55 56
    int need_emu  =  (unsigned) ix >= width  - w ||
                     (unsigned) iy >= height - h;
57 58 59 60 61 62

    if ( // non-constant fullpel offset (3% of blocks)
        ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) |
         (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) ||
        // uses more than 16 bits of subpel mv (only at huge resolution)
        (dxx | dxy | dyx | dyy) & 15 ||
63
        (need_emu && (h > MAX_H || stride > MAX_STRIDE))) {
64 65 66 67 68 69 70
        // FIXME could still use mmx for some of the rows
        ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy,
                 shift, r, width, height);
        return;
    }

    src += ix + iy * stride;
71 72 73 74
    if (need_emu) {
        ff_emulated_edge_mc_8(edge_buf, src, stride, stride, w + 1, h + 1, ix, iy, width, height);
        src = edge_buf;
    }
75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160

    __asm__ volatile (
        "movd         %0, %%mm6         \n\t"
        "pxor      %%mm7, %%mm7         \n\t"
        "punpcklwd %%mm6, %%mm6         \n\t"
        "punpcklwd %%mm6, %%mm6         \n\t"
        :: "r" (1 << shift));

    for (x = 0; x < w; x += 4) {
        uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0),
                            oxs - dxys + dxxs * (x + 1),
                            oxs - dxys + dxxs * (x + 2),
                            oxs - dxys + dxxs * (x + 3) };
        uint16_t dy4[4] = { oys - dyys + dyxs * (x + 0),
                            oys - dyys + dyxs * (x + 1),
                            oys - dyys + dyxs * (x + 2),
                            oys - dyys + dyxs * (x + 3) };

        for (y = 0; y < h; y++) {
            __asm__ volatile (
                "movq      %0, %%mm4    \n\t"
                "movq      %1, %%mm5    \n\t"
                "paddw     %2, %%mm4    \n\t"
                "paddw     %3, %%mm5    \n\t"
                "movq   %%mm4, %0       \n\t"
                "movq   %%mm5, %1       \n\t"
                "psrlw    $12, %%mm4    \n\t"
                "psrlw    $12, %%mm5    \n\t"
                : "+m" (*dx4), "+m" (*dy4)
                : "m" (*dxy4), "m" (*dyy4));

            __asm__ volatile (
                "movq      %%mm6, %%mm2 \n\t"
                "movq      %%mm6, %%mm1 \n\t"
                "psubw     %%mm4, %%mm2 \n\t"
                "psubw     %%mm5, %%mm1 \n\t"
                "movq      %%mm2, %%mm0 \n\t"
                "movq      %%mm4, %%mm3 \n\t"
                "pmullw    %%mm1, %%mm0 \n\t" // (s - dx) * (s - dy)
                "pmullw    %%mm5, %%mm3 \n\t" // dx * dy
                "pmullw    %%mm5, %%mm2 \n\t" // (s - dx) * dy
                "pmullw    %%mm4, %%mm1 \n\t" // dx * (s - dy)

                "movd         %4, %%mm5 \n\t"
                "movd         %3, %%mm4 \n\t"
                "punpcklbw %%mm7, %%mm5 \n\t"
                "punpcklbw %%mm7, %%mm4 \n\t"
                "pmullw    %%mm5, %%mm3 \n\t" // src[1, 1] * dx * dy
                "pmullw    %%mm4, %%mm2 \n\t" // src[0, 1] * (s - dx) * dy

                "movd         %2, %%mm5 \n\t"
                "movd         %1, %%mm4 \n\t"
                "punpcklbw %%mm7, %%mm5 \n\t"
                "punpcklbw %%mm7, %%mm4 \n\t"
                "pmullw    %%mm5, %%mm1 \n\t" // src[1, 0] * dx * (s - dy)
                "pmullw    %%mm4, %%mm0 \n\t" // src[0, 0] * (s - dx) * (s - dy)
                "paddw        %5, %%mm1 \n\t"
                "paddw     %%mm3, %%mm2 \n\t"
                "paddw     %%mm1, %%mm0 \n\t"
                "paddw     %%mm2, %%mm0 \n\t"

                "psrlw        %6, %%mm0 \n\t"
                "packuswb  %%mm0, %%mm0 \n\t"
                "movd      %%mm0, %0    \n\t"

                : "=m" (dst[x + y * stride])
                : "m" (src[0]), "m" (src[1]),
                  "m" (src[stride]), "m" (src[stride + 1]),
                  "m" (*r4), "m" (shift2));
            src += stride;
        }
        src += 4 - h * stride;
    }
}

#endif /* HAVE_INLINE_ASM */

av_cold void ff_mpegvideodsp_init_x86(MpegVideoDSPContext *c)
{
#if HAVE_INLINE_ASM
    int cpu_flags = av_get_cpu_flags();

    if (INLINE_MMX(cpu_flags))
        c->gmc = gmc_mmx;
#endif /* HAVE_INLINE_ASM */
}