yuv2yuv_altivec.c 8.51 KB
Newer Older
1
/*
Diego Biurrun's avatar
Diego Biurrun committed
2
 * AltiVec-enhanced yuv-to-yuv conversion routines.
3 4 5 6
 *
 * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
 * based on the equivalent C code in swscale.c
 *
7
 * This file is part of FFmpeg.
8
 *
9
 * FFmpeg is free software; you can redistribute it and/or
10 11 12 13
 * 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.
 *
14
 * FFmpeg is distributed in the hope that it will be useful,
15 16 17 18 19
 * 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
20
 * License along with FFmpeg; if not, write to the Free Software
21 22 23 24
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <inttypes.h>
25

26
#include "config.h"
27 28
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
29 30 31
#include "libswscale/swscale.h"
#include "libswscale/swscale_internal.h"

32 33
#if HAVE_ALTIVEC

34
static int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t *src[],
35
                                       int srcStride[], int srcSliceY,
36
                                       int srcSliceH, uint8_t *dstParam[],
37 38
                                       int dstStride_a[])
{
39 40 41 42 43 44 45 46 47
    uint8_t *dst = dstParam[0] + dstStride_a[0] * srcSliceY;
    // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH,
    //            srcStride[0], srcStride[1], dstStride[0]);
    const uint8_t *ysrc   = src[0];
    const uint8_t *usrc   = src[1];
    const uint8_t *vsrc   = src[2];
    const int width       = c->srcW;
    const int height      = srcSliceH;
    const int lumStride   = srcStride[0];
48
    const int chromStride = srcStride[1];
49
    const int dstStride   = dstStride_a[0];
50
    const vector unsigned char yperm = vec_lvsl(0, ysrc);
51
    const int vertLumPerChroma       = 2;
52 53 54
    register unsigned int y;

    /* This code assumes:
55 56 57 58 59 60
     *
     * 1) dst is 16 bytes-aligned
     * 2) dstStride is a multiple of 16
     * 3) width is a multiple of 16
     * 4) lum & chrom stride are multiples of 8
     */
61

62
    for (y = 0; y < height; y++) {
63
        int i;
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
        for (i = 0; i < width - 31; i += 32) {
            const unsigned int j          = i >> 1;
            vector unsigned char v_yA     = vec_ld(i, ysrc);
            vector unsigned char v_yB     = vec_ld(i + 16, ysrc);
            vector unsigned char v_yC     = vec_ld(i + 32, ysrc);
            vector unsigned char v_y1     = vec_perm(v_yA, v_yB, yperm);
            vector unsigned char v_y2     = vec_perm(v_yB, v_yC, yperm);
            vector unsigned char v_uA     = vec_ld(j, usrc);
            vector unsigned char v_uB     = vec_ld(j + 16, usrc);
            vector unsigned char v_u      = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
            vector unsigned char v_vA     = vec_ld(j, vsrc);
            vector unsigned char v_vB     = vec_ld(j + 16, vsrc);
            vector unsigned char v_v      = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
            vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
            vector unsigned char v_uv_b   = vec_mergel(v_u, v_v);
79 80 81 82 83 84 85 86 87 88
            vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
            vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
            vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
            vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
            vec_st(v_yuy2_0, (i << 1), dst);
            vec_st(v_yuy2_1, (i << 1) + 16, dst);
            vec_st(v_yuy2_2, (i << 1) + 32, dst);
            vec_st(v_yuy2_3, (i << 1) + 48, dst);
        }
        if (i < width) {
89 90 91 92 93
            const unsigned int j          = i >> 1;
            vector unsigned char v_y1     = vec_ld(i, ysrc);
            vector unsigned char v_u      = vec_ld(j, usrc);
            vector unsigned char v_v      = vec_ld(j, vsrc);
            vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
94 95 96 97 98
            vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
            vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
            vec_st(v_yuy2_0, (i << 1), dst);
            vec_st(v_yuy2_1, (i << 1) + 16, dst);
        }
99
        if ((y & (vertLumPerChroma - 1)) == vertLumPerChroma - 1) {
100 101 102 103
            usrc += chromStride;
            vsrc += chromStride;
        }
        ysrc += lumStride;
104
        dst  += dstStride;
105 106 107 108 109
    }

    return srcSliceH;
}

110
static int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t *src[],
111
                                       int srcStride[], int srcSliceY,
112
                                       int srcSliceH, uint8_t *dstParam[],
113 114
                                       int dstStride_a[])
{
115 116 117 118 119 120 121 122 123 124 125 126
    uint8_t *dst = dstParam[0] + dstStride_a[0] * srcSliceY;
    // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH,
    //            srcStride[0], srcStride[1], dstStride[0]);
    const uint8_t *ysrc              = src[0];
    const uint8_t *usrc              = src[1];
    const uint8_t *vsrc              = src[2];
    const int width                  = c->srcW;
    const int height                 = srcSliceH;
    const int lumStride              = srcStride[0];
    const int chromStride            = srcStride[1];
    const int dstStride              = dstStride_a[0];
    const int vertLumPerChroma       = 2;
127 128 129 130
    const vector unsigned char yperm = vec_lvsl(0, ysrc);
    register unsigned int y;

    /* This code assumes:
131 132 133 134 135 136
     *
     * 1) dst is 16 bytes-aligned
     * 2) dstStride is a multiple of 16
     * 3) width is a multiple of 16
     * 4) lum & chrom stride are multiples of 8
     */
137

138
    for (y = 0; y < height; y++) {
139
        int i;
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
        for (i = 0; i < width - 31; i += 32) {
            const unsigned int j          = i >> 1;
            vector unsigned char v_yA     = vec_ld(i, ysrc);
            vector unsigned char v_yB     = vec_ld(i + 16, ysrc);
            vector unsigned char v_yC     = vec_ld(i + 32, ysrc);
            vector unsigned char v_y1     = vec_perm(v_yA, v_yB, yperm);
            vector unsigned char v_y2     = vec_perm(v_yB, v_yC, yperm);
            vector unsigned char v_uA     = vec_ld(j, usrc);
            vector unsigned char v_uB     = vec_ld(j + 16, usrc);
            vector unsigned char v_u      = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
            vector unsigned char v_vA     = vec_ld(j, vsrc);
            vector unsigned char v_vB     = vec_ld(j + 16, vsrc);
            vector unsigned char v_v      = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
            vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
            vector unsigned char v_uv_b   = vec_mergel(v_u, v_v);
155 156 157 158 159 160 161 162 163 164
            vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
            vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
            vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
            vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
            vec_st(v_uyvy_0, (i << 1), dst);
            vec_st(v_uyvy_1, (i << 1) + 16, dst);
            vec_st(v_uyvy_2, (i << 1) + 32, dst);
            vec_st(v_uyvy_3, (i << 1) + 48, dst);
        }
        if (i < width) {
165 166 167 168 169
            const unsigned int j          = i >> 1;
            vector unsigned char v_y1     = vec_ld(i, ysrc);
            vector unsigned char v_u      = vec_ld(j, usrc);
            vector unsigned char v_v      = vec_ld(j, vsrc);
            vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
170 171 172 173 174
            vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
            vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
            vec_st(v_uyvy_0, (i << 1), dst);
            vec_st(v_uyvy_1, (i << 1) + 16, dst);
        }
175
        if ((y & (vertLumPerChroma - 1)) == vertLumPerChroma - 1) {
176 177 178 179
            usrc += chromStride;
            vsrc += chromStride;
        }
        ysrc += lumStride;
180
        dst  += dstStride;
181 182 183 184
    }
    return srcSliceH;
}

185 186
#endif /* HAVE_ALTIVEC */

187
av_cold void ff_get_unscaled_swscale_ppc(SwsContext *c)
188
{
189
#if HAVE_ALTIVEC
190
    if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
191
        return;
192 193 194

    if (!(c->srcW & 15) && !(c->flags & SWS_BITEXACT) &&
        c->srcFormat == AV_PIX_FMT_YUV420P) {
195
        enum AVPixelFormat dstFormat = c->dstFormat;
196 197

        // unscaled YV12 -> packed YUV, we want speed
198
        if (dstFormat == AV_PIX_FMT_YUYV422)
199
            c->swscale = yv12toyuy2_unscaled_altivec;
200
        else if (dstFormat == AV_PIX_FMT_UYVY422)
201
            c->swscale = yv12touyvy_unscaled_altivec;
202
    }
203
#endif /* HAVE_ALTIVEC */
204
}