FFmpeg: libswscale/ppc/yuv2yuv_altivec.c Source File

00001 /*
00002  * AltiVec-enhanced yuv-to-yuv convertion routines.
00003  *
00004  * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
00005  * based on the equivalent C code in swscale.c
00006  *
00007  * This file is part of FFmpeg.
00008  *
00009  * FFmpeg is free software; you can redistribute it and/or
00010  * modify it under the terms of the GNU Lesser General Public
00011  * License as published by the Free Software Foundation; either
00012  * version 2.1 of the License, or (at your option) any later version.
00013  *
00014  * FFmpeg is distributed in the hope that it will be useful,
00015  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00016  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00017  * Lesser General Public License for more details.
00018  *
00019  * You should have received a copy of the GNU Lesser General Public
00020  * License along with FFmpeg; if not, write to the Free Software
00021  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00022  */
00023 
00024 #include <inttypes.h>
00025 
00026 #include "config.h"
00027 #include "libswscale/swscale.h"
00028 #include "libswscale/swscale_internal.h"
00029 #include "libavutil/cpu.h"
00030 
00031 static int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t *src[],
00032                                        int srcStride[], int srcSliceY,
00033                                        int srcSliceH, uint8_t *dstParam[],
00034                                        int dstStride_a[])
00035 {
00036     uint8_t *dst = dstParam[0] + dstStride_a[0] * srcSliceY;
00037     // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH,
00038     //            srcStride[0], srcStride[1], dstStride[0]);
00039     const uint8_t *ysrc   = src[0];
00040     const uint8_t *usrc   = src[1];
00041     const uint8_t *vsrc   = src[2];
00042     const int width       = c->srcW;
00043     const int height      = srcSliceH;
00044     const int lumStride   = srcStride[0];
00045     const int chromStride = srcStride[1];
00046     const int dstStride   = dstStride_a[0];
00047     const vector unsigned char yperm = vec_lvsl(0, ysrc);
00048     const int vertLumPerChroma       = 2;
00049     register unsigned int y;
00050 
00051     /* This code assumes:
00052      *
00053      * 1) dst is 16 bytes-aligned
00054      * 2) dstStride is a multiple of 16
00055      * 3) width is a multiple of 16
00056      * 4) lum & chrom stride are multiples of 8
00057      */
00058 
00059     for (y = 0; y < height; y++) {
00060         int i;
00061         for (i = 0; i < width - 31; i += 32) {
00062             const unsigned int j          = i >> 1;
00063             vector unsigned char v_yA     = vec_ld(i, ysrc);
00064             vector unsigned char v_yB     = vec_ld(i + 16, ysrc);
00065             vector unsigned char v_yC     = vec_ld(i + 32, ysrc);
00066             vector unsigned char v_y1     = vec_perm(v_yA, v_yB, yperm);
00067             vector unsigned char v_y2     = vec_perm(v_yB, v_yC, yperm);
00068             vector unsigned char v_uA     = vec_ld(j, usrc);
00069             vector unsigned char v_uB     = vec_ld(j + 16, usrc);
00070             vector unsigned char v_u      = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
00071             vector unsigned char v_vA     = vec_ld(j, vsrc);
00072             vector unsigned char v_vB     = vec_ld(j + 16, vsrc);
00073             vector unsigned char v_v      = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
00074             vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
00075             vector unsigned char v_uv_b   = vec_mergel(v_u, v_v);
00076             vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
00077             vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
00078             vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
00079             vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
00080             vec_st(v_yuy2_0, (i << 1), dst);
00081             vec_st(v_yuy2_1, (i << 1) + 16, dst);
00082             vec_st(v_yuy2_2, (i << 1) + 32, dst);
00083             vec_st(v_yuy2_3, (i << 1) + 48, dst);
00084         }
00085         if (i < width) {
00086             const unsigned int j          = i >> 1;
00087             vector unsigned char v_y1     = vec_ld(i, ysrc);
00088             vector unsigned char v_u      = vec_ld(j, usrc);
00089             vector unsigned char v_v      = vec_ld(j, vsrc);
00090             vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
00091             vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
00092             vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
00093             vec_st(v_yuy2_0, (i << 1), dst);
00094             vec_st(v_yuy2_1, (i << 1) + 16, dst);
00095         }
00096         if ((y & (vertLumPerChroma - 1)) == vertLumPerChroma - 1) {
00097             usrc += chromStride;
00098             vsrc += chromStride;
00099         }
00100         ysrc += lumStride;
00101         dst  += dstStride;
00102     }
00103 
00104     return srcSliceH;
00105 }
00106 
00107 static int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t *src[],
00108                                        int srcStride[], int srcSliceY,
00109                                        int srcSliceH, uint8_t *dstParam[],
00110                                        int dstStride_a[])
00111 {
00112     uint8_t *dst = dstParam[0] + dstStride_a[0] * srcSliceY;
00113     // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH,
00114     //            srcStride[0], srcStride[1], dstStride[0]);
00115     const uint8_t *ysrc              = src[0];
00116     const uint8_t *usrc              = src[1];
00117     const uint8_t *vsrc              = src[2];
00118     const int width                  = c->srcW;
00119     const int height                 = srcSliceH;
00120     const int lumStride              = srcStride[0];
00121     const int chromStride            = srcStride[1];
00122     const int dstStride              = dstStride_a[0];
00123     const int vertLumPerChroma       = 2;
00124     const vector unsigned char yperm = vec_lvsl(0, ysrc);
00125     register unsigned int y;
00126 
00127     /* This code assumes:
00128      *
00129      * 1) dst is 16 bytes-aligned
00130      * 2) dstStride is a multiple of 16
00131      * 3) width is a multiple of 16
00132      * 4) lum & chrom stride are multiples of 8
00133      */
00134 
00135     for (y = 0; y < height; y++) {
00136         int i;
00137         for (i = 0; i < width - 31; i += 32) {
00138             const unsigned int j          = i >> 1;
00139             vector unsigned char v_yA     = vec_ld(i, ysrc);
00140             vector unsigned char v_yB     = vec_ld(i + 16, ysrc);
00141             vector unsigned char v_yC     = vec_ld(i + 32, ysrc);
00142             vector unsigned char v_y1     = vec_perm(v_yA, v_yB, yperm);
00143             vector unsigned char v_y2     = vec_perm(v_yB, v_yC, yperm);
00144             vector unsigned char v_uA     = vec_ld(j, usrc);
00145             vector unsigned char v_uB     = vec_ld(j + 16, usrc);
00146             vector unsigned char v_u      = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
00147             vector unsigned char v_vA     = vec_ld(j, vsrc);
00148             vector unsigned char v_vB     = vec_ld(j + 16, vsrc);
00149             vector unsigned char v_v      = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
00150             vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
00151             vector unsigned char v_uv_b   = vec_mergel(v_u, v_v);
00152             vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
00153             vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
00154             vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
00155             vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
00156             vec_st(v_uyvy_0, (i << 1), dst);
00157             vec_st(v_uyvy_1, (i << 1) + 16, dst);
00158             vec_st(v_uyvy_2, (i << 1) + 32, dst);
00159             vec_st(v_uyvy_3, (i << 1) + 48, dst);
00160         }
00161         if (i < width) {
00162             const unsigned int j          = i >> 1;
00163             vector unsigned char v_y1     = vec_ld(i, ysrc);
00164             vector unsigned char v_u      = vec_ld(j, usrc);
00165             vector unsigned char v_v      = vec_ld(j, vsrc);
00166             vector unsigned char v_uv_a   = vec_mergeh(v_u, v_v);
00167             vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
00168             vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
00169             vec_st(v_uyvy_0, (i << 1), dst);
00170             vec_st(v_uyvy_1, (i << 1) + 16, dst);
00171         }
00172         if ((y & (vertLumPerChroma - 1)) == vertLumPerChroma - 1) {
00173             usrc += chromStride;
00174             vsrc += chromStride;
00175         }
00176         ysrc += lumStride;
00177         dst  += dstStride;
00178     }
00179     return srcSliceH;
00180 }
00181 
00182 void ff_swscale_get_unscaled_altivec(SwsContext *c)
00183 {
00184     if ((av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) && !(c->srcW & 15) &&
00185         !(c->flags & SWS_BITEXACT) && c->srcFormat == PIX_FMT_YUV420P) {
00186         enum PixelFormat dstFormat = c->dstFormat;
00187 
00188         // unscaled YV12 -> packed YUV, we want speed
00189         if (dstFormat == PIX_FMT_YUYV422)
00190             c->swScale = yv12toyuy2_unscaled_altivec;
00191         else if (dstFormat == PIX_FMT_UYVY422)
00192             c->swScale = yv12touyvy_unscaled_altivec;
00193     }
00194 }