FFmpeg: libavcodec/ppc/idct_altivec.c Source File

00001 /*
00002  * Copyright (c) 2001 Michel Lespinasse
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 /*
00022  * NOTE: This code is based on GPL code from the libmpeg2 project.  The
00023  * author, Michel Lespinasses, has given explicit permission to release
00024  * under LGPL as part of FFmpeg.
00025  */
00026 
00027 /*
00028  * FFmpeg integration by Dieter Shirley
00029  *
00030  * This file is a direct copy of the AltiVec IDCT module from the libmpeg2
00031  * project.  I've deleted all of the libmpeg2-specific code, renamed the
00032  * functions and reordered the function parameters.  The only change to the
00033  * IDCT function itself was to factor out the partial transposition, and to
00034  * perform a full transpose at the end of the function.
00035  */
00036 
00037 
00038 #include <stdlib.h>                                      /* malloc(), free() */
00039 #include <string.h>
00040 #include "config.h"
00041 #if HAVE_ALTIVEC_H
00042 #include <altivec.h>
00043 #endif
00044 #include "libavcodec/dsputil.h"
00045 #include "types_altivec.h"
00046 #include "dsputil_altivec.h"
00047 
00048 #define IDCT_HALF                                       \
00049     /* 1st stage */                                     \
00050     t1 = vec_mradds (a1, vx7, vx1 );                    \
00051     t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7));    \
00052     t7 = vec_mradds (a2, vx5, vx3);                     \
00053     t3 = vec_mradds (ma2, vx3, vx5);                    \
00054                                                         \
00055     /* 2nd stage */                                     \
00056     t5 = vec_adds (vx0, vx4);                           \
00057     t0 = vec_subs (vx0, vx4);                           \
00058     t2 = vec_mradds (a0, vx6, vx2);                     \
00059     t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6));    \
00060     t6 = vec_adds (t8, t3);                             \
00061     t3 = vec_subs (t8, t3);                             \
00062     t8 = vec_subs (t1, t7);                             \
00063     t1 = vec_adds (t1, t7);                             \
00064                                                         \
00065     /* 3rd stage */                                     \
00066     t7 = vec_adds (t5, t2);                             \
00067     t2 = vec_subs (t5, t2);                             \
00068     t5 = vec_adds (t0, t4);                             \
00069     t0 = vec_subs (t0, t4);                             \
00070     t4 = vec_subs (t8, t3);                             \
00071     t3 = vec_adds (t8, t3);                             \
00072                                                         \
00073     /* 4th stage */                                     \
00074     vy0 = vec_adds (t7, t1);                            \
00075     vy7 = vec_subs (t7, t1);                            \
00076     vy1 = vec_mradds (c4, t3, t5);                      \
00077     vy6 = vec_mradds (mc4, t3, t5);                     \
00078     vy2 = vec_mradds (c4, t4, t0);                      \
00079     vy5 = vec_mradds (mc4, t4, t0);                     \
00080     vy3 = vec_adds (t2, t6);                            \
00081     vy4 = vec_subs (t2, t6);
00082 
00083 
00084 #define IDCT                                                            \
00085     vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7;                \
00086     vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                \
00087     vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias;                  \
00088     vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8;                    \
00089     vec_u16 shift;                                                 \
00090                                                                         \
00091     c4 = vec_splat (constants[0], 0);                                   \
00092     a0 = vec_splat (constants[0], 1);                                   \
00093     a1 = vec_splat (constants[0], 2);                                   \
00094     a2 = vec_splat (constants[0], 3);                                   \
00095     mc4 = vec_splat (constants[0], 4);                                  \
00096     ma2 = vec_splat (constants[0], 5);                                  \
00097     bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3);     \
00098                                                                         \
00099     zero = vec_splat_s16 (0);                                           \
00100     shift = vec_splat_u16 (4);                                          \
00101                                                                         \
00102     vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero);    \
00103     vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero);    \
00104     vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero);    \
00105     vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero);    \
00106     vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero);    \
00107     vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero);    \
00108     vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero);    \
00109     vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero);    \
00110                                                                         \
00111     IDCT_HALF                                                           \
00112                                                                         \
00113     vx0 = vec_mergeh (vy0, vy4);                                        \
00114     vx1 = vec_mergel (vy0, vy4);                                        \
00115     vx2 = vec_mergeh (vy1, vy5);                                        \
00116     vx3 = vec_mergel (vy1, vy5);                                        \
00117     vx4 = vec_mergeh (vy2, vy6);                                        \
00118     vx5 = vec_mergel (vy2, vy6);                                        \
00119     vx6 = vec_mergeh (vy3, vy7);                                        \
00120     vx7 = vec_mergel (vy3, vy7);                                        \
00121                                                                         \
00122     vy0 = vec_mergeh (vx0, vx4);                                        \
00123     vy1 = vec_mergel (vx0, vx4);                                        \
00124     vy2 = vec_mergeh (vx1, vx5);                                        \
00125     vy3 = vec_mergel (vx1, vx5);                                        \
00126     vy4 = vec_mergeh (vx2, vx6);                                        \
00127     vy5 = vec_mergel (vx2, vx6);                                        \
00128     vy6 = vec_mergeh (vx3, vx7);                                        \
00129     vy7 = vec_mergel (vx3, vx7);                                        \
00130                                                                         \
00131     vx0 = vec_adds (vec_mergeh (vy0, vy4), bias);                       \
00132     vx1 = vec_mergel (vy0, vy4);                                        \
00133     vx2 = vec_mergeh (vy1, vy5);                                        \
00134     vx3 = vec_mergel (vy1, vy5);                                        \
00135     vx4 = vec_mergeh (vy2, vy6);                                        \
00136     vx5 = vec_mergel (vy2, vy6);                                        \
00137     vx6 = vec_mergeh (vy3, vy7);                                        \
00138     vx7 = vec_mergel (vy3, vy7);                                        \
00139                                                                         \
00140     IDCT_HALF                                                           \
00141                                                                         \
00142     shift = vec_splat_u16 (6);                                          \
00143     vx0 = vec_sra (vy0, shift);                                         \
00144     vx1 = vec_sra (vy1, shift);                                         \
00145     vx2 = vec_sra (vy2, shift);                                         \
00146     vx3 = vec_sra (vy3, shift);                                         \
00147     vx4 = vec_sra (vy4, shift);                                         \
00148     vx5 = vec_sra (vy5, shift);                                         \
00149     vx6 = vec_sra (vy6, shift);                                         \
00150     vx7 = vec_sra (vy7, shift);
00151 
00152 
00153 static const vec_s16 constants[5] = {
00154     {23170, 13573,  6518, 21895, -23170, -21895,    32,    31},
00155     {16384, 22725, 21407, 19266,  16384,  19266, 21407, 22725},
00156     {22725, 31521, 29692, 26722,  22725,  26722, 29692, 31521},
00157     {21407, 29692, 27969, 25172,  21407,  25172, 27969, 29692},
00158     {19266, 26722, 25172, 22654,  19266,  22654, 25172, 26722}
00159 };
00160 
00161 void ff_idct_put_altivec(uint8_t* dest, int stride, int16_t *blk)
00162 {
00163     vec_s16 *block = (vec_s16*)blk;
00164     vec_u8 tmp;
00165 
00166     IDCT
00167 
00168 #define COPY(dest,src)                                          \
00169     tmp = vec_packsu (src, src);                                \
00170     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);       \
00171     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
00172 
00173     COPY (dest, vx0)    dest += stride;
00174     COPY (dest, vx1)    dest += stride;
00175     COPY (dest, vx2)    dest += stride;
00176     COPY (dest, vx3)    dest += stride;
00177     COPY (dest, vx4)    dest += stride;
00178     COPY (dest, vx5)    dest += stride;
00179     COPY (dest, vx6)    dest += stride;
00180     COPY (dest, vx7)
00181 }
00182 
00183 void ff_idct_add_altivec(uint8_t* dest, int stride, int16_t *blk)
00184 {
00185     vec_s16 *block = (vec_s16*)blk;
00186     vec_u8 tmp;
00187     vec_s16 tmp2, tmp3;
00188     vec_u8 perm0;
00189     vec_u8 perm1;
00190     vec_u8 p0, p1, p;
00191 
00192     IDCT
00193 
00194     p0 = vec_lvsl (0, dest);
00195     p1 = vec_lvsl (stride, dest);
00196     p = vec_splat_u8 (-1);
00197     perm0 = vec_mergeh (p, p0);
00198     perm1 = vec_mergeh (p, p1);
00199 
00200 #define ADD(dest,src,perm)                                              \
00201     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
00202     tmp = vec_ld (0, dest);                                             \
00203     tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm);       \
00204     tmp3 = vec_adds (tmp2, src);                                        \
00205     tmp = vec_packsu (tmp3, tmp3);                                      \
00206     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);               \
00207     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
00208 
00209     ADD (dest, vx0, perm0)      dest += stride;
00210     ADD (dest, vx1, perm1)      dest += stride;
00211     ADD (dest, vx2, perm0)      dest += stride;
00212     ADD (dest, vx3, perm1)      dest += stride;
00213     ADD (dest, vx4, perm0)      dest += stride;
00214     ADD (dest, vx5, perm1)      dest += stride;
00215     ADD (dest, vx6, perm0)      dest += stride;
00216     ADD (dest, vx7, perm1)
00217 }