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_ppc.h"
00047 #include "dsputil_altivec.h"
00048 
00049 #define IDCT_HALF                                       \
00050     /* 1st stage */                                     \
00051     t1 = vec_mradds (a1, vx7, vx1 );                    \
00052     t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7));    \
00053     t7 = vec_mradds (a2, vx5, vx3);                     \
00054     t3 = vec_mradds (ma2, vx3, vx5);                    \
00055                                                         \
00056     /* 2nd stage */                                     \
00057     t5 = vec_adds (vx0, vx4);                           \
00058     t0 = vec_subs (vx0, vx4);                           \
00059     t2 = vec_mradds (a0, vx6, vx2);                     \
00060     t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6));    \
00061     t6 = vec_adds (t8, t3);                             \
00062     t3 = vec_subs (t8, t3);                             \
00063     t8 = vec_subs (t1, t7);                             \
00064     t1 = vec_adds (t1, t7);                             \
00065                                                         \
00066     /* 3rd stage */                                     \
00067     t7 = vec_adds (t5, t2);                             \
00068     t2 = vec_subs (t5, t2);                             \
00069     t5 = vec_adds (t0, t4);                             \
00070     t0 = vec_subs (t0, t4);                             \
00071     t4 = vec_subs (t8, t3);                             \
00072     t3 = vec_adds (t8, t3);                             \
00073                                                         \
00074     /* 4th stage */                                     \
00075     vy0 = vec_adds (t7, t1);                            \
00076     vy7 = vec_subs (t7, t1);                            \
00077     vy1 = vec_mradds (c4, t3, t5);                      \
00078     vy6 = vec_mradds (mc4, t3, t5);                     \
00079     vy2 = vec_mradds (c4, t4, t0);                      \
00080     vy5 = vec_mradds (mc4, t4, t0);                     \
00081     vy3 = vec_adds (t2, t6);                            \
00082     vy4 = vec_subs (t2, t6);
00083 
00084 
00085 #define IDCT                                                            \
00086     vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7;                \
00087     vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                \
00088     vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias;                  \
00089     vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8;                    \
00090     vec_u16 shift;                                                 \
00091                                                                         \
00092     c4 = vec_splat (constants[0], 0);                                   \
00093     a0 = vec_splat (constants[0], 1);                                   \
00094     a1 = vec_splat (constants[0], 2);                                   \
00095     a2 = vec_splat (constants[0], 3);                                   \
00096     mc4 = vec_splat (constants[0], 4);                                  \
00097     ma2 = vec_splat (constants[0], 5);                                  \
00098     bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3);     \
00099                                                                         \
00100     zero = vec_splat_s16 (0);                                           \
00101     shift = vec_splat_u16 (4);                                          \
00102                                                                         \
00103     vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero);    \
00104     vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero);    \
00105     vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero);    \
00106     vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero);    \
00107     vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero);    \
00108     vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero);    \
00109     vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero);    \
00110     vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero);    \
00111                                                                         \
00112     IDCT_HALF                                                           \
00113                                                                         \
00114     vx0 = vec_mergeh (vy0, vy4);                                        \
00115     vx1 = vec_mergel (vy0, vy4);                                        \
00116     vx2 = vec_mergeh (vy1, vy5);                                        \
00117     vx3 = vec_mergel (vy1, vy5);                                        \
00118     vx4 = vec_mergeh (vy2, vy6);                                        \
00119     vx5 = vec_mergel (vy2, vy6);                                        \
00120     vx6 = vec_mergeh (vy3, vy7);                                        \
00121     vx7 = vec_mergel (vy3, vy7);                                        \
00122                                                                         \
00123     vy0 = vec_mergeh (vx0, vx4);                                        \
00124     vy1 = vec_mergel (vx0, vx4);                                        \
00125     vy2 = vec_mergeh (vx1, vx5);                                        \
00126     vy3 = vec_mergel (vx1, vx5);                                        \
00127     vy4 = vec_mergeh (vx2, vx6);                                        \
00128     vy5 = vec_mergel (vx2, vx6);                                        \
00129     vy6 = vec_mergeh (vx3, vx7);                                        \
00130     vy7 = vec_mergel (vx3, vx7);                                        \
00131                                                                         \
00132     vx0 = vec_adds (vec_mergeh (vy0, vy4), bias);                       \
00133     vx1 = vec_mergel (vy0, vy4);                                        \
00134     vx2 = vec_mergeh (vy1, vy5);                                        \
00135     vx3 = vec_mergel (vy1, vy5);                                        \
00136     vx4 = vec_mergeh (vy2, vy6);                                        \
00137     vx5 = vec_mergel (vy2, vy6);                                        \
00138     vx6 = vec_mergeh (vy3, vy7);                                        \
00139     vx7 = vec_mergel (vy3, vy7);                                        \
00140                                                                         \
00141     IDCT_HALF                                                           \
00142                                                                         \
00143     shift = vec_splat_u16 (6);                                          \
00144     vx0 = vec_sra (vy0, shift);                                         \
00145     vx1 = vec_sra (vy1, shift);                                         \
00146     vx2 = vec_sra (vy2, shift);                                         \
00147     vx3 = vec_sra (vy3, shift);                                         \
00148     vx4 = vec_sra (vy4, shift);                                         \
00149     vx5 = vec_sra (vy5, shift);                                         \
00150     vx6 = vec_sra (vy6, shift);                                         \
00151     vx7 = vec_sra (vy7, shift);
00152 
00153 
00154 static const vec_s16 constants[5] = {
00155     {23170, 13573,  6518, 21895, -23170, -21895,    32,    31},
00156     {16384, 22725, 21407, 19266,  16384,  19266, 21407, 22725},
00157     {22725, 31521, 29692, 26722,  22725,  26722, 29692, 31521},
00158     {21407, 29692, 27969, 25172,  21407,  25172, 27969, 29692},
00159     {19266, 26722, 25172, 22654,  19266,  22654, 25172, 26722}
00160 };
00161 
00162 void idct_put_altivec(uint8_t* dest, int stride, int16_t *blk)
00163 {
00164 POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
00165     vec_s16 *block = (vec_s16*)blk;
00166     vec_u8 tmp;
00167 
00168 #if CONFIG_POWERPC_PERF
00169 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
00170 #endif
00171     IDCT
00172 
00173 #define COPY(dest,src)                                          \
00174     tmp = vec_packsu (src, src);                                \
00175     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);       \
00176     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
00177 
00178     COPY (dest, vx0)    dest += stride;
00179     COPY (dest, vx1)    dest += stride;
00180     COPY (dest, vx2)    dest += stride;
00181     COPY (dest, vx3)    dest += stride;
00182     COPY (dest, vx4)    dest += stride;
00183     COPY (dest, vx5)    dest += stride;
00184     COPY (dest, vx6)    dest += stride;
00185     COPY (dest, vx7)
00186 
00187 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
00188 }
00189 
00190 void idct_add_altivec(uint8_t* dest, int stride, int16_t *blk)
00191 {
00192 POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
00193     vec_s16 *block = (vec_s16*)blk;
00194     vec_u8 tmp;
00195     vec_s16 tmp2, tmp3;
00196     vec_u8 perm0;
00197     vec_u8 perm1;
00198     vec_u8 p0, p1, p;
00199 
00200 #if CONFIG_POWERPC_PERF
00201 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
00202 #endif
00203 
00204     IDCT
00205 
00206     p0 = vec_lvsl (0, dest);
00207     p1 = vec_lvsl (stride, dest);
00208     p = vec_splat_u8 (-1);
00209     perm0 = vec_mergeh (p, p0);
00210     perm1 = vec_mergeh (p, p1);
00211 
00212 #define ADD(dest,src,perm)                                              \
00213     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
00214     tmp = vec_ld (0, dest);                                             \
00215     tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm);       \
00216     tmp3 = vec_adds (tmp2, src);                                        \
00217     tmp = vec_packsu (tmp3, tmp3);                                      \
00218     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);               \
00219     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
00220 
00221     ADD (dest, vx0, perm0)      dest += stride;
00222     ADD (dest, vx1, perm1)      dest += stride;
00223     ADD (dest, vx2, perm0)      dest += stride;
00224     ADD (dest, vx3, perm1)      dest += stride;
00225     ADD (dest, vx4, perm0)      dest += stride;
00226     ADD (dest, vx5, perm1)      dest += stride;
00227     ADD (dest, vx6, perm0)      dest += stride;
00228     ADD (dest, vx7, perm1)
00229 
00230 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
00231 }
00232