19cc93a98046cbbffab1ef86ccd077d9df231b18
[ffmpeg.git] / libavcodec / ppc / idct_altivec.c
1 /*
2  * Copyright (c) 2001 Michel Lespinasse
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * FFmpeg is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20
21 /*
22  * NOTE: This code is based on GPL code from the libmpeg2 project.  The
23  * author, Michel Lespinasses, has given explicit permission to release
24  * under LGPL as part of FFmpeg.
25  */
26
27 /*
28  * FFmpeg integration by Dieter Shirley
29  *
30  * This file is a direct copy of the AltiVec IDCT module from the libmpeg2
31  * project.  I've deleted all of the libmpeg2-specific code, renamed the
32  * functions and reordered the function parameters.  The only change to the
33  * IDCT function itself was to factor out the partial transposition, and to
34  * perform a full transpose at the end of the function.
35  */
36
37
38 #include <stdlib.h>                                      /* malloc(), free() */
39 #include <string.h>
40 #include "libavcodec/dsputil.h"
41
42 #include "gcc_fixes.h"
43 #include "types_altivec.h"
44 #include "dsputil_ppc.h"
45
46 #define IDCT_HALF                                       \
47     /* 1st stage */                                     \
48     t1 = vec_mradds (a1, vx7, vx1 );                    \
49     t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7));    \
50     t7 = vec_mradds (a2, vx5, vx3);                     \
51     t3 = vec_mradds (ma2, vx3, vx5);                    \
52                                                         \
53     /* 2nd stage */                                     \
54     t5 = vec_adds (vx0, vx4);                           \
55     t0 = vec_subs (vx0, vx4);                           \
56     t2 = vec_mradds (a0, vx6, vx2);                     \
57     t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6));    \
58     t6 = vec_adds (t8, t3);                             \
59     t3 = vec_subs (t8, t3);                             \
60     t8 = vec_subs (t1, t7);                             \
61     t1 = vec_adds (t1, t7);                             \
62                                                         \
63     /* 3rd stage */                                     \
64     t7 = vec_adds (t5, t2);                             \
65     t2 = vec_subs (t5, t2);                             \
66     t5 = vec_adds (t0, t4);                             \
67     t0 = vec_subs (t0, t4);                             \
68     t4 = vec_subs (t8, t3);                             \
69     t3 = vec_adds (t8, t3);                             \
70                                                         \
71     /* 4th stage */                                     \
72     vy0 = vec_adds (t7, t1);                            \
73     vy7 = vec_subs (t7, t1);                            \
74     vy1 = vec_mradds (c4, t3, t5);                      \
75     vy6 = vec_mradds (mc4, t3, t5);                     \
76     vy2 = vec_mradds (c4, t4, t0);                      \
77     vy5 = vec_mradds (mc4, t4, t0);                     \
78     vy3 = vec_adds (t2, t6);                            \
79     vy4 = vec_subs (t2, t6);
80
81
82 #define IDCT                                                            \
83     vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7;                \
84     vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                \
85     vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias;                  \
86     vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8;                    \
87     vec_u16 shift;                                                 \
88                                                                         \
89     c4 = vec_splat (constants[0], 0);                                   \
90     a0 = vec_splat (constants[0], 1);                                   \
91     a1 = vec_splat (constants[0], 2);                                   \
92     a2 = vec_splat (constants[0], 3);                                   \
93     mc4 = vec_splat (constants[0], 4);                                  \
94     ma2 = vec_splat (constants[0], 5);                                  \
95     bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3);     \
96                                                                         \
97     zero = vec_splat_s16 (0);                                           \
98     shift = vec_splat_u16 (4);                                          \
99                                                                         \
100     vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero);    \
101     vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero);    \
102     vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero);    \
103     vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero);    \
104     vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero);    \
105     vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero);    \
106     vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero);    \
107     vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero);    \
108                                                                         \
109     IDCT_HALF                                                           \
110                                                                         \
111     vx0 = vec_mergeh (vy0, vy4);                                        \
112     vx1 = vec_mergel (vy0, vy4);                                        \
113     vx2 = vec_mergeh (vy1, vy5);                                        \
114     vx3 = vec_mergel (vy1, vy5);                                        \
115     vx4 = vec_mergeh (vy2, vy6);                                        \
116     vx5 = vec_mergel (vy2, vy6);                                        \
117     vx6 = vec_mergeh (vy3, vy7);                                        \
118     vx7 = vec_mergel (vy3, vy7);                                        \
119                                                                         \
120     vy0 = vec_mergeh (vx0, vx4);                                        \
121     vy1 = vec_mergel (vx0, vx4);                                        \
122     vy2 = vec_mergeh (vx1, vx5);                                        \
123     vy3 = vec_mergel (vx1, vx5);                                        \
124     vy4 = vec_mergeh (vx2, vx6);                                        \
125     vy5 = vec_mergel (vx2, vx6);                                        \
126     vy6 = vec_mergeh (vx3, vx7);                                        \
127     vy7 = vec_mergel (vx3, vx7);                                        \
128                                                                         \
129     vx0 = vec_adds (vec_mergeh (vy0, vy4), bias);                       \
130     vx1 = vec_mergel (vy0, vy4);                                        \
131     vx2 = vec_mergeh (vy1, vy5);                                        \
132     vx3 = vec_mergel (vy1, vy5);                                        \
133     vx4 = vec_mergeh (vy2, vy6);                                        \
134     vx5 = vec_mergel (vy2, vy6);                                        \
135     vx6 = vec_mergeh (vy3, vy7);                                        \
136     vx7 = vec_mergel (vy3, vy7);                                        \
137                                                                         \
138     IDCT_HALF                                                           \
139                                                                         \
140     shift = vec_splat_u16 (6);                                          \
141     vx0 = vec_sra (vy0, shift);                                         \
142     vx1 = vec_sra (vy1, shift);                                         \
143     vx2 = vec_sra (vy2, shift);                                         \
144     vx3 = vec_sra (vy3, shift);                                         \
145     vx4 = vec_sra (vy4, shift);                                         \
146     vx5 = vec_sra (vy5, shift);                                         \
147     vx6 = vec_sra (vy6, shift);                                         \
148     vx7 = vec_sra (vy7, shift);
149
150
151 static const vec_s16 constants[5] = {
152     {23170, 13573,  6518, 21895, -23170, -21895,    32,    31},
153     {16384, 22725, 21407, 19266,  16384,  19266, 21407, 22725},
154     {22725, 31521, 29692, 26722,  22725,  26722, 29692, 31521},
155     {21407, 29692, 27969, 25172,  21407,  25172, 27969, 29692},
156     {19266, 26722, 25172, 22654,  19266,  22654, 25172, 26722}
157 };
158
159 void idct_put_altivec(uint8_t* dest, int stride, vec_s16* block)
160 {
161 POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
162     vec_u8 tmp;
163
164 #if CONFIG_POWERPC_PERF
165 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
166 #endif
167     IDCT
168
169 #define COPY(dest,src)                                          \
170     tmp = vec_packsu (src, src);                                \
171     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);       \
172     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
173
174     COPY (dest, vx0)    dest += stride;
175     COPY (dest, vx1)    dest += stride;
176     COPY (dest, vx2)    dest += stride;
177     COPY (dest, vx3)    dest += stride;
178     COPY (dest, vx4)    dest += stride;
179     COPY (dest, vx5)    dest += stride;
180     COPY (dest, vx6)    dest += stride;
181     COPY (dest, vx7)
182
183 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
184 }
185
186 void idct_add_altivec(uint8_t* dest, int stride, vec_s16* block)
187 {
188 POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
189     vec_u8 tmp;
190     vec_s16 tmp2, tmp3;
191     vec_u8 perm0;
192     vec_u8 perm1;
193     vec_u8 p0, p1, p;
194
195 #if CONFIG_POWERPC_PERF
196 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
197 #endif
198
199     IDCT
200
201     p0 = vec_lvsl (0, dest);
202     p1 = vec_lvsl (stride, dest);
203     p = vec_splat_u8 (-1);
204     perm0 = vec_mergeh (p, p0);
205     perm1 = vec_mergeh (p, p1);
206
207 #define ADD(dest,src,perm)                                              \
208     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
209     tmp = vec_ld (0, dest);                                             \
210     tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm);       \
211     tmp3 = vec_adds (tmp2, src);                                        \
212     tmp = vec_packsu (tmp3, tmp3);                                      \
213     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);               \
214     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
215
216     ADD (dest, vx0, perm0)      dest += stride;
217     ADD (dest, vx1, perm1)      dest += stride;
218     ADD (dest, vx2, perm0)      dest += stride;
219     ADD (dest, vx3, perm1)      dest += stride;
220     ADD (dest, vx4, perm0)      dest += stride;
221     ADD (dest, vx5, perm1)      dest += stride;
222     ADD (dest, vx6, perm0)      dest += stride;
223     ADD (dest, vx7, perm1)
224
225 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
226 }
227