7acef487a691684f593540a5e57244febfb92402
[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 functions and
32  * re-ordered the function parameters.  The only change to the IDCT function
33  * itself was to factor out the partial transposition, and to perform a full
34  * 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
44 #include "dsputil_ppc.h"
45
46 #define vector_s16_t vector signed short
47 #define const_vector_s16_t const vector signed short
48 #define vector_u16_t vector unsigned short
49 #define vector_s8_t vector signed char
50 #define vector_u8_t vector unsigned char
51 #define vector_s32_t vector signed int
52 #define vector_u32_t vector unsigned int
53
54 #define IDCT_HALF                                       \
55     /* 1st stage */                                     \
56     t1 = vec_mradds (a1, vx7, vx1 );                    \
57     t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7));    \
58     t7 = vec_mradds (a2, vx5, vx3);                     \
59     t3 = vec_mradds (ma2, vx3, vx5);                    \
60                                                         \
61     /* 2nd stage */                                     \
62     t5 = vec_adds (vx0, vx4);                           \
63     t0 = vec_subs (vx0, vx4);                           \
64     t2 = vec_mradds (a0, vx6, vx2);                     \
65     t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6));    \
66     t6 = vec_adds (t8, t3);                             \
67     t3 = vec_subs (t8, t3);                             \
68     t8 = vec_subs (t1, t7);                             \
69     t1 = vec_adds (t1, t7);                             \
70                                                         \
71     /* 3rd stage */                                     \
72     t7 = vec_adds (t5, t2);                             \
73     t2 = vec_subs (t5, t2);                             \
74     t5 = vec_adds (t0, t4);                             \
75     t0 = vec_subs (t0, t4);                             \
76     t4 = vec_subs (t8, t3);                             \
77     t3 = vec_adds (t8, t3);                             \
78                                                         \
79     /* 4th stage */                                     \
80     vy0 = vec_adds (t7, t1);                            \
81     vy7 = vec_subs (t7, t1);                            \
82     vy1 = vec_mradds (c4, t3, t5);                      \
83     vy6 = vec_mradds (mc4, t3, t5);                     \
84     vy2 = vec_mradds (c4, t4, t0);                      \
85     vy5 = vec_mradds (mc4, t4, t0);                     \
86     vy3 = vec_adds (t2, t6);                            \
87     vy4 = vec_subs (t2, t6);
88
89
90 #define IDCT                                                            \
91     vector_s16_t vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7;                \
92     vector_s16_t vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                \
93     vector_s16_t a0, a1, a2, ma2, c4, mc4, zero, bias;                  \
94     vector_s16_t t0, t1, t2, t3, t4, t5, t6, t7, t8;                    \
95     vector_u16_t shift;                                                 \
96                                                                         \
97     c4 = vec_splat (constants[0], 0);                                   \
98     a0 = vec_splat (constants[0], 1);                                   \
99     a1 = vec_splat (constants[0], 2);                                   \
100     a2 = vec_splat (constants[0], 3);                                   \
101     mc4 = vec_splat (constants[0], 4);                                  \
102     ma2 = vec_splat (constants[0], 5);                                  \
103     bias = (vector_s16_t)vec_splat ((vector_s32_t)constants[0], 3);     \
104                                                                         \
105     zero = vec_splat_s16 (0);                                           \
106     shift = vec_splat_u16 (4);                                          \
107                                                                         \
108     vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero);    \
109     vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero);    \
110     vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero);    \
111     vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero);    \
112     vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero);    \
113     vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero);    \
114     vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero);    \
115     vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero);    \
116                                                                         \
117     IDCT_HALF                                                           \
118                                                                         \
119     vx0 = vec_mergeh (vy0, vy4);                                        \
120     vx1 = vec_mergel (vy0, vy4);                                        \
121     vx2 = vec_mergeh (vy1, vy5);                                        \
122     vx3 = vec_mergel (vy1, vy5);                                        \
123     vx4 = vec_mergeh (vy2, vy6);                                        \
124     vx5 = vec_mergel (vy2, vy6);                                        \
125     vx6 = vec_mergeh (vy3, vy7);                                        \
126     vx7 = vec_mergel (vy3, vy7);                                        \
127                                                                         \
128     vy0 = vec_mergeh (vx0, vx4);                                        \
129     vy1 = vec_mergel (vx0, vx4);                                        \
130     vy2 = vec_mergeh (vx1, vx5);                                        \
131     vy3 = vec_mergel (vx1, vx5);                                        \
132     vy4 = vec_mergeh (vx2, vx6);                                        \
133     vy5 = vec_mergel (vx2, vx6);                                        \
134     vy6 = vec_mergeh (vx3, vx7);                                        \
135     vy7 = vec_mergel (vx3, vx7);                                        \
136                                                                         \
137     vx0 = vec_adds (vec_mergeh (vy0, vy4), bias);                       \
138     vx1 = vec_mergel (vy0, vy4);                                        \
139     vx2 = vec_mergeh (vy1, vy5);                                        \
140     vx3 = vec_mergel (vy1, vy5);                                        \
141     vx4 = vec_mergeh (vy2, vy6);                                        \
142     vx5 = vec_mergel (vy2, vy6);                                        \
143     vx6 = vec_mergeh (vy3, vy7);                                        \
144     vx7 = vec_mergel (vy3, vy7);                                        \
145                                                                         \
146     IDCT_HALF                                                           \
147                                                                         \
148     shift = vec_splat_u16 (6);                                          \
149     vx0 = vec_sra (vy0, shift);                                         \
150     vx1 = vec_sra (vy1, shift);                                         \
151     vx2 = vec_sra (vy2, shift);                                         \
152     vx3 = vec_sra (vy3, shift);                                         \
153     vx4 = vec_sra (vy4, shift);                                         \
154     vx5 = vec_sra (vy5, shift);                                         \
155     vx6 = vec_sra (vy6, shift);                                         \
156     vx7 = vec_sra (vy7, shift);
157
158
159 static const_vector_s16_t constants[5] = {
160     {23170, 13573,  6518, 21895, -23170, -21895,    32,    31},
161     {16384, 22725, 21407, 19266,  16384,  19266, 21407, 22725},
162     {22725, 31521, 29692, 26722,  22725,  26722, 29692, 31521},
163     {21407, 29692, 27969, 25172,  21407,  25172, 27969, 29692},
164     {19266, 26722, 25172, 22654,  19266,  22654, 25172, 26722}
165 };
166
167 void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
168 {
169 POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
170     vector_u8_t tmp;
171
172 #ifdef CONFIG_POWERPC_PERF
173 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
174 #endif
175     IDCT
176
177 #define COPY(dest,src)                                          \
178     tmp = vec_packsu (src, src);                                \
179     vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest);       \
180     vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
181
182     COPY (dest, vx0)    dest += stride;
183     COPY (dest, vx1)    dest += stride;
184     COPY (dest, vx2)    dest += stride;
185     COPY (dest, vx3)    dest += stride;
186     COPY (dest, vx4)    dest += stride;
187     COPY (dest, vx5)    dest += stride;
188     COPY (dest, vx6)    dest += stride;
189     COPY (dest, vx7)
190
191 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
192 }
193
194 void idct_add_altivec(uint8_t* dest, int stride, vector_s16_t* block)
195 {
196 POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
197     vector_u8_t tmp;
198     vector_s16_t tmp2, tmp3;
199     vector_u8_t perm0;
200     vector_u8_t perm1;
201     vector_u8_t p0, p1, p;
202
203 #ifdef CONFIG_POWERPC_PERF
204 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
205 #endif
206
207     IDCT
208
209     p0 = vec_lvsl (0, dest);
210     p1 = vec_lvsl (stride, dest);
211     p = vec_splat_u8 (-1);
212     perm0 = vec_mergeh (p, p0);
213     perm1 = vec_mergeh (p, p1);
214
215 #define ADD(dest,src,perm)                                              \
216     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
217     tmp = vec_ld (0, dest);                                             \
218     tmp2 = (vector_s16_t)vec_perm (tmp, (vector_u8_t)zero, perm);       \
219     tmp3 = vec_adds (tmp2, src);                                        \
220     tmp = vec_packsu (tmp3, tmp3);                                      \
221     vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest);               \
222     vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
223
224     ADD (dest, vx0, perm0)      dest += stride;
225     ADD (dest, vx1, perm1)      dest += stride;
226     ADD (dest, vx2, perm0)      dest += stride;
227     ADD (dest, vx3, perm1)      dest += stride;
228     ADD (dest, vx4, perm0)      dest += stride;
229     ADD (dest, vx5, perm1)      dest += stride;
230     ADD (dest, vx6, perm0)      dest += stride;
231     ADD (dest, vx7, perm1)
232
233 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
234 }
235