d821ecd223212b18934e6d733a7aa28140266518
[ffmpeg.git] / libavcodec / ppc / idct_altivec.c
1 /*
2  * Copyright (c) 2001 Michel Lespinasse
3  *
4  * This library is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2 of the License, or (at your option) any later version.
8  *
9  * This library is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with this library; if not, write to the Free Software
16  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
17  *
18  */
19
20 /*
21  * NOTE: This code is based on GPL code from the libmpeg2 project.  The
22  * author, Michel Lespinasses, has given explicit permission to release
23  * under LGPL as part of ffmpeg.
24  *
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 "../dsputil.h"
41
42 #include "gcc_fixes.h"
43
44 #include "dsputil_altivec.h"
45
46 #define vector_s16_t vector signed short
47 #define vector_u16_t vector unsigned short
48 #define vector_s8_t vector signed char
49 #define vector_u8_t vector unsigned char
50 #define vector_s32_t vector signed int
51 #define vector_u32_t vector unsigned int
52
53 #define IDCT_HALF                                       \
54     /* 1st stage */                                     \
55     t1 = vec_mradds (a1, vx7, vx1 );                    \
56     t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7));    \
57     t7 = vec_mradds (a2, vx5, vx3);                     \
58     t3 = vec_mradds (ma2, vx3, vx5);                    \
59                                                         \
60     /* 2nd stage */                                     \
61     t5 = vec_adds (vx0, vx4);                           \
62     t0 = vec_subs (vx0, vx4);                           \
63     t2 = vec_mradds (a0, vx6, vx2);                     \
64     t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6));    \
65     t6 = vec_adds (t8, t3);                             \
66     t3 = vec_subs (t8, t3);                             \
67     t8 = vec_subs (t1, t7);                             \
68     t1 = vec_adds (t1, t7);                             \
69                                                         \
70     /* 3rd stage */                                     \
71     t7 = vec_adds (t5, t2);                             \
72     t2 = vec_subs (t5, t2);                             \
73     t5 = vec_adds (t0, t4);                             \
74     t0 = vec_subs (t0, t4);                             \
75     t4 = vec_subs (t8, t3);                             \
76     t3 = vec_adds (t8, t3);                             \
77                                                         \
78     /* 4th stage */                                     \
79     vy0 = vec_adds (t7, t1);                            \
80     vy7 = vec_subs (t7, t1);                            \
81     vy1 = vec_mradds (c4, t3, t5);                      \
82     vy6 = vec_mradds (mc4, t3, t5);                     \
83     vy2 = vec_mradds (c4, t4, t0);                      \
84     vy5 = vec_mradds (mc4, t4, t0);                     \
85     vy3 = vec_adds (t2, t6);                            \
86     vy4 = vec_subs (t2, t6);
87
88         
89 #define IDCT                                                            \
90     vector_s16_t vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7;                \
91     vector_s16_t vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                \
92     vector_s16_t a0, a1, a2, ma2, c4, mc4, zero, bias;                  \
93     vector_s16_t t0, t1, t2, t3, t4, t5, t6, t7, t8;                    \
94     vector_u16_t shift;                                                 \
95                                                                         \
96     c4 = vec_splat (constants[0], 0);                                   \
97     a0 = vec_splat (constants[0], 1);                                   \
98     a1 = vec_splat (constants[0], 2);                                   \
99     a2 = vec_splat (constants[0], 3);                                   \
100     mc4 = vec_splat (constants[0], 4);                                  \
101     ma2 = vec_splat (constants[0], 5);                                  \
102     bias = (vector_s16_t)vec_splat ((vector_s32_t)constants[0], 3);     \
103                                                                         \
104     zero = vec_splat_s16 (0);                                           \
105     shift = vec_splat_u16 (4);                                          \
106                                                                         \
107     vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero);    \
108     vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero);    \
109     vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero);    \
110     vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero);    \
111     vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero);    \
112     vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero);    \
113     vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero);    \
114     vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero);    \
115                                                                         \
116     IDCT_HALF                                                           \
117                                                                         \
118     vx0 = vec_mergeh (vy0, vy4);                                        \
119     vx1 = vec_mergel (vy0, vy4);                                        \
120     vx2 = vec_mergeh (vy1, vy5);                                        \
121     vx3 = vec_mergel (vy1, vy5);                                        \
122     vx4 = vec_mergeh (vy2, vy6);                                        \
123     vx5 = vec_mergel (vy2, vy6);                                        \
124     vx6 = vec_mergeh (vy3, vy7);                                        \
125     vx7 = vec_mergel (vy3, vy7);                                        \
126                                                                         \
127     vy0 = vec_mergeh (vx0, vx4);                                        \
128     vy1 = vec_mergel (vx0, vx4);                                        \
129     vy2 = vec_mergeh (vx1, vx5);                                        \
130     vy3 = vec_mergel (vx1, vx5);                                        \
131     vy4 = vec_mergeh (vx2, vx6);                                        \
132     vy5 = vec_mergel (vx2, vx6);                                        \
133     vy6 = vec_mergeh (vx3, vx7);                                        \
134     vy7 = vec_mergel (vx3, vx7);                                        \
135                                                                         \
136     vx0 = vec_adds (vec_mergeh (vy0, vy4), bias);                       \
137     vx1 = vec_mergel (vy0, vy4);                                        \
138     vx2 = vec_mergeh (vy1, vy5);                                        \
139     vx3 = vec_mergel (vy1, vy5);                                        \
140     vx4 = vec_mergeh (vy2, vy6);                                        \
141     vx5 = vec_mergel (vy2, vy6);                                        \
142     vx6 = vec_mergeh (vy3, vy7);                                        \
143     vx7 = vec_mergel (vy3, vy7);                                        \
144                                                                         \
145     IDCT_HALF                                                           \
146                                                                         \
147     shift = vec_splat_u16 (6);                                          \
148     vx0 = vec_sra (vy0, shift);                                         \
149     vx1 = vec_sra (vy1, shift);                                         \
150     vx2 = vec_sra (vy2, shift);                                         \
151     vx3 = vec_sra (vy3, shift);                                         \
152     vx4 = vec_sra (vy4, shift);                                         \
153     vx5 = vec_sra (vy5, shift);                                         \
154     vx6 = vec_sra (vy6, shift);                                         \
155     vx7 = vec_sra (vy7, shift);
156
157
158 static const vector_s16_t constants[5] = {
159     (vector_s16_t) AVV(23170, 13573, 6518, 21895, -23170, -21895, 32, 31),
160     (vector_s16_t) AVV(16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725),
161     (vector_s16_t) AVV(22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521),
162     (vector_s16_t) AVV(21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692),
163     (vector_s16_t) AVV(19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722)
164 };
165
166 void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
167 {
168 POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
169 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
170 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
171     void simple_idct_put(uint8_t *dest, int line_size, int16_t *block);
172     simple_idct_put(dest, stride, (int16_t*)block);
173 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
174 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
175     vector_u8_t tmp;
176
177 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
178
179     IDCT
180
181 #define COPY(dest,src)                                          \
182     tmp = vec_packsu (src, src);                                \
183     vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest);       \
184     vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
185
186     COPY (dest, vx0)    dest += stride;
187     COPY (dest, vx1)    dest += stride;
188     COPY (dest, vx2)    dest += stride;
189     COPY (dest, vx3)    dest += stride;
190     COPY (dest, vx4)    dest += stride;
191     COPY (dest, vx5)    dest += stride;
192     COPY (dest, vx6)    dest += stride;
193     COPY (dest, vx7)
194
195 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
196 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
197 }
198
199 void idct_add_altivec(uint8_t* dest, int stride, vector_s16_t* block)
200 {
201 POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
202 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
203 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
204     void simple_idct_add(uint8_t *dest, int line_size, int16_t *block);
205     simple_idct_add(dest, stride, (int16_t*)block);
206 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
207 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
208     vector_u8_t tmp;
209     vector_s16_t tmp2, tmp3;
210     vector_u8_t perm0;
211     vector_u8_t perm1;
212     vector_u8_t p0, p1, p;
213
214 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
215
216     IDCT
217
218     p0 = vec_lvsl (0, dest);
219     p1 = vec_lvsl (stride, dest);
220     p = vec_splat_u8 (-1);
221     perm0 = vec_mergeh (p, p0);
222     perm1 = vec_mergeh (p, p1);
223
224 #define ADD(dest,src,perm)                                              \
225     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
226     tmp = vec_ld (0, dest);                                             \
227     tmp2 = (vector_s16_t)vec_perm (tmp, (vector_u8_t)zero, perm);       \
228     tmp3 = vec_adds (tmp2, src);                                        \
229     tmp = vec_packsu (tmp3, tmp3);                                      \
230     vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest);               \
231     vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
232
233     ADD (dest, vx0, perm0)      dest += stride;
234     ADD (dest, vx1, perm1)      dest += stride;
235     ADD (dest, vx2, perm0)      dest += stride;
236     ADD (dest, vx3, perm1)      dest += stride;
237     ADD (dest, vx4, perm0)      dest += stride;
238     ADD (dest, vx5, perm1)      dest += stride;
239     ADD (dest, vx6, perm0)      dest += stride;
240     ADD (dest, vx7, perm1)
241
242 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
243 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
244 }
245