9857
|
1 /*
|
|
2 * idct_altivec.c
|
10270
|
3 * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
|
9857
|
4 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
|
|
5 *
|
|
6 * This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
|
|
7 * See http://libmpeg2.sourceforge.net/ for updates.
|
|
8 *
|
|
9 * mpeg2dec is free software; you can redistribute it and/or modify
|
|
10 * it under the terms of the GNU General Public License as published by
|
|
11 * the Free Software Foundation; either version 2 of the License, or
|
|
12 * (at your option) any later version.
|
|
13 *
|
|
14 * mpeg2dec is distributed in the hope that it will be useful,
|
|
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
17 * GNU General Public License for more details.
|
|
18 *
|
|
19 * You should have received a copy of the GNU General Public License
|
|
20 * along with this program; if not, write to the Free Software
|
|
21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
22 */
|
|
23
|
|
24 #include "config.h"
|
|
25
|
|
26 #ifdef ARCH_PPC
|
|
27
|
10270
|
28 #ifdef HAVE_ALTIVEC_H
|
|
29 #include <altivec.h>
|
|
30 #endif
|
9857
|
31 #include <inttypes.h>
|
|
32
|
|
33 #include "mpeg2.h"
|
12932
|
34 #include "attributes.h"
|
9857
|
35 #include "mpeg2_internal.h"
|
|
36
|
10270
|
37 typedef vector signed char vector_s8_t;
|
|
38 typedef vector unsigned char vector_u8_t;
|
|
39 typedef vector signed short vector_s16_t;
|
|
40 typedef vector unsigned short vector_u16_t;
|
|
41 typedef vector signed int vector_s32_t;
|
|
42 typedef vector unsigned int vector_u32_t;
|
9857
|
43
|
15483
|
44 #if defined( HAVE_ALTIVEC_H ) && !defined( __APPLE_ALTIVEC__ ) && (__GNUC__ * 100 + __GNUC_MINOR__ < 303)
|
10270
|
45 /* work around gcc <3.3 vec_mergel bug */
|
|
46 static inline vector_s16_t my_vec_mergel (vector_s16_t const A,
|
|
47 vector_s16_t const B)
|
|
48 {
|
|
49 static const vector_u8_t mergel = {
|
|
50 0x08, 0x09, 0x18, 0x19, 0x0a, 0x0b, 0x1a, 0x1b,
|
|
51 0x0c, 0x0d, 0x1c, 0x1d, 0x0e, 0x0f, 0x1e, 0x1f
|
|
52 };
|
|
53 return vec_perm (A, B, mergel);
|
|
54 }
|
|
55 #undef vec_mergel
|
|
56 #define vec_mergel my_vec_mergel
|
|
57 #endif
|
9857
|
58
|
15483
|
59 #if defined( __APPLE_CC__ ) && defined( __APPLE_ALTIVEC__ ) /* apple */
|
|
60 #define VEC_S16(a,b,c,d,e,f,g,h) (vector_s16_t) (a, b, c, d, e, f, g, h)
|
|
61 #else /* gnu */
|
10270
|
62 #define VEC_S16(a,b,c,d,e,f,g,h) {a, b, c, d, e, f, g, h}
|
|
63 #endif
|
9857
|
64
|
10270
|
65 static const vector_s16_t constants ATTR_ALIGN(16) =
|
|
66 VEC_S16 (23170, 13573, 6518, 21895, -23170, -21895, 32, 31);
|
|
67 static const vector_s16_t constants_1 ATTR_ALIGN(16) =
|
|
68 VEC_S16 (16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725);
|
|
69 static const vector_s16_t constants_2 ATTR_ALIGN(16) =
|
12932
|
70 VEC_S16 (16069, 22289, 20995, 18895, 16069, 18895, 20995, 22289);
|
10270
|
71 static const vector_s16_t constants_3 ATTR_ALIGN(16) =
|
|
72 VEC_S16 (21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692);
|
|
73 static const vector_s16_t constants_4 ATTR_ALIGN(16) =
|
12932
|
74 VEC_S16 (13623, 18895, 17799, 16019, 13623, 16019, 17799, 18895);
|
9857
|
75
|
|
76 #define IDCT \
|
|
77 vector_s16_t vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \
|
|
78 vector_s16_t vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \
|
|
79 vector_s16_t a0, a1, a2, ma2, c4, mc4, zero, bias; \
|
|
80 vector_s16_t t0, t1, t2, t3, t4, t5, t6, t7, t8; \
|
|
81 vector_u16_t shift; \
|
|
82 \
|
10270
|
83 c4 = vec_splat (constants, 0); \
|
|
84 a0 = vec_splat (constants, 1); \
|
|
85 a1 = vec_splat (constants, 2); \
|
|
86 a2 = vec_splat (constants, 3); \
|
|
87 mc4 = vec_splat (constants, 4); \
|
|
88 ma2 = vec_splat (constants, 5); \
|
|
89 bias = (vector_s16_t)vec_splat ((vector_s32_t)constants, 3); \
|
9857
|
90 \
|
|
91 zero = vec_splat_s16 (0); \
|
12932
|
92 \
|
|
93 vx0 = vec_adds (block[0], block[4]); \
|
|
94 vx4 = vec_subs (block[0], block[4]); \
|
|
95 t5 = vec_mradds (vx0, constants_1, zero); \
|
|
96 t0 = vec_mradds (vx4, constants_1, zero); \
|
|
97 \
|
|
98 vx1 = vec_mradds (a1, block[7], block[1]); \
|
|
99 vx7 = vec_mradds (a1, block[1], vec_subs (zero, block[7])); \
|
|
100 t1 = vec_mradds (vx1, constants_2, zero); \
|
|
101 t8 = vec_mradds (vx7, constants_2, zero); \
|
|
102 \
|
|
103 vx2 = vec_mradds (a0, block[6], block[2]); \
|
|
104 vx6 = vec_mradds (a0, block[2], vec_subs (zero, block[6])); \
|
|
105 t2 = vec_mradds (vx2, constants_3, zero); \
|
|
106 t4 = vec_mradds (vx6, constants_3, zero); \
|
|
107 \
|
|
108 vx3 = vec_mradds (block[3], constants_4, zero); \
|
|
109 vx5 = vec_mradds (block[5], constants_4, zero); \
|
|
110 t7 = vec_mradds (a2, vx5, vx3); \
|
|
111 t3 = vec_mradds (ma2, vx3, vx5); \
|
9857
|
112 \
|
12932
|
113 t6 = vec_adds (t8, t3); \
|
|
114 t3 = vec_subs (t8, t3); \
|
|
115 t8 = vec_subs (t1, t7); \
|
|
116 t1 = vec_adds (t1, t7); \
|
|
117 t6 = vec_mradds (a0, t6, t6); /* a0+1 == 2*c4 */ \
|
|
118 t1 = vec_mradds (a0, t1, t1); /* a0+1 == 2*c4 */ \
|
9857
|
119 \
|
12932
|
120 t7 = vec_adds (t5, t2); \
|
|
121 t2 = vec_subs (t5, t2); \
|
|
122 t5 = vec_adds (t0, t4); \
|
|
123 t0 = vec_subs (t0, t4); \
|
|
124 t4 = vec_subs (t8, t3); \
|
|
125 t3 = vec_adds (t8, t3); \
|
|
126 \
|
|
127 vy0 = vec_adds (t7, t1); \
|
|
128 vy7 = vec_subs (t7, t1); \
|
|
129 vy1 = vec_adds (t5, t3); \
|
|
130 vy6 = vec_subs (t5, t3); \
|
|
131 vy2 = vec_adds (t0, t4); \
|
|
132 vy5 = vec_subs (t0, t4); \
|
|
133 vy3 = vec_adds (t2, t6); \
|
|
134 vy4 = vec_subs (t2, t6); \
|
9857
|
135 \
|
|
136 vx0 = vec_mergeh (vy0, vy4); \
|
|
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 vy0 = vec_mergeh (vx0, vx4); \
|
|
146 vy1 = vec_mergel (vx0, vx4); \
|
|
147 vy2 = vec_mergeh (vx1, vx5); \
|
|
148 vy3 = vec_mergel (vx1, vx5); \
|
|
149 vy4 = vec_mergeh (vx2, vx6); \
|
|
150 vy5 = vec_mergel (vx2, vx6); \
|
|
151 vy6 = vec_mergeh (vx3, vx7); \
|
|
152 vy7 = vec_mergel (vx3, vx7); \
|
|
153 \
|
12932
|
154 vx0 = vec_mergeh (vy0, vy4); \
|
9857
|
155 vx1 = vec_mergel (vy0, vy4); \
|
|
156 vx2 = vec_mergeh (vy1, vy5); \
|
|
157 vx3 = vec_mergel (vy1, vy5); \
|
|
158 vx4 = vec_mergeh (vy2, vy6); \
|
|
159 vx5 = vec_mergel (vy2, vy6); \
|
|
160 vx6 = vec_mergeh (vy3, vy7); \
|
|
161 vx7 = vec_mergel (vy3, vy7); \
|
|
162 \
|
12932
|
163 vx0 = vec_adds (vx0, bias); \
|
|
164 t5 = vec_adds (vx0, vx4); \
|
|
165 t0 = vec_subs (vx0, vx4); \
|
|
166 \
|
|
167 t1 = vec_mradds (a1, vx7, vx1); \
|
|
168 t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7)); \
|
|
169 \
|
|
170 t2 = vec_mradds (a0, vx6, vx2); \
|
|
171 t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6)); \
|
|
172 \
|
|
173 t7 = vec_mradds (a2, vx5, vx3); \
|
|
174 t3 = vec_mradds (ma2, vx3, vx5); \
|
|
175 \
|
|
176 t6 = vec_adds (t8, t3); \
|
|
177 t3 = vec_subs (t8, t3); \
|
|
178 t8 = vec_subs (t1, t7); \
|
|
179 t1 = vec_adds (t1, t7); \
|
|
180 \
|
|
181 t7 = vec_adds (t5, t2); \
|
|
182 t2 = vec_subs (t5, t2); \
|
|
183 t5 = vec_adds (t0, t4); \
|
|
184 t0 = vec_subs (t0, t4); \
|
|
185 t4 = vec_subs (t8, t3); \
|
|
186 t3 = vec_adds (t8, t3); \
|
|
187 \
|
|
188 vy0 = vec_adds (t7, t1); \
|
|
189 vy7 = vec_subs (t7, t1); \
|
|
190 vy1 = vec_mradds (c4, t3, t5); \
|
|
191 vy6 = vec_mradds (mc4, t3, t5); \
|
|
192 vy2 = vec_mradds (c4, t4, t0); \
|
|
193 vy5 = vec_mradds (mc4, t4, t0); \
|
|
194 vy3 = vec_adds (t2, t6); \
|
|
195 vy4 = vec_subs (t2, t6); \
|
9857
|
196 \
|
|
197 shift = vec_splat_u16 (6); \
|
|
198 vx0 = vec_sra (vy0, shift); \
|
|
199 vx1 = vec_sra (vy1, shift); \
|
|
200 vx2 = vec_sra (vy2, shift); \
|
|
201 vx3 = vec_sra (vy3, shift); \
|
|
202 vx4 = vec_sra (vy4, shift); \
|
|
203 vx5 = vec_sra (vy5, shift); \
|
|
204 vx6 = vec_sra (vy6, shift); \
|
|
205 vx7 = vec_sra (vy7, shift);
|
|
206
|
10270
|
207 void mpeg2_idct_copy_altivec (int16_t * const _block, uint8_t * dest,
|
9857
|
208 const int stride)
|
|
209 {
|
10270
|
210 vector_s16_t * const block = (vector_s16_t *)_block;
|
9857
|
211 vector_u8_t tmp;
|
|
212
|
|
213 IDCT
|
|
214
|
|
215 #define COPY(dest,src) \
|
|
216 tmp = vec_packsu (src, src); \
|
|
217 vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \
|
|
218 vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
|
|
219
|
|
220 COPY (dest, vx0) dest += stride;
|
|
221 COPY (dest, vx1) dest += stride;
|
|
222 COPY (dest, vx2) dest += stride;
|
|
223 COPY (dest, vx3) dest += stride;
|
|
224 COPY (dest, vx4) dest += stride;
|
|
225 COPY (dest, vx5) dest += stride;
|
|
226 COPY (dest, vx6) dest += stride;
|
|
227 COPY (dest, vx7)
|
|
228
|
10270
|
229 block[0] = block[1] = block[2] = block[3] = zero;
|
|
230 block[4] = block[5] = block[6] = block[7] = zero;
|
9857
|
231 }
|
|
232
|
10270
|
233 void mpeg2_idct_add_altivec (const int last, int16_t * const _block,
|
|
234 uint8_t * dest, const int stride)
|
9857
|
235 {
|
10270
|
236 vector_s16_t * const block = (vector_s16_t *)_block;
|
9857
|
237 vector_u8_t tmp;
|
|
238 vector_s16_t tmp2, tmp3;
|
|
239 vector_u8_t perm0;
|
|
240 vector_u8_t perm1;
|
|
241 vector_u8_t p0, p1, p;
|
|
242
|
|
243 IDCT
|
|
244
|
|
245 p0 = vec_lvsl (0, dest);
|
|
246 p1 = vec_lvsl (stride, dest);
|
|
247 p = vec_splat_u8 (-1);
|
|
248 perm0 = vec_mergeh (p, p0);
|
|
249 perm1 = vec_mergeh (p, p1);
|
|
250
|
|
251 #define ADD(dest,src,perm) \
|
|
252 /* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \
|
|
253 tmp = vec_ld (0, dest); \
|
|
254 tmp2 = (vector_s16_t)vec_perm (tmp, (vector_u8_t)zero, perm); \
|
|
255 tmp3 = vec_adds (tmp2, src); \
|
|
256 tmp = vec_packsu (tmp3, tmp3); \
|
|
257 vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \
|
|
258 vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
|
|
259
|
|
260 ADD (dest, vx0, perm0) dest += stride;
|
|
261 ADD (dest, vx1, perm1) dest += stride;
|
|
262 ADD (dest, vx2, perm0) dest += stride;
|
|
263 ADD (dest, vx3, perm1) dest += stride;
|
|
264 ADD (dest, vx4, perm0) dest += stride;
|
|
265 ADD (dest, vx5, perm1) dest += stride;
|
|
266 ADD (dest, vx6, perm0) dest += stride;
|
|
267 ADD (dest, vx7, perm1)
|
|
268
|
10270
|
269 block[0] = block[1] = block[2] = block[3] = zero;
|
|
270 block[4] = block[5] = block[6] = block[7] = zero;
|
9857
|
271 }
|
|
272
|
10270
|
273 void mpeg2_idct_altivec_init (void)
|
|
274 {
|
|
275 extern uint8_t mpeg2_scan_norm[64];
|
|
276 extern uint8_t mpeg2_scan_alt[64];
|
|
277 int i, j;
|
|
278
|
|
279 /* the altivec idct uses a transposed input, so we patch scan tables */
|
|
280 for (i = 0; i < 64; i++) {
|
|
281 j = mpeg2_scan_norm[i];
|
|
282 mpeg2_scan_norm[i] = (j >> 3) | ((j & 7) << 3);
|
|
283 j = mpeg2_scan_alt[i];
|
|
284 mpeg2_scan_alt[i] = (j >> 3) | ((j & 7) << 3);
|
|
285 }
|
|
286 }
|
|
287
|
|
288 #endif
|