64153ad6cd0d3ba9aefd7bbf4d72c91ab91a98a7
[libav.git] / libavcodec / armv4l / dsputil_arm.c
1 /*
2 * ARMv4L optimized DSP utils
3 * Copyright (c) 2001 Lionel Ulmer.
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "libavcodec/dsputil.h"
23 #ifdef HAVE_IPP
24 #include <ipp.h>
25 #endif
26
27 extern void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx);
28 extern void ff_float_init_arm_vfp(DSPContext* c, AVCodecContext *avctx);
29
30 extern void j_rev_dct_ARM(DCTELEM *data);
31 extern void simple_idct_ARM(DCTELEM *data);
32
33 extern void simple_idct_armv5te(DCTELEM *data);
34 extern void simple_idct_put_armv5te(uint8_t *dest, int line_size,
35 DCTELEM *data);
36 extern void simple_idct_add_armv5te(uint8_t *dest, int line_size,
37 DCTELEM *data);
38
39 extern void ff_simple_idct_armv6(DCTELEM *data);
40 extern void ff_simple_idct_put_armv6(uint8_t *dest, int line_size,
41 DCTELEM *data);
42 extern void ff_simple_idct_add_armv6(uint8_t *dest, int line_size,
43 DCTELEM *data);
44
45 /* XXX: local hack */
46 static void (*ff_put_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
47 static void (*ff_add_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
48
49 void put_pixels8_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
50 void put_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
51 void put_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
52 void put_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
53
54 void put_no_rnd_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
55 void put_no_rnd_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
56 void put_no_rnd_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
57
58 void put_pixels16_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
59
60 extern void ff_prefetch_arm(void *mem, int stride, int h);
61
62 CALL_2X_PIXELS(put_pixels16_x2_arm , put_pixels8_x2_arm , 8)
63 CALL_2X_PIXELS(put_pixels16_y2_arm , put_pixels8_y2_arm , 8)
64 CALL_2X_PIXELS(put_pixels16_xy2_arm, put_pixels8_xy2_arm, 8)
65 CALL_2X_PIXELS(put_no_rnd_pixels16_x2_arm , put_no_rnd_pixels8_x2_arm , 8)
66 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_arm , put_no_rnd_pixels8_y2_arm , 8)
67 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_arm, put_no_rnd_pixels8_xy2_arm, 8)
68
69 extern void ff_add_pixels_clamped_ARM(short *block, unsigned char *dest,
70 int line_size);
71
72 /* XXX: those functions should be suppressed ASAP when all IDCTs are
73 converted */
74 static void j_rev_dct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
75 {
76 j_rev_dct_ARM (block);
77 ff_put_pixels_clamped(block, dest, line_size);
78 }
79 static void j_rev_dct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
80 {
81 j_rev_dct_ARM (block);
82 ff_add_pixels_clamped(block, dest, line_size);
83 }
84 static void simple_idct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
85 {
86 simple_idct_ARM (block);
87 ff_put_pixels_clamped(block, dest, line_size);
88 }
89 static void simple_idct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
90 {
91 simple_idct_ARM (block);
92 ff_add_pixels_clamped(block, dest, line_size);
93 }
94
95 #ifdef HAVE_IPP
96 static void simple_idct_ipp(DCTELEM *block)
97 {
98 ippiDCT8x8Inv_Video_16s_C1I(block);
99 }
100 static void simple_idct_ipp_put(uint8_t *dest, int line_size, DCTELEM *block)
101 {
102 ippiDCT8x8Inv_Video_16s8u_C1R(block, dest, line_size);
103 }
104
105 void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size);
106
107 static void simple_idct_ipp_add(uint8_t *dest, int line_size, DCTELEM *block)
108 {
109 ippiDCT8x8Inv_Video_16s_C1I(block);
110 #ifdef HAVE_IWMMXT
111 add_pixels_clamped_iwmmxt(block, dest, line_size);
112 #else
113 ff_add_pixels_clamped_ARM(block, dest, line_size);
114 #endif
115 }
116 #endif
117
118 int mm_support(void)
119 {
120 return ENABLE_IWMMXT * MM_IWMMXT;
121 }
122
123 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx)
124 {
125 int idct_algo= avctx->idct_algo;
126
127 ff_put_pixels_clamped = c->put_pixels_clamped;
128 ff_add_pixels_clamped = c->add_pixels_clamped;
129
130 if (avctx->lowres == 0) {
131 if(idct_algo == FF_IDCT_AUTO){
132 #if defined(HAVE_IPP)
133 idct_algo = FF_IDCT_IPP;
134 #elif defined(HAVE_ARMV6)
135 idct_algo = FF_IDCT_SIMPLEARMV6;
136 #elif defined(HAVE_ARMV5TE)
137 idct_algo = FF_IDCT_SIMPLEARMV5TE;
138 #else
139 idct_algo = FF_IDCT_ARM;
140 #endif
141 }
142
143 if(idct_algo==FF_IDCT_ARM){
144 c->idct_put= j_rev_dct_ARM_put;
145 c->idct_add= j_rev_dct_ARM_add;
146 c->idct = j_rev_dct_ARM;
147 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;/* FF_NO_IDCT_PERM */
148 } else if (idct_algo==FF_IDCT_SIMPLEARM){
149 c->idct_put= simple_idct_ARM_put;
150 c->idct_add= simple_idct_ARM_add;
151 c->idct = simple_idct_ARM;
152 c->idct_permutation_type= FF_NO_IDCT_PERM;
153 #ifdef HAVE_ARMV6
154 } else if (idct_algo==FF_IDCT_SIMPLEARMV6){
155 c->idct_put= ff_simple_idct_put_armv6;
156 c->idct_add= ff_simple_idct_add_armv6;
157 c->idct = ff_simple_idct_armv6;
158 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
159 #endif
160 #ifdef HAVE_ARMV5TE
161 } else if (idct_algo==FF_IDCT_SIMPLEARMV5TE){
162 c->idct_put= simple_idct_put_armv5te;
163 c->idct_add= simple_idct_add_armv5te;
164 c->idct = simple_idct_armv5te;
165 c->idct_permutation_type = FF_NO_IDCT_PERM;
166 #endif
167 #ifdef HAVE_IPP
168 } else if (idct_algo==FF_IDCT_IPP){
169 c->idct_put= simple_idct_ipp_put;
170 c->idct_add= simple_idct_ipp_add;
171 c->idct = simple_idct_ipp;
172 c->idct_permutation_type= FF_NO_IDCT_PERM;
173 #endif
174 }
175 }
176
177 c->put_pixels_tab[0][0] = put_pixels16_arm;
178 c->put_pixels_tab[0][1] = put_pixels16_x2_arm; //OK!
179 c->put_pixels_tab[0][2] = put_pixels16_y2_arm; //OK!
180 c->put_pixels_tab[0][3] = put_pixels16_xy2_arm;
181 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_arm;
182 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_arm; // OK
183 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_arm; //OK
184 c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_arm;
185 c->put_pixels_tab[1][0] = put_pixels8_arm; //OK
186 c->put_pixels_tab[1][1] = put_pixels8_x2_arm; //OK
187 c->put_pixels_tab[1][2] = put_pixels8_y2_arm;
188 c->put_pixels_tab[1][3] = put_pixels8_xy2_arm;
189 c->put_no_rnd_pixels_tab[1][0] = put_pixels8_arm;//OK
190 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_arm; //OK
191 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_arm; //OK
192 c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_arm;
193
194 #ifdef HAVE_ARMV5TE
195 c->prefetch = ff_prefetch_arm;
196 #endif
197
198 #ifdef HAVE_IWMMXT
199 dsputil_init_iwmmxt(c, avctx);
200 #endif
201 #ifdef HAVE_ARMVFP
202 ff_float_init_arm_vfp(c, avctx);
203 #endif
204 }