2 * ARMv4L optimized DSP utils
3 * Copyright (c) 2001 Lionel Ulmer.
5 * This file is part of FFmpeg.
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.
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.
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
27 extern void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx);
29 extern void j_rev_dct_ARM(DCTELEM *data);
30 extern void simple_idct_ARM(DCTELEM *data);
32 extern void simple_idct_armv5te(DCTELEM *data);
33 extern void simple_idct_put_armv5te(uint8_t *dest, int line_size,
35 extern void simple_idct_add_armv5te(uint8_t *dest, int line_size,
38 extern void ff_simple_idct_armv6(DCTELEM *data);
39 extern void ff_simple_idct_put_armv6(uint8_t *dest, int line_size,
41 extern void ff_simple_idct_add_armv6(uint8_t *dest, int line_size,
45 static void (*ff_put_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
46 static void (*ff_add_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
48 void put_pixels8_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
49 void put_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
50 void put_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
51 void put_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
53 void put_no_rnd_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
54 void put_no_rnd_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
55 void put_no_rnd_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
57 void put_pixels16_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
59 CALL_2X_PIXELS(put_pixels16_x2_arm , put_pixels8_x2_arm , 8)
60 CALL_2X_PIXELS(put_pixels16_y2_arm , put_pixels8_y2_arm , 8)
61 CALL_2X_PIXELS(put_pixels16_xy2_arm, put_pixels8_xy2_arm, 8)
62 CALL_2X_PIXELS(put_no_rnd_pixels16_x2_arm , put_no_rnd_pixels8_x2_arm , 8)
63 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_arm , put_no_rnd_pixels8_y2_arm , 8)
64 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_arm, put_no_rnd_pixels8_xy2_arm, 8)
66 static void add_pixels_clamped_ARM(short *block, unsigned char *dest, int line_size)
75 /* block[0] and block[1]*/
77 "ldrsh r7, [%0, #2] \n\t"
78 "and r6, r4, #0xFF \n\t"
79 "and r8, r4, #0xFF00 \n\t"
81 "add r8, r7, r8, lsr #8 \n\t"
85 "movne r6, r5, lsr #24 \n\t"
87 "movne r8, r7, lsr #24 \n\t"
89 "ldrsh r5, [%0, #4] \n\t" /* moved form [A] */
90 "orr r9, r9, r8, lsl #8 \n\t"
91 /* block[2] and block[3] */
93 "ldrsh r7, [%0, #6] \n\t"
94 "and r6, r4, #0xFF0000 \n\t"
95 "and r8, r4, #0xFF000000 \n\t"
96 "add r6, r5, r6, lsr #16 \n\t"
97 "add r8, r7, r8, lsr #24 \n\t"
100 "tst r6, #0x100 \n\t"
101 "movne r6, r5, lsr #24 \n\t"
102 "tst r8, #0x100 \n\t"
103 "movne r8, r7, lsr #24 \n\t"
104 "orr r9, r9, r6, lsl #16 \n\t"
105 "ldr r4, [%1, #4] \n\t" /* moved form [B] */
106 "orr r9, r9, r8, lsl #24 \n\t"
108 "ldrsh r5, [%0, #8] \n\t" /* moved form [C] */
113 /* block[4] and block[5] */
115 "ldrsh r7, [%0, #10] \n\t"
116 "and r6, r4, #0xFF \n\t"
117 "and r8, r4, #0xFF00 \n\t"
118 "add r6, r5, r6 \n\t"
119 "add r8, r7, r8, lsr #8 \n\t"
122 "tst r6, #0x100 \n\t"
123 "movne r6, r5, lsr #24 \n\t"
124 "tst r8, #0x100 \n\t"
125 "movne r8, r7, lsr #24 \n\t"
127 "ldrsh r5, [%0, #12] \n\t" /* moved from [D] */
128 "orr r9, r9, r8, lsl #8 \n\t"
129 /* block[6] and block[7] */
131 "ldrsh r7, [%0, #14] \n\t"
132 "and r6, r4, #0xFF0000 \n\t"
133 "and r8, r4, #0xFF000000 \n\t"
134 "add r6, r5, r6, lsr #16 \n\t"
135 "add r8, r7, r8, lsr #24 \n\t"
138 "tst r6, #0x100 \n\t"
139 "movne r6, r5, lsr #24 \n\t"
140 "tst r8, #0x100 \n\t"
141 "movne r8, r7, lsr #24 \n\t"
142 "orr r9, r9, r6, lsl #16 \n\t"
143 "add %0, %0, #16 \n\t" /* moved from [E] */
144 "orr r9, r9, r8, lsl #24 \n\t"
145 "subs r10, r10, #1 \n\t" /* moved from [F] */
147 "str r9, [%1, #4] \n\t"
151 "add %1, %1, %2 \n\t"
156 : "r4", "r5", "r6", "r7", "r8", "r9", "r10", "cc", "memory" );
159 /* XXX: those functions should be suppressed ASAP when all IDCTs are
161 static void j_rev_dct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
163 j_rev_dct_ARM (block);
164 ff_put_pixels_clamped(block, dest, line_size);
166 static void j_rev_dct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
168 j_rev_dct_ARM (block);
169 ff_add_pixels_clamped(block, dest, line_size);
171 static void simple_idct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
173 simple_idct_ARM (block);
174 ff_put_pixels_clamped(block, dest, line_size);
176 static void simple_idct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
178 simple_idct_ARM (block);
179 ff_add_pixels_clamped(block, dest, line_size);
183 static void simple_idct_ipp(DCTELEM *block)
185 ippiDCT8x8Inv_Video_16s_C1I(block);
187 static void simple_idct_ipp_put(uint8_t *dest, int line_size, DCTELEM *block)
189 ippiDCT8x8Inv_Video_16s8u_C1R(block, dest, line_size);
192 void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size);
194 static void simple_idct_ipp_add(uint8_t *dest, int line_size, DCTELEM *block)
196 ippiDCT8x8Inv_Video_16s_C1I(block);
198 add_pixels_clamped_iwmmxt(block, dest, line_size);
200 add_pixels_clamped_ARM(block, dest, line_size);
205 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx)
207 int idct_algo= avctx->idct_algo;
209 ff_put_pixels_clamped = c->put_pixels_clamped;
210 ff_add_pixels_clamped = c->add_pixels_clamped;
212 if (avctx->lowres == 0) {
213 if(idct_algo == FF_IDCT_AUTO){
214 #if defined(HAVE_IPP)
215 idct_algo = FF_IDCT_IPP;
216 #elif defined(HAVE_ARMV6)
217 idct_algo = FF_IDCT_SIMPLEARMV6;
218 #elif defined(HAVE_ARMV5TE)
219 idct_algo = FF_IDCT_SIMPLEARMV5TE;
221 idct_algo = FF_IDCT_ARM;
225 if(idct_algo==FF_IDCT_ARM){
226 c->idct_put= j_rev_dct_ARM_put;
227 c->idct_add= j_rev_dct_ARM_add;
228 c->idct = j_rev_dct_ARM;
229 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;/* FF_NO_IDCT_PERM */
230 } else if (idct_algo==FF_IDCT_SIMPLEARM){
231 c->idct_put= simple_idct_ARM_put;
232 c->idct_add= simple_idct_ARM_add;
233 c->idct = simple_idct_ARM;
234 c->idct_permutation_type= FF_NO_IDCT_PERM;
236 } else if (idct_algo==FF_IDCT_SIMPLEARMV6){
237 c->idct_put= ff_simple_idct_put_armv6;
238 c->idct_add= ff_simple_idct_add_armv6;
239 c->idct = ff_simple_idct_armv6;
240 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
243 } else if (idct_algo==FF_IDCT_SIMPLEARMV5TE){
244 c->idct_put= simple_idct_put_armv5te;
245 c->idct_add= simple_idct_add_armv5te;
246 c->idct = simple_idct_armv5te;
247 c->idct_permutation_type = FF_NO_IDCT_PERM;
250 } else if (idct_algo==FF_IDCT_IPP){
251 c->idct_put= simple_idct_ipp_put;
252 c->idct_add= simple_idct_ipp_add;
253 c->idct = simple_idct_ipp;
254 c->idct_permutation_type= FF_NO_IDCT_PERM;
259 c->put_pixels_tab[0][0] = put_pixels16_arm;
260 c->put_pixels_tab[0][1] = put_pixels16_x2_arm; //OK!
261 c->put_pixels_tab[0][2] = put_pixels16_y2_arm; //OK!
262 c->put_pixels_tab[0][3] = put_pixels16_xy2_arm;
263 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_arm;
264 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_arm; // OK
265 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_arm; //OK
266 c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_arm;
267 c->put_pixels_tab[1][0] = put_pixels8_arm; //OK
268 c->put_pixels_tab[1][1] = put_pixels8_x2_arm; //OK
269 c->put_pixels_tab[1][2] = put_pixels8_y2_arm;
270 c->put_pixels_tab[1][3] = put_pixels8_xy2_arm;
271 c->put_no_rnd_pixels_tab[1][0] = put_pixels8_arm;//OK
272 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_arm; //OK
273 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_arm; //OK
274 c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_arm;
277 dsputil_init_iwmmxt(c, avctx);