2 * Blackfin Pixel Operations
3 * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "config_bfin.h"
28 * Halfpel motion compensation with rounding (a+b+1)>>1.
29 * This is an array[4][4] of motion compensation funcions for 4
30 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
31 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
32 * @param block destination where the result is stored
33 * @param pixels source
34 * @param line_size number of bytes in a horizontal line of block
39 DEFUN(put_pixels8uc,mL1,
40 (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
41 int dest_size, int line_size, int h)):
45 r0=[sp+12]; // dest_size
46 r2=[sp+16]; // line_size
53 LSETUP(pp8$0,pp8$1) LC0=P0;
54 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
56 pp8$0: DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
57 R6 = BYTEOP1P(R1:0,R3:2) || R0 = [I0++M0]|| R2 =[I1++M0];
58 R7 = BYTEOP1P(R1:0,R3:2)(R) || R0 = [I0++] || [I3++] = R6 ;
59 pp8$1: DISALGNEXCPT || R2 = [I1++] || [I3++M3] = R7;
63 DEFUN_END(put_pixels8uc)
65 DEFUN(put_pixels16uc,mL1,
66 (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
67 int dest_size, int line_size, int h)):
73 r0=[fp+20]; // dest_size
74 r2=[fp+24]; // line_size
83 LSETUP(pp16$0,pp16$1) LC0=P0;
84 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
86 pp16$0: DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
87 R6 = BYTEOP1P(R1:0,R3:2) || R0 = [I0++] || R2 =[I1++];
88 R7 = BYTEOP1P(R1:0,R3:2)(R) || R1 = [I0++] || R3 =[I1++];
90 R6 = BYTEOP1P(R1:0,R3:2) || R0 = [I0++M0] || R2 =[I1++M0];
91 R7 = BYTEOP1P(R1:0,R3:2)(R) || R0 = [I0++] || [I3++] = R7 ;
93 pp16$1: DISALGNEXCPT || R2 = [I1++] || [I3++M3] = R7;
98 DEFUN_END(put_pixels16uc)
105 DEFUN(put_pixels8uc_nornd,mL1,
106 (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
107 int line_size, int h)):
111 r2=[sp+12]; // line_size
118 LSETUP(pp8$2,pp8$3) LC0=P0;
119 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
121 pp8$2: DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
122 R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++M0]|| R2 =[I1++M0];
123 R7 = BYTEOP1P(R1:0,R3:2)(T,R) || R0 = [I0++] || [I3++] = R6 ;
124 pp8$3: DISALGNEXCPT || R2 = [I1++] || [I3++M3] = R7;
128 DEFUN_END(put_pixels8uc_nornd)
130 DEFUN(put_pixels16uc_nornd,mL1,
131 (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
132 int line_size, int h)):
136 r2=[sp+12]; // line_size
145 LSETUP(pp16$2,pp16$3) LC0=P0;
146 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
149 DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
150 R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++] || R2 =[I1++];
151 R7 = BYTEOP1P(R1:0,R3:2)(T,R) || R1 = [I0++] || R3 =[I1++];
154 R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++M0] || R2 =[I1++M0];
155 R7 = BYTEOP1P(R1:0,R3:2)(T,R) || R0 = [I0++] || [I3++] = R7 ;
157 pp16$3: DISALGNEXCPT || R2 = [I1++] || [I3++M3] = R7;
162 DEFUN_END(put_pixels16uc_nornd)
164 DEFUN(z_put_pixels16_xy2,mL1,
165 (uint8_t *block, const uint8_t *s0,
166 int dest_size, int line_size, int h)):
170 i0=r1; // src0--> pixels
171 i1=r1; // src1--> pixels + line_size
173 m2=r2; // m2=dest_width-4
178 i1+=m3; /* src1 + line_size */
179 m0=r2; /* line-size - 20 */
185 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
187 LSETUP(LS$16E,LE$16E) LC0=P0;
188 LS$16E: DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
189 R4 = BYTEOP2P (R3:2,R1:0) (RNDL) || R0 = [I0++] || R2 =[I1++];
190 R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R1 = [I0++] || [I3++] = R4 ;
191 DISALGNEXCPT || R3 = [I1++] || [I3++] = R5;
192 R4 = BYTEOP2P (R3:2,R1:0) (RNDL) || R0 = [I0++M0]|| R2 = [I1++M0];
193 R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R0 = [I0++] || [I3++] = R4 ;
194 LE$16E: DISALGNEXCPT || R2 = [I1++] || [I3++M2] = R5;
204 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
205 LSETUP(LS$16O,LE$16O) LC0=P0;
206 LS$16O: DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
207 R4 = BYTEOP2P (R3:2,R1:0) (RNDH) || R0 = [I0++] || R2 =[I1++];
208 R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R1 = [I0++] || R6 =[I3++];
209 R4 = R4 +|+ R6 || R7 = [I3--];
210 R5 = R5 +|+ R7 || [I3++] = R4;
211 DISALGNEXCPT || R3 =[I1++] || [I3++] = R5;
212 R4 = BYTEOP2P (R3:2,R1:0) (RNDH) || R0 = [I0++M0]|| R2 = [I1++M0];
213 R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R0 = [I0++] || R6 = [I3++];
214 R4 = R4 +|+ R6 || R7 = [I3--];
215 R5 = R5 +|+ R7 || [I3++] = R4;
216 LE$16O: DISALGNEXCPT || R2 = [I1++] || [I3++M2] = R5;
221 DEFUN_END(z_put_pixels16_xy2)
223 DEFUN(put_pixels16_xy2_nornd,mL1,
224 (uint8_t *block, const uint8_t *s0,
225 int line_size, int h)):
229 i0=r1; // src0--> pixels
230 i1=r1; // src1--> pixels + line_size
235 i1+=m3; /* src1 + line_size */
236 m0=r2; /* line-size - 20 */
243 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
245 LSETUP(LS$16ET,LE$16ET) LC0=P0;
246 LS$16ET:DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
247 R4 = BYTEOP2P (R3:2,R1:0) (TL) || R0 = [I0++] || R2 =[I1++];
248 R5 = BYTEOP2P (R3:2,R1:0) (TL,R) || R1 = [I0++] || [I3++] = R4 ;
249 DISALGNEXCPT || R3 = [I1++] || [I3++] = R5;
250 R4 = BYTEOP2P (R3:2,R1:0) (TL) || R0 = [I0++M0]|| R2 = [I1++M0];
251 R5 = BYTEOP2P (R3:2,R1:0) (TL,R) || R0 = [I0++] || [I3++] = R4 ;
252 LE$16ET:DISALGNEXCPT || R2 = [I1++] || [I3++M2] = R5;
262 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
263 LSETUP(LS$16OT,LE$16OT) LC0=P0;
264 LS$16OT:DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
265 R4 = BYTEOP2P (R3:2,R1:0) (TH) || R0 = [I0++] || R2 =[I1++];
266 R5 = BYTEOP2P (R3:2,R1:0) (TH,R) || R1 = [I0++] || R6 =[I3++];
267 R4 = R4 +|+ R6 || R7 = [I3--];
268 R5 = R5 +|+ R7 || [I3++] = R4;
269 DISALGNEXCPT || R3 =[I1++] || [I3++] = R5;
270 R4 = BYTEOP2P (R3:2,R1:0) (TH) || R0 = [I0++M0]|| R2 = [I1++M0];
271 R5 = BYTEOP2P (R3:2,R1:0) (TH,R) || R0 = [I0++] || R6 = [I3++];
272 R4 = R4 +|+ R6 || R7 = [I3--];
273 R5 = R5 +|+ R7 || [I3++] = R4;
274 LE$16OT:DISALGNEXCPT || R2 = [I1++] || [I3++M2] = R5;
279 DEFUN_END(put_pixels16_xy2_nornd)
281 DEFUN(z_put_pixels8_xy2,mL1,
282 (uint8_t *block, const uint8_t *s0,
283 int dest_size, int line_size, int h)):
287 i0=r1; // src0--> pixels
288 i1=r1; // src1--> pixels + line_size
290 m2=r2; // m2=dest_width-4
295 i1+=m3; /* src1 + line_size */
296 m0=r2; /* line-size - 20 */
302 LSETUP(LS$8E,LE$8E) LC0=P0;
303 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
304 LS$8E: DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
305 R4 = BYTEOP2P (R3:2,R1:0) (RNDL) || R0 = [I0++M0] || R2 =[I1++M0];
306 R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R0 = [I0++] || [I3++] = R4 ;
307 LE$8E: DISALGNEXCPT || R2 = [I1++] || [I3++M2] = R5;
317 LSETUP(LS$8O,LE$8O) LC0=P0;
318 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
319 LS$8O: DISALGNEXCPT || R1 = [I0++] || R3 =[I1++];
320 R4 = BYTEOP2P (R3:2,R1:0) (RNDH) || R0 = [I0++M0] || R2 =[I1++M0];
321 R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R0 = [I0++] || R6 =[I3++];
322 R4 = R4 +|+ R6 || R7 = [I3--];
323 R5 = R5 +|+ R7 || [I3++] = R4;
324 LE$8O: DISALGNEXCPT || R2 =[I1++] || [I3++M2] = R5;
329 DEFUN_END(z_put_pixels8_xy2)
331 DEFUN(put_pixels8_xy2_nornd,mL1,
332 (uint8_t *block, const uint8_t *s0, int line_size, int h)):
336 i0=r1; // src0--> pixels
337 i1=r1; // src1--> pixels + line_size
342 i1+=m3; /* src1 + line_size */
343 m0=r2; /* line-size - 20 */
351 LSETUP(LS$8ET,LE$8ET) LC0=P0;
352 DISALGNEXCPT || R0 = [I0++] || R2 =[I1++];
354 LS$8ET: DISALGNEXCPT || R1 = [I0++] || R3 = [I1++];
355 R4 = BYTEOP2P (R3:2,R1:0) (TL) || R0 = [I0++M0] || R2 = [I1++M0];
356 R5 = BYTEOP2P (R3:2,R1:0) (TL,R) || R0 = [I0++] || [I3++] = R4 ;
357 LE$8ET: DISALGNEXCPT || R2 = [I1++] || [I3++M2] = R5;
367 LSETUP(LS$8OT,LE$8OT) LC0=P0;
368 DISALGNEXCPT || R0 = [I0++] || R2 = [I1++];
370 LS$8OT: DISALGNEXCPT || R1 = [I0++] || R3 = [I1++];
371 R4 = BYTEOP2P (R3:2,R1:0) (TH) || R0 = [I0++M0] || R2 = [I1++M0];
372 R5 = BYTEOP2P (R3:2,R1:0) (TH,R) || R0 = [I0++] || R6 = [I3++];
373 R4 = R4 +|+ R6 || R7 = [I3--];
374 R5 = R5 +|+ R7 || [I3++] = R4;
375 LE$8OT: DISALGNEXCPT || R2 =[I1++] || [I3++M2] = R5;
380 DEFUN_END(put_pixels8_xy2_nornd)