2 * Copyright (c) 2002 Brian Foley
3 * Copyright (c) 2002 Dieter Shirley
4 * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 #include "libavutil/attributes.h"
30 #include "libavutil/cpu.h"
31 #include "libavutil/ppc/cpu.h"
32 #include "libavutil/ppc/types_altivec.h"
33 #include "libavutil/ppc/util_altivec.h"
34 #include "libavcodec/hpeldsp.h"
35 #include "hpeldsp_altivec.h"
38 /* next one assumes that ((line_size % 16) == 0) */
39 void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
41 register vector unsigned char pixelsv1;
42 register vector unsigned char pixelsv1B;
43 register vector unsigned char pixelsv1C;
44 register vector unsigned char pixelsv1D;
47 register ptrdiff_t line_size_2 = line_size << 1;
48 register ptrdiff_t line_size_3 = line_size + line_size_2;
49 register ptrdiff_t line_size_4 = line_size << 2;
51 // hand-unrolling the loop by 4 gains about 15%
52 // mininum execution time goes from 74 to 60 cycles
53 // it's faster than -funroll-loops, but using
54 // -funroll-loops w/ this is bad - 74 cycles again.
55 // all this is on a 7450, tuning for the 7450
56 for (i = 0; i < h; i += 4) {
57 pixelsv1 = unaligned_load( 0, pixels);
58 pixelsv1B = unaligned_load(line_size, pixels);
59 pixelsv1C = unaligned_load(line_size_2, pixels);
60 pixelsv1D = unaligned_load(line_size_3, pixels);
61 VEC_ST(pixelsv1, 0, (unsigned char*)block);
62 VEC_ST(pixelsv1B, line_size, (unsigned char*)block);
63 VEC_ST(pixelsv1C, line_size_2, (unsigned char*)block);
64 VEC_ST(pixelsv1D, line_size_3, (unsigned char*)block);
70 /* next one assumes that ((line_size % 16) == 0) */
71 #define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
72 void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
74 register vector unsigned char pixelsv, blockv;
77 for (i = 0; i < h; i++) {
78 blockv = vec_ld(0, block);
79 pixelsv = VEC_LD( 0, pixels);
80 blockv = vec_avg(blockv,pixelsv);
81 vec_st(blockv, 0, (unsigned char*)block);
87 /* next one assumes that ((line_size % 8) == 0) */
88 static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
90 register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
93 for (i = 0; i < h; i++) {
94 /* block is 8 bytes-aligned, so we're either in the
95 left block (16 bytes-aligned) or in the right block (not) */
96 int rightside = ((unsigned long)block & 0x0000000F);
98 blockv = vec_ld(0, block);
99 pixelsv = VEC_LD( 0, pixels);
102 pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1));
104 pixelsv = vec_perm(blockv, pixelsv, vcprm(s0,s1,2,3));
107 blockv = vec_avg(blockv, pixelsv);
109 vec_st(blockv, 0, block);
116 /* next one assumes that ((line_size % 8) == 0) */
117 static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
120 register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
121 register vector unsigned char blockv;
122 register vector unsigned short pixelssum1, pixelssum2, temp3;
123 register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
124 register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
126 pixelsv1 = VEC_LD(0, pixels);
127 pixelsv2 = VEC_LD(1, pixels);
128 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
129 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
131 pixelssum1 = vec_add((vector unsigned short)pixelsv1,
132 (vector unsigned short)pixelsv2);
133 pixelssum1 = vec_add(pixelssum1, vctwo);
135 for (i = 0; i < h ; i++) {
136 int rightside = ((unsigned long)block & 0x0000000F);
137 blockv = vec_ld(0, block);
139 pixelsv1 = unaligned_load(line_size, pixels);
140 pixelsv2 = unaligned_load(line_size+1, pixels);
141 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
142 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
143 pixelssum2 = vec_add((vector unsigned short)pixelsv1,
144 (vector unsigned short)pixelsv2);
145 temp3 = vec_add(pixelssum1, pixelssum2);
146 temp3 = vec_sra(temp3, vctwo);
147 pixelssum1 = vec_add(pixelssum2, vctwo);
148 pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
151 blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
153 blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
156 vec_st(blockv, 0, block);
163 /* next one assumes that ((line_size % 8) == 0) */
164 static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
167 register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
168 register vector unsigned char blockv;
169 register vector unsigned short pixelssum1, pixelssum2, temp3;
170 register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
171 register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
172 register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
174 pixelsv1 = VEC_LD(0, pixels);
175 pixelsv2 = VEC_LD(1, pixels);
176 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
177 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
178 pixelssum1 = vec_add((vector unsigned short)pixelsv1,
179 (vector unsigned short)pixelsv2);
180 pixelssum1 = vec_add(pixelssum1, vcone);
182 for (i = 0; i < h ; i++) {
183 int rightside = ((unsigned long)block & 0x0000000F);
184 blockv = vec_ld(0, block);
186 pixelsv1 = unaligned_load(line_size, pixels);
187 pixelsv2 = unaligned_load(line_size+1, pixels);
188 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
189 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
190 pixelssum2 = vec_add((vector unsigned short)pixelsv1,
191 (vector unsigned short)pixelsv2);
192 temp3 = vec_add(pixelssum1, pixelssum2);
193 temp3 = vec_sra(temp3, vctwo);
194 pixelssum1 = vec_add(pixelssum2, vcone);
195 pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
198 blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
200 blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
203 vec_st(blockv, 0, block);
210 /* next one assumes that ((line_size % 16) == 0) */
211 static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
214 register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
215 register vector unsigned char blockv;
216 register vector unsigned short temp3, temp4,
217 pixelssum1, pixelssum2, pixelssum3, pixelssum4;
218 register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
219 register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
221 pixelsv1 = VEC_LD(0, pixels);
222 pixelsv2 = VEC_LD(1, pixels);
223 pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
224 pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
225 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
226 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
227 pixelssum3 = vec_add((vector unsigned short)pixelsv3,
228 (vector unsigned short)pixelsv4);
229 pixelssum3 = vec_add(pixelssum3, vctwo);
230 pixelssum1 = vec_add((vector unsigned short)pixelsv1,
231 (vector unsigned short)pixelsv2);
232 pixelssum1 = vec_add(pixelssum1, vctwo);
234 for (i = 0; i < h ; i++) {
235 blockv = vec_ld(0, block);
237 pixelsv1 = unaligned_load(line_size, pixels);
238 pixelsv2 = unaligned_load(line_size+1, pixels);
240 pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
241 pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
242 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
243 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
244 pixelssum4 = vec_add((vector unsigned short)pixelsv3,
245 (vector unsigned short)pixelsv4);
246 pixelssum2 = vec_add((vector unsigned short)pixelsv1,
247 (vector unsigned short)pixelsv2);
248 temp4 = vec_add(pixelssum3, pixelssum4);
249 temp4 = vec_sra(temp4, vctwo);
250 temp3 = vec_add(pixelssum1, pixelssum2);
251 temp3 = vec_sra(temp3, vctwo);
253 pixelssum3 = vec_add(pixelssum4, vctwo);
254 pixelssum1 = vec_add(pixelssum2, vctwo);
256 blockv = vec_packsu(temp3, temp4);
258 vec_st(blockv, 0, block);
265 /* next one assumes that ((line_size % 16) == 0) */
266 static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
269 register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
270 register vector unsigned char blockv;
271 register vector unsigned short temp3, temp4,
272 pixelssum1, pixelssum2, pixelssum3, pixelssum4;
273 register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
274 register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
275 register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
277 pixelsv1 = VEC_LD(0, pixels);
278 pixelsv2 = VEC_LD(1, pixels);
279 pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
280 pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
281 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
282 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
283 pixelssum3 = vec_add((vector unsigned short)pixelsv3,
284 (vector unsigned short)pixelsv4);
285 pixelssum3 = vec_add(pixelssum3, vcone);
286 pixelssum1 = vec_add((vector unsigned short)pixelsv1,
287 (vector unsigned short)pixelsv2);
288 pixelssum1 = vec_add(pixelssum1, vcone);
290 for (i = 0; i < h ; i++) {
291 pixelsv1 = unaligned_load(line_size, pixels);
292 pixelsv2 = unaligned_load(line_size+1, pixels);
294 pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
295 pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
296 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
297 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
298 pixelssum4 = vec_add((vector unsigned short)pixelsv3,
299 (vector unsigned short)pixelsv4);
300 pixelssum2 = vec_add((vector unsigned short)pixelsv1,
301 (vector unsigned short)pixelsv2);
302 temp4 = vec_add(pixelssum3, pixelssum4);
303 temp4 = vec_sra(temp4, vctwo);
304 temp3 = vec_add(pixelssum1, pixelssum2);
305 temp3 = vec_sra(temp3, vctwo);
307 pixelssum3 = vec_add(pixelssum4, vcone);
308 pixelssum1 = vec_add(pixelssum2, vcone);
310 blockv = vec_packsu(temp3, temp4);
312 VEC_ST(blockv, 0, block);
319 /* next one assumes that ((line_size % 8) == 0) */
320 static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
323 register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
324 register vector unsigned char blockv, blocktemp;
325 register vector unsigned short pixelssum1, pixelssum2, temp3;
327 register const vector unsigned char vczero = (const vector unsigned char)
329 register const vector unsigned short vctwo = (const vector unsigned short)
332 pixelsv1 = VEC_LD(0, pixels);
333 pixelsv2 = VEC_LD(1, pixels);
334 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
335 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
336 pixelssum1 = vec_add((vector unsigned short)pixelsv1,
337 (vector unsigned short)pixelsv2);
338 pixelssum1 = vec_add(pixelssum1, vctwo);
340 for (i = 0; i < h ; i++) {
341 int rightside = ((unsigned long)block & 0x0000000F);
342 blockv = vec_ld(0, block);
344 pixelsv1 = unaligned_load(line_size, pixels);
345 pixelsv2 = unaligned_load(line_size+1, pixels);
347 pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
348 pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
349 pixelssum2 = vec_add((vector unsigned short)pixelsv1,
350 (vector unsigned short)pixelsv2);
351 temp3 = vec_add(pixelssum1, pixelssum2);
352 temp3 = vec_sra(temp3, vctwo);
353 pixelssum1 = vec_add(pixelssum2, vctwo);
354 pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
357 blocktemp = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
359 blocktemp = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
362 blockv = vec_avg(blocktemp, blockv);
363 vec_st(blockv, 0, block);
369 #endif /* HAVE_ALTIVEC */
371 av_cold void ff_hpeldsp_init_ppc(HpelDSPContext *c, int flags)
374 if (!PPC_ALTIVEC(av_get_cpu_flags()))
377 c->avg_pixels_tab[0][0] = ff_avg_pixels16_altivec;
378 c->avg_pixels_tab[1][0] = avg_pixels8_altivec;
379 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_altivec;
381 c->put_pixels_tab[0][0] = ff_put_pixels16_altivec;
382 c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec;
383 c->put_pixels_tab[0][3] = put_pixels16_xy2_altivec;
385 c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_altivec;
386 c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec;
387 c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec;
388 #endif /* HAVE_ALTIVEC */