2 * Alpha optimized DSP utils
3 * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
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
22 #include "libavutil/attributes.h"
23 #include "libavcodec/me_cmp.h"
26 int pix_abs16x16_mvi_asm(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h);
28 static inline uint64_t avg2(uint64_t a, uint64_t b)
30 return (a | b) - (((a ^ b) & BYTE_VEC(0xfe)) >> 1);
33 static inline uint64_t avg4(uint64_t l1, uint64_t l2, uint64_t l3, uint64_t l4)
35 uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
36 + ((l2 & ~BYTE_VEC(0x03)) >> 2)
37 + ((l3 & ~BYTE_VEC(0x03)) >> 2)
38 + ((l4 & ~BYTE_VEC(0x03)) >> 2);
39 uint64_t r2 = (( (l1 & BYTE_VEC(0x03))
40 + (l2 & BYTE_VEC(0x03))
41 + (l3 & BYTE_VEC(0x03))
42 + (l4 & BYTE_VEC(0x03))
43 + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03);
47 static int pix_abs8x8_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
51 if ((size_t) pix2 & 0x7) {
52 /* works only when pix2 is actually unaligned */
53 do { /* do 8 pixel a time */
58 result += perr(p1, p2);
69 result += perr(p1, p2);
79 #if 0 /* now done in assembly */
80 int pix_abs16x16_mvi(uint8_t *pix1, uint8_t *pix2, int line_size)
85 if ((size_t) pix2 & 0x7) {
86 /* works only when pix2 is actually unaligned */
87 do { /* do 16 pixel a time */
88 uint64_t p1_l, p1_r, p2_l, p2_r;
94 p2_l = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
95 p2_r = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);
99 result += perr(p1_l, p2_l)
104 uint64_t p1_l, p1_r, p2_l, p2_r;
107 p1_r = ldq(pix1 + 8);
109 p2_r = ldq(pix2 + 8);
113 result += perr(p1_l, p2_l)
122 static int pix_abs16x16_x2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
125 uint64_t disalign = (size_t) pix2 & 0x7;
130 uint64_t p1_l, p1_r, p2_l, p2_r;
134 p1_r = ldq(pix1 + 8);
137 p2_l = avg2(l, (l >> 8) | ((uint64_t) r << 56));
138 p2_r = avg2(r, (r >> 8) | ((uint64_t) pix2[16] << 56));
142 result += perr(p1_l, p2_l)
147 /* |.......l|lllllllr|rrrrrrr*|
148 This case is special because disalign1 would be 8, which
149 gets treated as 0 by extqh. At least it is a bit faster
152 uint64_t p1_l, p1_r, p2_l, p2_r;
156 p1_r = ldq(pix1 + 8);
159 r = ldq_u(pix2 + 16);
160 p2_l = avg2(extql(l, disalign) | extqh(m, disalign), m);
161 p2_r = avg2(extql(m, disalign) | extqh(r, disalign), r);
165 result += perr(p1_l, p2_l)
171 uint64_t disalign1 = disalign + 1;
172 uint64_t p1_l, p1_r, p2_l, p2_r;
176 p1_r = ldq(pix1 + 8);
179 r = ldq_u(pix2 + 16);
180 p2_l = avg2(extql(l, disalign) | extqh(m, disalign),
181 extql(l, disalign1) | extqh(m, disalign1));
182 p2_r = avg2(extql(m, disalign) | extqh(r, disalign),
183 extql(m, disalign1) | extqh(r, disalign1));
187 result += perr(p1_l, p2_l)
195 static int pix_abs16x16_y2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
199 if ((size_t) pix2 & 0x7) {
200 uint64_t t, p2_l, p2_r;
202 p2_l = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
203 p2_r = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);
206 uint64_t p1_l, p1_r, np2_l, np2_r;
210 p1_r = ldq(pix1 + 8);
213 np2_l = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
214 np2_r = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);
216 result += perr(p1_l, avg2(p2_l, np2_l))
217 + perr(p1_r, avg2(p2_r, np2_r));
227 p2_r = ldq(pix2 + 8);
229 uint64_t p1_l, p1_r, np2_l, np2_r;
232 p1_r = ldq(pix1 + 8);
235 np2_r = ldq(pix2 + 8);
237 result += perr(p1_l, avg2(p2_l, np2_l))
238 + perr(p1_r, avg2(p2_r, np2_r));
248 static int pix_abs16x16_xy2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
253 uint64_t p2_l, p2_r, p2_x;
256 p1_r = ldq(pix1 + 8);
258 if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
260 p2_r = uldq(pix2 + 8);
261 p2_x = (uint64_t) pix2[16] << 56;
264 p2_r = ldq(pix2 + 8);
265 p2_x = ldq(pix2 + 16) << 56;
269 uint64_t np1_l, np1_r;
270 uint64_t np2_l, np2_r, np2_x;
276 np1_r = ldq(pix1 + 8);
278 if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
280 np2_r = uldq(pix2 + 8);
281 np2_x = (uint64_t) pix2[16] << 56;
284 np2_r = ldq(pix2 + 8);
285 np2_x = ldq(pix2 + 16) << 56;
289 avg4( p2_l, ( p2_l >> 8) | ((uint64_t) p2_r << 56),
290 np2_l, (np2_l >> 8) | ((uint64_t) np2_r << 56)))
292 avg4( p2_r, ( p2_r >> 8) | ((uint64_t) p2_x),
293 np2_r, (np2_r >> 8) | ((uint64_t) np2_x)));
305 av_cold void ff_me_cmp_init_alpha(MECmpContext *c, AVCodecContext *avctx)
307 /* amask clears all bits that correspond to present features. */
308 if (amask(AMASK_MVI) == 0) {
309 c->sad[0] = pix_abs16x16_mvi_asm;
310 c->sad[1] = pix_abs8x8_mvi;
311 c->pix_abs[0][0] = pix_abs16x16_mvi_asm;
312 c->pix_abs[1][0] = pix_abs8x8_mvi;
313 c->pix_abs[0][1] = pix_abs16x16_x2_mvi;
314 c->pix_abs[0][2] = pix_abs16x16_y2_mvi;
315 c->pix_abs[0][3] = pix_abs16x16_xy2_mvi;