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(struct MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
27 ptrdiff_t line_size, int h);
29 static inline uint64_t avg2(uint64_t a, uint64_t b)
31 return (a | b) - (((a ^ b) & BYTE_VEC(0xfe)) >> 1);
34 static inline uint64_t avg4(uint64_t l1, uint64_t l2, uint64_t l3, uint64_t l4)
36 uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
37 + ((l2 & ~BYTE_VEC(0x03)) >> 2)
38 + ((l3 & ~BYTE_VEC(0x03)) >> 2)
39 + ((l4 & ~BYTE_VEC(0x03)) >> 2);
40 uint64_t r2 = (( (l1 & BYTE_VEC(0x03))
41 + (l2 & BYTE_VEC(0x03))
42 + (l3 & BYTE_VEC(0x03))
43 + (l4 & BYTE_VEC(0x03))
44 + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03);
48 static int pix_abs8x8_mvi(struct MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
49 ptrdiff_t line_size, int h)
53 if ((size_t) pix2 & 0x7) {
54 /* works only when pix2 is actually unaligned */
55 do { /* do 8 pixel a time */
60 result += perr(p1, p2);
71 result += perr(p1, p2);
81 static int pix_abs16x16_x2_mvi(struct MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
82 ptrdiff_t line_size, int h)
85 uint64_t disalign = (size_t) pix2 & 0x7;
90 uint64_t p1_l, p1_r, p2_l, p2_r;
97 p2_l = avg2(l, (l >> 8) | ((uint64_t) r << 56));
98 p2_r = avg2(r, (r >> 8) | ((uint64_t) pix2[16] << 56));
102 result += perr(p1_l, p2_l)
107 /* |.......l|lllllllr|rrrrrrr*|
108 This case is special because disalign1 would be 8, which
109 gets treated as 0 by extqh. At least it is a bit faster
112 uint64_t p1_l, p1_r, p2_l, p2_r;
116 p1_r = ldq(pix1 + 8);
119 r = ldq_u(pix2 + 16);
120 p2_l = avg2(extql(l, disalign) | extqh(m, disalign), m);
121 p2_r = avg2(extql(m, disalign) | extqh(r, disalign), r);
125 result += perr(p1_l, p2_l)
131 uint64_t disalign1 = disalign + 1;
132 uint64_t p1_l, p1_r, p2_l, p2_r;
136 p1_r = ldq(pix1 + 8);
139 r = ldq_u(pix2 + 16);
140 p2_l = avg2(extql(l, disalign) | extqh(m, disalign),
141 extql(l, disalign1) | extqh(m, disalign1));
142 p2_r = avg2(extql(m, disalign) | extqh(r, disalign),
143 extql(m, disalign1) | extqh(r, disalign1));
147 result += perr(p1_l, p2_l)
155 static int pix_abs16x16_y2_mvi(struct MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
156 ptrdiff_t line_size, int h)
160 if ((size_t) pix2 & 0x7) {
161 uint64_t t, p2_l, p2_r;
163 p2_l = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
164 p2_r = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);
167 uint64_t p1_l, p1_r, np2_l, np2_r;
171 p1_r = ldq(pix1 + 8);
174 np2_l = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
175 np2_r = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);
177 result += perr(p1_l, avg2(p2_l, np2_l))
178 + perr(p1_r, avg2(p2_r, np2_r));
188 p2_r = ldq(pix2 + 8);
190 uint64_t p1_l, p1_r, np2_l, np2_r;
193 p1_r = ldq(pix1 + 8);
196 np2_r = ldq(pix2 + 8);
198 result += perr(p1_l, avg2(p2_l, np2_l))
199 + perr(p1_r, avg2(p2_r, np2_r));
209 static int pix_abs16x16_xy2_mvi(struct MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
210 ptrdiff_t line_size, int h)
215 uint64_t p2_l, p2_r, p2_x;
218 p1_r = ldq(pix1 + 8);
220 if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
222 p2_r = uldq(pix2 + 8);
223 p2_x = (uint64_t) pix2[16] << 56;
226 p2_r = ldq(pix2 + 8);
227 p2_x = ldq(pix2 + 16) << 56;
231 uint64_t np1_l, np1_r;
232 uint64_t np2_l, np2_r, np2_x;
238 np1_r = ldq(pix1 + 8);
240 if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
242 np2_r = uldq(pix2 + 8);
243 np2_x = (uint64_t) pix2[16] << 56;
246 np2_r = ldq(pix2 + 8);
247 np2_x = ldq(pix2 + 16) << 56;
251 avg4( p2_l, ( p2_l >> 8) | ((uint64_t) p2_r << 56),
252 np2_l, (np2_l >> 8) | ((uint64_t) np2_r << 56)))
254 avg4( p2_r, ( p2_r >> 8) | ((uint64_t) p2_x),
255 np2_r, (np2_r >> 8) | ((uint64_t) np2_x)));
267 av_cold void ff_me_cmp_init_alpha(MECmpContext *c, AVCodecContext *avctx)
269 /* amask clears all bits that correspond to present features. */
270 if (amask(AMASK_MVI) == 0) {
271 c->sad[0] = pix_abs16x16_mvi_asm;
272 c->sad[1] = pix_abs8x8_mvi;
273 c->pix_abs[0][0] = pix_abs16x16_mvi_asm;
274 c->pix_abs[1][0] = pix_abs8x8_mvi;
275 c->pix_abs[0][1] = pix_abs16x16_x2_mvi;
276 c->pix_abs[0][2] = pix_abs16x16_y2_mvi;
277 c->pix_abs[0][3] = pix_abs16x16_xy2_mvi;