2 * ARM NEON optimised DSP functions
3 * Copyright (c) 2008 Mans Rullgard <mans@mansr.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
24 #include "libavutil/attributes.h"
25 #include "libavcodec/avcodec.h"
26 #include "dsputil_arm.h"
28 void ff_simple_idct_neon(int16_t *data);
29 void ff_simple_idct_put_neon(uint8_t *dest, int line_size, int16_t *data);
30 void ff_simple_idct_add_neon(uint8_t *dest, int line_size, int16_t *data);
32 void ff_clear_block_neon(int16_t *block);
33 void ff_clear_blocks_neon(int16_t *blocks);
35 void ff_put_pixels16_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
36 void ff_put_pixels16_x2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
37 void ff_put_pixels16_y2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
38 void ff_put_pixels16_xy2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
39 void ff_put_pixels8_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
40 void ff_put_pixels8_x2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
41 void ff_put_pixels8_y2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
42 void ff_put_pixels8_xy2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
43 void ff_put_pixels16_x2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
44 void ff_put_pixels16_y2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
45 void ff_put_pixels16_xy2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
46 void ff_put_pixels8_x2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
47 void ff_put_pixels8_y2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
48 void ff_put_pixels8_xy2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
50 void ff_avg_pixels16_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
51 void ff_avg_pixels16_x2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
52 void ff_avg_pixels16_y2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
53 void ff_avg_pixels16_xy2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
54 void ff_avg_pixels8_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
55 void ff_avg_pixels8_x2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
56 void ff_avg_pixels8_y2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
57 void ff_avg_pixels8_xy2_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
58 void ff_avg_pixels16_x2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
59 void ff_avg_pixels16_y2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
60 void ff_avg_pixels16_xy2_no_rnd_neon(uint8_t *, const uint8_t *, ptrdiff_t, int);
62 void ff_add_pixels_clamped_neon(const int16_t *, uint8_t *, int);
63 void ff_put_pixels_clamped_neon(const int16_t *, uint8_t *, int);
64 void ff_put_signed_pixels_clamped_neon(const int16_t *, uint8_t *, int);
66 void ff_vector_clipf_neon(float *dst, const float *src, float min, float max,
68 void ff_vector_clip_int32_neon(int32_t *dst, const int32_t *src, int32_t min,
69 int32_t max, unsigned int len);
71 int32_t ff_scalarproduct_int16_neon(const int16_t *v1, const int16_t *v2, int len);
72 int32_t ff_scalarproduct_and_madd_int16_neon(int16_t *v1, const int16_t *v2,
73 const int16_t *v3, int len, int mul);
75 void ff_apply_window_int16_neon(int16_t *dst, const int16_t *src,
76 const int16_t *window, unsigned n);
78 av_cold void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
80 const int high_bit_depth = avctx->bits_per_raw_sample > 8;
82 if (avctx->bits_per_raw_sample <= 8) {
83 if (avctx->idct_algo == FF_IDCT_AUTO ||
84 avctx->idct_algo == FF_IDCT_SIMPLENEON) {
85 c->idct_put = ff_simple_idct_put_neon;
86 c->idct_add = ff_simple_idct_add_neon;
87 c->idct = ff_simple_idct_neon;
88 c->idct_permutation_type = FF_PARTTRANS_IDCT_PERM;
92 if (!high_bit_depth) {
93 c->clear_block = ff_clear_block_neon;
94 c->clear_blocks = ff_clear_blocks_neon;
96 c->put_pixels_tab[0][0] = ff_put_pixels16_neon;
97 c->put_pixels_tab[0][1] = ff_put_pixels16_x2_neon;
98 c->put_pixels_tab[0][2] = ff_put_pixels16_y2_neon;
99 c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_neon;
100 c->put_pixels_tab[1][0] = ff_put_pixels8_neon;
101 c->put_pixels_tab[1][1] = ff_put_pixels8_x2_neon;
102 c->put_pixels_tab[1][2] = ff_put_pixels8_y2_neon;
103 c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_neon;
105 c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_neon;
106 c->put_no_rnd_pixels_tab[0][1] = ff_put_pixels16_x2_no_rnd_neon;
107 c->put_no_rnd_pixels_tab[0][2] = ff_put_pixels16_y2_no_rnd_neon;
108 c->put_no_rnd_pixels_tab[0][3] = ff_put_pixels16_xy2_no_rnd_neon;
109 c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_neon;
110 c->put_no_rnd_pixels_tab[1][1] = ff_put_pixels8_x2_no_rnd_neon;
111 c->put_no_rnd_pixels_tab[1][2] = ff_put_pixels8_y2_no_rnd_neon;
112 c->put_no_rnd_pixels_tab[1][3] = ff_put_pixels8_xy2_no_rnd_neon;
114 c->avg_pixels_tab[0][0] = ff_avg_pixels16_neon;
115 c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_neon;
116 c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_neon;
117 c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_neon;
118 c->avg_pixels_tab[1][0] = ff_avg_pixels8_neon;
119 c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_neon;
120 c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_neon;
121 c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_neon;
123 c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_neon;
124 c->avg_no_rnd_pixels_tab[1] = ff_avg_pixels16_x2_no_rnd_neon;
125 c->avg_no_rnd_pixels_tab[2] = ff_avg_pixels16_y2_no_rnd_neon;
126 c->avg_no_rnd_pixels_tab[3] = ff_avg_pixels16_xy2_no_rnd_neon;
129 c->add_pixels_clamped = ff_add_pixels_clamped_neon;
130 c->put_pixels_clamped = ff_put_pixels_clamped_neon;
131 c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_neon;
133 c->vector_clipf = ff_vector_clipf_neon;
134 c->vector_clip_int32 = ff_vector_clip_int32_neon;
136 c->scalarproduct_int16 = ff_scalarproduct_int16_neon;
137 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_neon;
139 c->apply_window_int16 = ff_apply_window_int16_neon;