2 * Copyright (c) 2014 Seppo Tomperi <seppo.tomperi@vtt.fi>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/attributes.h"
22 #include "libavutil/arm/cpu.h"
23 #include "libavcodec/hevcdsp.h"
24 #include "hevcdsp_arm.h"
26 void ff_hevc_v_loop_filter_luma_neon(uint8_t *_pix, ptrdiff_t _stride, int _beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
27 void ff_hevc_h_loop_filter_luma_neon(uint8_t *_pix, ptrdiff_t _stride, int _beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
28 void ff_hevc_v_loop_filter_chroma_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
29 void ff_hevc_h_loop_filter_chroma_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
30 void ff_hevc_transform_4x4_neon_8(int16_t *coeffs, int col_limit);
31 void ff_hevc_transform_8x8_neon_8(int16_t *coeffs, int col_limit);
32 void ff_hevc_idct_4x4_dc_neon_8(int16_t *coeffs);
33 void ff_hevc_idct_8x8_dc_neon_8(int16_t *coeffs);
34 void ff_hevc_idct_16x16_dc_neon_8(int16_t *coeffs);
35 void ff_hevc_idct_32x32_dc_neon_8(int16_t *coeffs);
36 void ff_hevc_transform_luma_4x4_neon_8(int16_t *coeffs);
37 void ff_hevc_transform_add_4x4_neon_8(uint8_t *_dst, int16_t *coeffs,
39 void ff_hevc_transform_add_8x8_neon_8(uint8_t *_dst, int16_t *coeffs,
41 void ff_hevc_transform_add_16x16_neon_8(uint8_t *_dst, int16_t *coeffs,
43 void ff_hevc_transform_add_32x32_neon_8(uint8_t *_dst, int16_t *coeffs,
46 #define PUT_PIXELS(name) \
47 void name(int16_t *dst, uint8_t *src, \
48 ptrdiff_t srcstride, int height, \
49 intptr_t mx, intptr_t my, int width)
50 PUT_PIXELS(ff_hevc_put_pixels_w2_neon_8);
51 PUT_PIXELS(ff_hevc_put_pixels_w4_neon_8);
52 PUT_PIXELS(ff_hevc_put_pixels_w6_neon_8);
53 PUT_PIXELS(ff_hevc_put_pixels_w8_neon_8);
54 PUT_PIXELS(ff_hevc_put_pixels_w12_neon_8);
55 PUT_PIXELS(ff_hevc_put_pixels_w16_neon_8);
56 PUT_PIXELS(ff_hevc_put_pixels_w24_neon_8);
57 PUT_PIXELS(ff_hevc_put_pixels_w32_neon_8);
58 PUT_PIXELS(ff_hevc_put_pixels_w48_neon_8);
59 PUT_PIXELS(ff_hevc_put_pixels_w64_neon_8);
62 static void (*put_hevc_qpel_neon[4][4])(int16_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
63 int height, int width);
64 static void (*put_hevc_qpel_uw_neon[4][4])(uint8_t *dst, ptrdiff_t dststride, uint8_t *_src, ptrdiff_t _srcstride,
65 int width, int height, int16_t* src2, ptrdiff_t src2stride);
66 void ff_hevc_put_qpel_neon_wrapper(int16_t *dst, uint8_t *src, ptrdiff_t srcstride,
67 int height, intptr_t mx, intptr_t my, int width);
68 void ff_hevc_put_qpel_uni_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
69 int height, intptr_t mx, intptr_t my, int width);
70 void ff_hevc_put_qpel_bi_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
72 int height, intptr_t mx, intptr_t my, int width);
73 #define QPEL_FUNC(name) \
74 void name(int16_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride, \
75 int height, int width)
77 QPEL_FUNC(ff_hevc_put_qpel_v1_neon_8);
78 QPEL_FUNC(ff_hevc_put_qpel_v2_neon_8);
79 QPEL_FUNC(ff_hevc_put_qpel_v3_neon_8);
80 QPEL_FUNC(ff_hevc_put_qpel_h1_neon_8);
81 QPEL_FUNC(ff_hevc_put_qpel_h2_neon_8);
82 QPEL_FUNC(ff_hevc_put_qpel_h3_neon_8);
83 QPEL_FUNC(ff_hevc_put_qpel_h1v1_neon_8);
84 QPEL_FUNC(ff_hevc_put_qpel_h1v2_neon_8);
85 QPEL_FUNC(ff_hevc_put_qpel_h1v3_neon_8);
86 QPEL_FUNC(ff_hevc_put_qpel_h2v1_neon_8);
87 QPEL_FUNC(ff_hevc_put_qpel_h2v2_neon_8);
88 QPEL_FUNC(ff_hevc_put_qpel_h2v3_neon_8);
89 QPEL_FUNC(ff_hevc_put_qpel_h3v1_neon_8);
90 QPEL_FUNC(ff_hevc_put_qpel_h3v2_neon_8);
91 QPEL_FUNC(ff_hevc_put_qpel_h3v3_neon_8);
94 #define QPEL_FUNC_UW_PIX(name) \
95 void name(uint8_t *dst, ptrdiff_t dststride, uint8_t *_src, ptrdiff_t _srcstride, \
96 int height, intptr_t mx, intptr_t my, int width);
97 QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w4_neon_8);
98 QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w8_neon_8);
99 QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w16_neon_8);
100 QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w24_neon_8);
101 QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w32_neon_8);
102 QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w48_neon_8);
103 QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w64_neon_8);
104 #undef QPEL_FUNC_UW_PIX
106 #define QPEL_FUNC_UW(name) \
107 void name(uint8_t *dst, ptrdiff_t dststride, uint8_t *_src, ptrdiff_t _srcstride, \
108 int width, int height, int16_t* src2, ptrdiff_t src2stride);
109 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_pixels_neon_8);
110 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_v1_neon_8);
111 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_v2_neon_8);
112 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_v3_neon_8);
113 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1_neon_8);
114 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2_neon_8);
115 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3_neon_8);
116 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1v1_neon_8);
117 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1v2_neon_8);
118 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1v3_neon_8);
119 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2v1_neon_8);
120 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2v2_neon_8);
121 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2v3_neon_8);
122 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3v1_neon_8);
123 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3v2_neon_8);
124 QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3v3_neon_8);
127 void ff_hevc_put_qpel_neon_wrapper(int16_t *dst, uint8_t *src, ptrdiff_t srcstride,
128 int height, intptr_t mx, intptr_t my, int width) {
130 put_hevc_qpel_neon[my][mx](dst, MAX_PB_SIZE, src, srcstride, height, width);
133 void ff_hevc_put_qpel_uni_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
134 int height, intptr_t mx, intptr_t my, int width) {
136 put_hevc_qpel_uw_neon[my][mx](dst, dststride, src, srcstride, width, height, NULL, 0);
139 void ff_hevc_put_qpel_bi_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
141 int height, intptr_t mx, intptr_t my, int width) {
142 put_hevc_qpel_uw_neon[my][mx](dst, dststride, src, srcstride, width, height, src2, MAX_PB_SIZE);
145 av_cold void ff_hevcdsp_init_neon(HEVCDSPContext *c, const int bit_depth)
147 if (bit_depth == 8) {
149 c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_neon;
150 c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_neon;
151 c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_neon;
152 c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_neon;
153 c->idct[0] = ff_hevc_transform_4x4_neon_8;
154 c->idct[1] = ff_hevc_transform_8x8_neon_8;
155 c->idct_dc[0] = ff_hevc_idct_4x4_dc_neon_8;
156 c->idct_dc[1] = ff_hevc_idct_8x8_dc_neon_8;
157 c->idct_dc[2] = ff_hevc_idct_16x16_dc_neon_8;
158 c->idct_dc[3] = ff_hevc_idct_32x32_dc_neon_8;
159 c->transform_add[0] = ff_hevc_transform_add_4x4_neon_8;
160 c->transform_add[1] = ff_hevc_transform_add_8x8_neon_8;
161 c->transform_add[2] = ff_hevc_transform_add_16x16_neon_8;
162 c->transform_add[3] = ff_hevc_transform_add_32x32_neon_8;
163 c->idct_4x4_luma = ff_hevc_transform_luma_4x4_neon_8;
164 put_hevc_qpel_neon[1][0] = ff_hevc_put_qpel_v1_neon_8;
165 put_hevc_qpel_neon[2][0] = ff_hevc_put_qpel_v2_neon_8;
166 put_hevc_qpel_neon[3][0] = ff_hevc_put_qpel_v3_neon_8;
167 put_hevc_qpel_neon[0][1] = ff_hevc_put_qpel_h1_neon_8;
168 put_hevc_qpel_neon[0][2] = ff_hevc_put_qpel_h2_neon_8;
169 put_hevc_qpel_neon[0][3] = ff_hevc_put_qpel_h3_neon_8;
170 put_hevc_qpel_neon[1][1] = ff_hevc_put_qpel_h1v1_neon_8;
171 put_hevc_qpel_neon[1][2] = ff_hevc_put_qpel_h2v1_neon_8;
172 put_hevc_qpel_neon[1][3] = ff_hevc_put_qpel_h3v1_neon_8;
173 put_hevc_qpel_neon[2][1] = ff_hevc_put_qpel_h1v2_neon_8;
174 put_hevc_qpel_neon[2][2] = ff_hevc_put_qpel_h2v2_neon_8;
175 put_hevc_qpel_neon[2][3] = ff_hevc_put_qpel_h3v2_neon_8;
176 put_hevc_qpel_neon[3][1] = ff_hevc_put_qpel_h1v3_neon_8;
177 put_hevc_qpel_neon[3][2] = ff_hevc_put_qpel_h2v3_neon_8;
178 put_hevc_qpel_neon[3][3] = ff_hevc_put_qpel_h3v3_neon_8;
179 put_hevc_qpel_uw_neon[1][0] = ff_hevc_put_qpel_uw_v1_neon_8;
180 put_hevc_qpel_uw_neon[2][0] = ff_hevc_put_qpel_uw_v2_neon_8;
181 put_hevc_qpel_uw_neon[3][0] = ff_hevc_put_qpel_uw_v3_neon_8;
182 put_hevc_qpel_uw_neon[0][1] = ff_hevc_put_qpel_uw_h1_neon_8;
183 put_hevc_qpel_uw_neon[0][2] = ff_hevc_put_qpel_uw_h2_neon_8;
184 put_hevc_qpel_uw_neon[0][3] = ff_hevc_put_qpel_uw_h3_neon_8;
185 put_hevc_qpel_uw_neon[1][1] = ff_hevc_put_qpel_uw_h1v1_neon_8;
186 put_hevc_qpel_uw_neon[1][2] = ff_hevc_put_qpel_uw_h2v1_neon_8;
187 put_hevc_qpel_uw_neon[1][3] = ff_hevc_put_qpel_uw_h3v1_neon_8;
188 put_hevc_qpel_uw_neon[2][1] = ff_hevc_put_qpel_uw_h1v2_neon_8;
189 put_hevc_qpel_uw_neon[2][2] = ff_hevc_put_qpel_uw_h2v2_neon_8;
190 put_hevc_qpel_uw_neon[2][3] = ff_hevc_put_qpel_uw_h3v2_neon_8;
191 put_hevc_qpel_uw_neon[3][1] = ff_hevc_put_qpel_uw_h1v3_neon_8;
192 put_hevc_qpel_uw_neon[3][2] = ff_hevc_put_qpel_uw_h2v3_neon_8;
193 put_hevc_qpel_uw_neon[3][3] = ff_hevc_put_qpel_uw_h3v3_neon_8;
194 for (x = 0; x < 10; x++) {
195 c->put_hevc_qpel[x][1][0] = ff_hevc_put_qpel_neon_wrapper;
196 c->put_hevc_qpel[x][0][1] = ff_hevc_put_qpel_neon_wrapper;
197 c->put_hevc_qpel[x][1][1] = ff_hevc_put_qpel_neon_wrapper;
198 c->put_hevc_qpel_uni[x][1][0] = ff_hevc_put_qpel_uni_neon_wrapper;
199 c->put_hevc_qpel_uni[x][0][1] = ff_hevc_put_qpel_uni_neon_wrapper;
200 c->put_hevc_qpel_uni[x][1][1] = ff_hevc_put_qpel_uni_neon_wrapper;
201 c->put_hevc_qpel_bi[x][1][0] = ff_hevc_put_qpel_bi_neon_wrapper;
202 c->put_hevc_qpel_bi[x][0][1] = ff_hevc_put_qpel_bi_neon_wrapper;
203 c->put_hevc_qpel_bi[x][1][1] = ff_hevc_put_qpel_bi_neon_wrapper;
205 c->put_hevc_qpel[0][0][0] = ff_hevc_put_pixels_w2_neon_8;
206 c->put_hevc_qpel[1][0][0] = ff_hevc_put_pixels_w4_neon_8;
207 c->put_hevc_qpel[2][0][0] = ff_hevc_put_pixels_w6_neon_8;
208 c->put_hevc_qpel[3][0][0] = ff_hevc_put_pixels_w8_neon_8;
209 c->put_hevc_qpel[4][0][0] = ff_hevc_put_pixels_w12_neon_8;
210 c->put_hevc_qpel[5][0][0] = ff_hevc_put_pixels_w16_neon_8;
211 c->put_hevc_qpel[6][0][0] = ff_hevc_put_pixels_w24_neon_8;
212 c->put_hevc_qpel[7][0][0] = ff_hevc_put_pixels_w32_neon_8;
213 c->put_hevc_qpel[8][0][0] = ff_hevc_put_pixels_w48_neon_8;
214 c->put_hevc_qpel[9][0][0] = ff_hevc_put_pixels_w64_neon_8;
216 c->put_hevc_qpel_uni[1][0][0] = ff_hevc_put_qpel_uw_pixels_w4_neon_8;
217 c->put_hevc_qpel_uni[3][0][0] = ff_hevc_put_qpel_uw_pixels_w8_neon_8;
218 c->put_hevc_qpel_uni[5][0][0] = ff_hevc_put_qpel_uw_pixels_w16_neon_8;
219 c->put_hevc_qpel_uni[6][0][0] = ff_hevc_put_qpel_uw_pixels_w24_neon_8;
220 c->put_hevc_qpel_uni[7][0][0] = ff_hevc_put_qpel_uw_pixels_w32_neon_8;
221 c->put_hevc_qpel_uni[8][0][0] = ff_hevc_put_qpel_uw_pixels_w48_neon_8;
222 c->put_hevc_qpel_uni[9][0][0] = ff_hevc_put_qpel_uw_pixels_w64_neon_8;