1 /*
2 * Copyright © 2018, Niklas Haas
3 * Copyright © 2018, VideoLAN and dav1d authors
4 * Copyright © 2018, Two Orioles, LLC
5 * Copyright © 2021, Martin Storsjo
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright notice, this
12 * list of conditions and the following disclaimer.
13 *
14 * 2. Redistributions in binary form must reproduce the above copyright notice,
15 * this list of conditions and the following disclaimer in the documentation
16 * and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
20 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
21 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
22 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
23 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
24 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
25 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
27 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
30 #include "src/cpu.h"
31 #include "src/filmgrain.h"
32 #include "asm-offsets.h"
33
34 CHECK_OFFSET(Dav1dFilmGrainData, seed, FGD_SEED);
35 CHECK_OFFSET(Dav1dFilmGrainData, ar_coeff_lag, FGD_AR_COEFF_LAG);
36 CHECK_OFFSET(Dav1dFilmGrainData, ar_coeffs_y, FGD_AR_COEFFS_Y);
37 CHECK_OFFSET(Dav1dFilmGrainData, ar_coeffs_uv, FGD_AR_COEFFS_UV);
38 CHECK_OFFSET(Dav1dFilmGrainData, ar_coeff_shift, FGD_AR_COEFF_SHIFT);
39 CHECK_OFFSET(Dav1dFilmGrainData, grain_scale_shift, FGD_GRAIN_SCALE_SHIFT);
40
41 CHECK_OFFSET(Dav1dFilmGrainData, scaling_shift, FGD_SCALING_SHIFT);
42 CHECK_OFFSET(Dav1dFilmGrainData, uv_mult, FGD_UV_MULT);
43 CHECK_OFFSET(Dav1dFilmGrainData, uv_luma_mult, FGD_UV_LUMA_MULT);
44 CHECK_OFFSET(Dav1dFilmGrainData, uv_offset, FGD_UV_OFFSET);
45 CHECK_OFFSET(Dav1dFilmGrainData, clip_to_restricted_range, FGD_CLIP_TO_RESTRICTED_RANGE);
46
47 void BF(dav1d_generate_grain_y, neon)(entry buf[][GRAIN_WIDTH],
48 const Dav1dFilmGrainData *const data
49 HIGHBD_DECL_SUFFIX);
50
51 #define GEN_GRAIN_UV(suff) \
52 void BF(dav1d_generate_grain_uv_ ## suff, neon)(entry buf[][GRAIN_WIDTH], \
53 const entry buf_y[][GRAIN_WIDTH], \
54 const Dav1dFilmGrainData *const data, \
55 const intptr_t uv \
56 HIGHBD_DECL_SUFFIX)
57
58 GEN_GRAIN_UV(420);
59 GEN_GRAIN_UV(422);
60 GEN_GRAIN_UV(444);
61
62 // Use ptrdiff_t instead of int for the last few parameters, to get the
63 // same layout of parameters on the stack across platforms.
64 void BF(dav1d_fgy_32x32, neon)(pixel *const dst,
65 const pixel *const src,
66 const ptrdiff_t stride,
67 const uint8_t scaling[SCALING_SIZE],
68 const int scaling_shift,
69 const entry grain_lut[][GRAIN_WIDTH],
70 const int offsets[][2],
71 const int h, const ptrdiff_t clip,
72 const ptrdiff_t type
73 HIGHBD_DECL_SUFFIX);
74
fgy_32x32xn_neon(pixel * const dst_row,const pixel * const src_row,const ptrdiff_t stride,const Dav1dFilmGrainData * const data,const size_t pw,const uint8_t scaling[SCALING_SIZE],const entry grain_lut[][GRAIN_WIDTH],const int bh,const int row_num HIGHBD_DECL_SUFFIX)75 static void fgy_32x32xn_neon(pixel *const dst_row, const pixel *const src_row,
76 const ptrdiff_t stride,
77 const Dav1dFilmGrainData *const data, const size_t pw,
78 const uint8_t scaling[SCALING_SIZE],
79 const entry grain_lut[][GRAIN_WIDTH],
80 const int bh, const int row_num HIGHBD_DECL_SUFFIX)
81 {
82 const int rows = 1 + (data->overlap_flag && row_num > 0);
83
84 // seed[0] contains the current row, seed[1] contains the previous
85 unsigned seed[2];
86 for (int i = 0; i < rows; i++) {
87 seed[i] = data->seed;
88 seed[i] ^= (((row_num - i) * 37 + 178) & 0xFF) << 8;
89 seed[i] ^= (((row_num - i) * 173 + 105) & 0xFF);
90 }
91
92 int offsets[2 /* col offset */][2 /* row offset */];
93
94 // process this row in FG_BLOCK_SIZE^2 blocks
95 for (unsigned bx = 0; bx < pw; bx += FG_BLOCK_SIZE) {
96
97 if (data->overlap_flag && bx) {
98 // shift previous offsets left
99 for (int i = 0; i < rows; i++)
100 offsets[1][i] = offsets[0][i];
101 }
102
103 // update current offsets
104 for (int i = 0; i < rows; i++)
105 offsets[0][i] = get_random_number(8, &seed[i]);
106
107 int type = 0;
108 if (data->overlap_flag && row_num)
109 type |= 1; /* overlap y */
110 if (data->overlap_flag && bx)
111 type |= 2; /* overlap x */
112
113 BF(dav1d_fgy_32x32, neon)(dst_row + bx, src_row + bx, stride,
114 scaling, data->scaling_shift,
115 grain_lut, offsets, bh,
116 data->clip_to_restricted_range, type
117 HIGHBD_TAIL_SUFFIX);
118 }
119 }
120
121 // Use ptrdiff_t instead of int for the last few parameters, to get the
122 // parameters on the stack with the same layout across platforms.
123 #define FGUV(nm, sx, sy) \
124 void BF(dav1d_fguv_32x32_##nm, neon)(pixel *const dst, \
125 const pixel *const src, \
126 const ptrdiff_t stride, \
127 const uint8_t scaling[SCALING_SIZE], \
128 const Dav1dFilmGrainData *const data, \
129 const entry grain_lut[][GRAIN_WIDTH], \
130 const pixel *const luma_row, \
131 const ptrdiff_t luma_stride, \
132 const int offsets[][2], \
133 const ptrdiff_t h, const ptrdiff_t uv, \
134 const ptrdiff_t is_id, \
135 const ptrdiff_t type \
136 HIGHBD_DECL_SUFFIX); \
137 static void \
138 fguv_32x32xn_##nm##_neon(pixel *const dst_row, const pixel *const src_row, \
139 const ptrdiff_t stride, const Dav1dFilmGrainData *const data, \
140 const size_t pw, const uint8_t scaling[SCALING_SIZE], \
141 const entry grain_lut[][GRAIN_WIDTH], const int bh, \
142 const int row_num, const pixel *const luma_row, \
143 const ptrdiff_t luma_stride, const int uv, const int is_id \
144 HIGHBD_DECL_SUFFIX) \
145 { \
146 const int rows = 1 + (data->overlap_flag && row_num > 0); \
147 \
148 /* seed[0] contains the current row, seed[1] contains the previous */ \
149 unsigned seed[2]; \
150 for (int i = 0; i < rows; i++) { \
151 seed[i] = data->seed; \
152 seed[i] ^= (((row_num - i) * 37 + 178) & 0xFF) << 8; \
153 seed[i] ^= (((row_num - i) * 173 + 105) & 0xFF); \
154 } \
155 \
156 int offsets[2 /* col offset */][2 /* row offset */]; \
157 \
158 /* process this row in FG_BLOCK_SIZE^2 blocks (subsampled) */ \
159 for (unsigned bx = 0; bx < pw; bx += FG_BLOCK_SIZE >> sx) { \
160 if (data->overlap_flag && bx) { \
161 /* shift previous offsets left */ \
162 for (int i = 0; i < rows; i++) \
163 offsets[1][i] = offsets[0][i]; \
164 } \
165 \
166 /* update current offsets */ \
167 for (int i = 0; i < rows; i++) \
168 offsets[0][i] = get_random_number(8, &seed[i]); \
169 \
170 int type = 0; \
171 if (data->overlap_flag && row_num) \
172 type |= 1; /* overlap y */ \
173 if (data->overlap_flag && bx) \
174 type |= 2; /* overlap x */ \
175 if (data->chroma_scaling_from_luma) \
176 type |= 4; \
177 \
178 BF(dav1d_fguv_32x32_##nm, neon)(dst_row + bx, src_row + bx, stride, \
179 scaling, data, grain_lut, \
180 luma_row + (bx << sx), luma_stride, \
181 offsets, bh, uv, is_id, type \
182 HIGHBD_TAIL_SUFFIX); \
183 } \
184 }
185
186 FGUV(420, 1, 1);
187 FGUV(422, 1, 0);
188 FGUV(444, 0, 0);
189
film_grain_dsp_init_arm(Dav1dFilmGrainDSPContext * const c)190 static ALWAYS_INLINE void film_grain_dsp_init_arm(Dav1dFilmGrainDSPContext *const c) {
191 const unsigned flags = dav1d_get_cpu_flags();
192
193 if (!(flags & DAV1D_ARM_CPU_FLAG_NEON)) return;
194
195 c->generate_grain_y = BF(dav1d_generate_grain_y, neon);
196 c->generate_grain_uv[DAV1D_PIXEL_LAYOUT_I420 - 1] = BF(dav1d_generate_grain_uv_420, neon);
197 c->generate_grain_uv[DAV1D_PIXEL_LAYOUT_I422 - 1] = BF(dav1d_generate_grain_uv_422, neon);
198 c->generate_grain_uv[DAV1D_PIXEL_LAYOUT_I444 - 1] = BF(dav1d_generate_grain_uv_444, neon);
199
200 c->fgy_32x32xn = fgy_32x32xn_neon;
201 c->fguv_32x32xn[DAV1D_PIXEL_LAYOUT_I420 - 1] = fguv_32x32xn_420_neon;
202 c->fguv_32x32xn[DAV1D_PIXEL_LAYOUT_I422 - 1] = fguv_32x32xn_422_neon;
203 c->fguv_32x32xn[DAV1D_PIXEL_LAYOUT_I444 - 1] = fguv_32x32xn_444_neon;
204 }
205