| 1 | /* Copyright (C) 2010-2020 The RetroArch team |
| 2 | * |
| 3 | * --------------------------------------------------------------------------------------- |
| 4 | * The following license statement only applies to this file (scaler_filter.c). |
| 5 | * --------------------------------------------------------------------------------------- |
| 6 | * |
| 7 | * Permission is hereby granted, free of charge, |
| 8 | * to any person obtaining a copy of this software and associated documentation files (the "Software"), |
| 9 | * to deal in the Software without restriction, including without limitation the rights to |
| 10 | * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, |
| 11 | * and to permit persons to whom the Software is furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. |
| 14 | * |
| 15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, |
| 16 | * INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. |
| 18 | * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, |
| 19 | * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 20 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. |
| 21 | */ |
| 22 | |
| 23 | #include <stdio.h> |
| 24 | #include <string.h> |
| 25 | |
| 26 | #include <gfx/scaler/filter.h> |
| 27 | #include <gfx/scaler/scaler_int.h> |
| 28 | #include <retro_inline.h> |
| 29 | #include <filters.h> |
| 30 | #include <retro_math.h> |
| 31 | |
| 32 | #define FILTER_UNITY (1 << 14) |
| 33 | |
| 34 | static INLINE void gen_filter_point_sub(struct scaler_filter *filter, |
| 35 | int len, int pos, int step) |
| 36 | { |
| 37 | int i; |
| 38 | for (i = 0; i < len; i++, pos += step) |
| 39 | { |
| 40 | filter->filter_pos[i] = pos >> 16; |
| 41 | filter->filter[i] = FILTER_UNITY; |
| 42 | } |
| 43 | } |
| 44 | |
| 45 | static INLINE void gen_filter_bilinear_sub(struct scaler_filter *filter, |
| 46 | int len, int pos, int step) |
| 47 | { |
| 48 | int i; |
| 49 | for (i = 0; i < len; i++, pos += step) |
| 50 | { |
| 51 | filter->filter_pos[i] = pos >> 16; |
| 52 | filter->filter[i * 2 + 1] = (pos & 0xffff) >> 2; |
| 53 | filter->filter[i * 2 + 0] = FILTER_UNITY - filter->filter[i * 2 + 1]; |
| 54 | } |
| 55 | } |
| 56 | |
| 57 | static INLINE void gen_filter_sinc_sub(struct scaler_filter *filter, |
| 58 | int len, int pos, int step, double phase_mul) |
| 59 | { |
| 60 | int i, j; |
| 61 | const int sinc_size = filter->filter_len; |
| 62 | |
| 63 | for (i = 0; i < len; i++, pos += step) |
| 64 | { |
| 65 | filter->filter_pos[i] = pos >> 16; |
| 66 | |
| 67 | for (j = 0; j < sinc_size; j++) |
| 68 | { |
| 69 | double sinc_phase = M_PI * ((double)((sinc_size << 15) + (pos & 0xffff)) / 0x10000 - j); |
| 70 | double lanczos_phase = sinc_phase / ((sinc_size >> 1)); |
| 71 | int16_t sinc_val = FILTER_UNITY * sinc(sinc_phase * phase_mul) * sinc(lanczos_phase) * phase_mul; |
| 72 | |
| 73 | filter->filter[i * sinc_size + j] = sinc_val; |
| 74 | } |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | static bool validate_filter(struct scaler_ctx *ctx) |
| 79 | { |
| 80 | int i; |
| 81 | int max_h_pos; |
| 82 | int max_w_pos = ctx->in_width - ctx->horiz.filter_len; |
| 83 | |
| 84 | for (i = 0; i < ctx->out_width; i++) |
| 85 | { |
| 86 | if (ctx->horiz.filter_pos[i] > max_w_pos || ctx->horiz.filter_pos[i] < 0) |
| 87 | { |
| 88 | #ifndef NDEBUG |
| 89 | fprintf(stderr, "Out X = %d => In X = %d\n", i, ctx->horiz.filter_pos[i]); |
| 90 | #endif |
| 91 | return false; |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | max_h_pos = ctx->in_height - ctx->vert.filter_len; |
| 96 | |
| 97 | for (i = 0; i < ctx->out_height; i++) |
| 98 | { |
| 99 | if (ctx->vert.filter_pos[i] > max_h_pos || ctx->vert.filter_pos[i] < 0) |
| 100 | { |
| 101 | #ifndef NDEBUG |
| 102 | fprintf(stderr, "Out Y = %d => In Y = %d\n", i, ctx->vert.filter_pos[i]); |
| 103 | #endif |
| 104 | return false; |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | return true; |
| 109 | } |
| 110 | |
| 111 | static void fixup_filter_sub(struct scaler_filter *filter, |
| 112 | int out_len, int in_len) |
| 113 | { |
| 114 | int i; |
| 115 | int max_pos = in_len - filter->filter_len; |
| 116 | |
| 117 | for (i = 0; i < out_len; i++) |
| 118 | { |
| 119 | int postsample = filter->filter_pos[i] - max_pos; |
| 120 | int presample = -filter->filter_pos[i]; |
| 121 | |
| 122 | if (postsample > 0) |
| 123 | { |
| 124 | int16_t *base_filter = NULL; |
| 125 | |
| 126 | filter->filter_pos[i] -= postsample; |
| 127 | |
| 128 | base_filter = filter->filter + i * filter->filter_stride; |
| 129 | |
| 130 | if (postsample > (int)filter->filter_len) |
| 131 | memset(base_filter, 0, filter->filter_len * sizeof(int16_t)); |
| 132 | else |
| 133 | { |
| 134 | memmove(base_filter + postsample, base_filter, |
| 135 | (filter->filter_len - postsample) * sizeof(int16_t)); |
| 136 | memset(base_filter, 0, postsample * sizeof(int16_t)); |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | if (presample > 0) |
| 141 | { |
| 142 | int16_t *base_filter = NULL; |
| 143 | |
| 144 | filter->filter_pos[i] += presample; |
| 145 | base_filter = filter->filter + i * filter->filter_stride; |
| 146 | |
| 147 | if (presample > (int)filter->filter_len) |
| 148 | memset(base_filter, 0, filter->filter_len * sizeof(int16_t)); |
| 149 | else |
| 150 | { |
| 151 | memmove(base_filter, base_filter + presample, |
| 152 | (filter->filter_len - presample) * sizeof(int16_t)); |
| 153 | memset(base_filter + (filter->filter_len - presample), |
| 154 | 0, presample * sizeof(int16_t)); |
| 155 | } |
| 156 | } |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | bool scaler_gen_filter(struct scaler_ctx *ctx) |
| 161 | { |
| 162 | int x_pos, x_step, y_pos, y_step; |
| 163 | int sinc_size = 0; |
| 164 | |
| 165 | switch (ctx->scaler_type) |
| 166 | { |
| 167 | case SCALER_TYPE_POINT: |
| 168 | ctx->horiz.filter_len = 1; |
| 169 | ctx->horiz.filter_stride = 1; |
| 170 | ctx->vert.filter_len = 1; |
| 171 | ctx->vert.filter_stride = 1; |
| 172 | break; |
| 173 | case SCALER_TYPE_BILINEAR: |
| 174 | ctx->horiz.filter_len = 2; |
| 175 | ctx->horiz.filter_stride = 2; |
| 176 | ctx->vert.filter_len = 2; |
| 177 | ctx->vert.filter_stride = 2; |
| 178 | break; |
| 179 | case SCALER_TYPE_SINC: |
| 180 | sinc_size = 8 * ((ctx->in_width > ctx->out_width) |
| 181 | ? next_pow2(ctx->in_width / ctx->out_width) : 1); |
| 182 | ctx->horiz.filter_len = sinc_size; |
| 183 | ctx->horiz.filter_stride = sinc_size; |
| 184 | ctx->vert.filter_len = sinc_size; |
| 185 | ctx->vert.filter_stride = sinc_size; |
| 186 | break; |
| 187 | case SCALER_TYPE_UNKNOWN: |
| 188 | default: |
| 189 | return false; |
| 190 | } |
| 191 | |
| 192 | ctx->horiz.filter = (int16_t*)calloc(sizeof(int16_t), ctx->horiz.filter_stride * ctx->out_width); |
| 193 | ctx->horiz.filter_pos = (int*)calloc(sizeof(int), ctx->out_width); |
| 194 | |
| 195 | ctx->vert.filter = (int16_t*)calloc(sizeof(int16_t), ctx->vert.filter_stride * ctx->out_height); |
| 196 | ctx->vert.filter_pos = (int*)calloc(sizeof(int), ctx->out_height); |
| 197 | |
| 198 | if (!ctx->horiz.filter || !ctx->vert.filter) |
| 199 | return false; |
| 200 | |
| 201 | x_step = (1 << 16) * ctx->in_width / ctx->out_width; |
| 202 | y_step = (1 << 16) * ctx->in_height / ctx->out_height; |
| 203 | |
| 204 | switch (ctx->scaler_type) |
| 205 | { |
| 206 | case SCALER_TYPE_POINT: |
| 207 | x_pos = (1 << 15) * ctx->in_width / ctx->out_width - (1 << 15); |
| 208 | y_pos = (1 << 15) * ctx->in_height / ctx->out_height - (1 << 15); |
| 209 | |
| 210 | gen_filter_point_sub(&ctx->horiz, ctx->out_width, x_pos, x_step); |
| 211 | gen_filter_point_sub(&ctx->vert, ctx->out_height, y_pos, y_step); |
| 212 | |
| 213 | ctx->scaler_special = scaler_argb8888_point_special; |
| 214 | break; |
| 215 | |
| 216 | case SCALER_TYPE_BILINEAR: |
| 217 | x_pos = (1 << 15) * ctx->in_width / ctx->out_width - (1 << 15); |
| 218 | y_pos = (1 << 15) * ctx->in_height / ctx->out_height - (1 << 15); |
| 219 | |
| 220 | gen_filter_bilinear_sub(&ctx->horiz, ctx->out_width, x_pos, x_step); |
| 221 | gen_filter_bilinear_sub(&ctx->vert, ctx->out_height, y_pos, y_step); |
| 222 | break; |
| 223 | |
| 224 | case SCALER_TYPE_SINC: |
| 225 | /* Need to expand the filter when downsampling |
| 226 | * to get a proper low-pass effect. */ |
| 227 | |
| 228 | x_pos = (1 << 15) * ctx->in_width / ctx->out_width - (1 << 15) - (sinc_size << 15); |
| 229 | y_pos = (1 << 15) * ctx->in_height / ctx->out_height - (1 << 15) - (sinc_size << 15); |
| 230 | |
| 231 | gen_filter_sinc_sub(&ctx->horiz, ctx->out_width, x_pos, x_step, |
| 232 | ctx->in_width > ctx->out_width ? (double)ctx->out_width / ctx->in_width : 1.0); |
| 233 | gen_filter_sinc_sub(&ctx->vert, ctx->out_height, y_pos, y_step, |
| 234 | ctx->in_height > ctx->out_height ? (double)ctx->out_height / ctx->in_height : 1.0 |
| 235 | ); |
| 236 | break; |
| 237 | case SCALER_TYPE_UNKNOWN: |
| 238 | break; |
| 239 | } |
| 240 | |
| 241 | /* Makes sure that we never sample outside our rectangle. */ |
| 242 | fixup_filter_sub(&ctx->horiz, ctx->out_width, ctx->in_width); |
| 243 | fixup_filter_sub(&ctx->vert, ctx->out_height, ctx->in_height); |
| 244 | |
| 245 | return validate_filter(ctx); |
| 246 | } |