1 /************************************************************************** 2 * 3 * Copyright 2012 Christian Knig. 4 * All Rights Reserved. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the 8 * "Software"), to deal in the Software without restriction, including 9 * without limitation the rights to use, copy, modify, merge, publish, 10 * distribute, sub license, and/or sell copies of the Software, and to 11 * permit persons to whom the Software is furnished to do so, subject to 12 * the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the 15 * next paragraph) shall be included in all copies or substantial portions 16 * of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 * 26 **************************************************************************/ 27 28 #include "pipe/p_context.h" 29 30 #include "tgsi/tgsi_ureg.h" 31 32 #include "util/u_draw.h" 33 #include "util/u_memory.h" 34 #include "util/u_math.h" 35 36 #include "vl_types.h" 37 #include "vl_vertex_buffers.h" 38 #include "vl_median_filter.h" 39 40 enum VS_OUTPUT 41 { 42 VS_O_VPOS = 0, 43 VS_O_VTEX = 0 44 }; 45 46 static void * 47 create_vert_shader(struct vl_median_filter *filter) 48 { 49 struct ureg_program *shader; 50 struct ureg_src i_vpos; 51 struct ureg_dst o_vpos, o_vtex; 52 53 shader = ureg_create(TGSI_PROCESSOR_VERTEX); 54 if (!shader) 55 return NULL; 56 57 i_vpos = ureg_DECL_vs_input(shader, 0); 58 o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS); 59 o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX); 60 61 ureg_MOV(shader, o_vpos, i_vpos); 62 ureg_MOV(shader, o_vtex, i_vpos); 63 64 ureg_END(shader); 65 66 return ureg_create_shader_and_destroy(shader, filter->pipe); 67 } 68 69 static inline bool 70 is_vec_zero(struct vertex2f v) 71 { 72 return v.x == 0.0f && v.y == 0.0f; 73 } 74 75 static void * 76 create_frag_shader(struct vl_median_filter *filter, 77 struct vertex2f *offsets, 78 unsigned num_offsets) 79 { 80 struct pipe_screen *screen = filter->pipe->screen; 81 struct ureg_program *shader; 82 struct ureg_src i_vtex; 83 struct ureg_src sampler; 84 struct ureg_dst *t_array = MALLOC(sizeof(struct ureg_dst) * num_offsets); 85 struct ureg_dst o_fragment; 86 const unsigned median = num_offsets >> 1; 87 int i, j; 88 89 assert(num_offsets & 1); /* we need an odd number of offsets */ 90 if (!(num_offsets & 1)) { /* yeah, we REALLY need an odd number of offsets!!! */ 91 FREE(t_array); 92 return NULL; 93 } 94 95 if (num_offsets > screen->get_shader_param( 96 screen, TGSI_PROCESSOR_FRAGMENT, PIPE_SHADER_CAP_MAX_TEMPS)) { 97 98 FREE(t_array); 99 return NULL; 100 } 101 102 shader = ureg_create(TGSI_PROCESSOR_FRAGMENT); 103 if (!shader) { 104 FREE(t_array); 105 return NULL; 106 } 107 108 i_vtex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX, TGSI_INTERPOLATE_LINEAR); 109 sampler = ureg_DECL_sampler(shader, 0); 110 111 for (i = 0; i < num_offsets; ++i) 112 t_array[i] = ureg_DECL_temporary(shader); 113 o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0); 114 115 /* 116 * t_array[0..*] = vtex + offset[0..*] 117 * t_array[0..*] = tex(t_array[0..*], sampler) 118 * result = partial_bubblesort(t_array)[mid] 119 */ 120 121 for (i = 0; i < num_offsets; ++i) { 122 if (!is_vec_zero(offsets[i])) { 123 ureg_ADD(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_XY), 124 i_vtex, ureg_imm2f(shader, offsets[i].x, offsets[i].y)); 125 ureg_MOV(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_ZW), 126 ureg_imm1f(shader, 0.0f)); 127 } 128 } 129 130 for (i = 0; i < num_offsets; ++i) { 131 struct ureg_src src = is_vec_zero(offsets[i]) ? i_vtex : ureg_src(t_array[i]); 132 ureg_TEX(shader, t_array[i], TGSI_TEXTURE_2D, src, sampler); 133 } 134 135 // TODO: Couldn't this be improved even more? 136 for (i = 0; i <= median; ++i) { 137 for (j = 1; j < (num_offsets - i - 1); ++j) { 138 struct ureg_dst tmp = ureg_DECL_temporary(shader); 139 ureg_MOV(shader, tmp, ureg_src(t_array[j])); 140 ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1])); 141 ureg_MIN(shader, t_array[j - 1], ureg_src(tmp), ureg_src(t_array[j - 1])); 142 ureg_release_temporary(shader, tmp); 143 } 144 if (i == median) 145 ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1])); 146 else 147 ureg_MIN(shader, t_array[j - 1], ureg_src(t_array[j]), ureg_src(t_array[j - 1])); 148 } 149 ureg_MOV(shader, o_fragment, ureg_src(t_array[median])); 150 151 ureg_END(shader); 152 153 FREE(t_array); 154 return ureg_create_shader_and_destroy(shader, filter->pipe); 155 } 156 157 static void 158 generate_offsets(enum vl_median_filter_shape shape, unsigned size, 159 struct vertex2f **offsets, unsigned *num_offsets) 160 { 161 int i = 0, half_size; 162 struct vertex2f v; 163 164 assert(offsets && num_offsets); 165 166 /* size needs to be odd */ 167 size = align(size + 1, 2) - 1; 168 half_size = size >> 1; 169 170 switch(shape) { 171 case VL_MEDIAN_FILTER_BOX: 172 *num_offsets = size*size; 173 break; 174 175 case VL_MEDIAN_FILTER_CROSS: 176 case VL_MEDIAN_FILTER_X: 177 *num_offsets = size + size - 1; 178 break; 179 180 case VL_MEDIAN_FILTER_HORIZONTAL: 181 case VL_MEDIAN_FILTER_VERTICAL: 182 *num_offsets = size; 183 break; 184 185 default: 186 *num_offsets = 0; 187 return; 188 } 189 190 *offsets = MALLOC(sizeof(struct vertex2f) * *num_offsets); 191 if (!*offsets) 192 return; 193 194 switch(shape) { 195 case VL_MEDIAN_FILTER_BOX: 196 for (v.x = -half_size; v.x <= half_size; ++v.x) 197 for (v.y = -half_size; v.y <= half_size; ++v.y) 198 (*offsets)[i++] = v; 199 break; 200 201 case VL_MEDIAN_FILTER_CROSS: 202 v.y = 0.0f; 203 for (v.x = -half_size; v.x <= half_size; ++v.x) 204 (*offsets)[i++] = v; 205 206 v.x = 0.0f; 207 for (v.y = -half_size; v.y <= half_size; ++v.y) 208 if (v.y != 0.0f) 209 (*offsets)[i++] = v; 210 break; 211 212 case VL_MEDIAN_FILTER_X: 213 for (v.x = v.y = -half_size; v.x <= half_size; ++v.x, ++v.y) 214 (*offsets)[i++] = v; 215 216 for (v.x = -half_size, v.y = half_size; v.x <= half_size; ++v.x, --v.y) 217 if (v.y != 0.0f) 218 (*offsets)[i++] = v; 219 break; 220 221 case VL_MEDIAN_FILTER_HORIZONTAL: 222 v.y = 0.0f; 223 for (v.x = -half_size; v.x <= half_size; ++v.x) 224 (*offsets)[i++] = v; 225 break; 226 227 case VL_MEDIAN_FILTER_VERTICAL: 228 v.x = 0.0f; 229 for (v.y = -half_size; v.y <= half_size; ++v.y) 230 (*offsets)[i++] = v; 231 break; 232 } 233 234 assert(i == *num_offsets); 235 } 236 237 bool 238 vl_median_filter_init(struct vl_median_filter *filter, struct pipe_context *pipe, 239 unsigned width, unsigned height, unsigned size, 240 enum vl_median_filter_shape shape) 241 { 242 struct pipe_rasterizer_state rs_state; 243 struct pipe_blend_state blend; 244 struct pipe_sampler_state sampler; 245 struct vertex2f *offsets = NULL; 246 struct pipe_vertex_element ve; 247 unsigned i, num_offsets = 0; 248 249 assert(filter && pipe); 250 assert(width && height); 251 assert(size > 1 && size < 20); 252 253 memset(filter, 0, sizeof(*filter)); 254 filter->pipe = pipe; 255 256 memset(&rs_state, 0, sizeof(rs_state)); 257 rs_state.gl_rasterization_rules = true; 258 rs_state.depth_clip = 1; 259 filter->rs_state = pipe->create_rasterizer_state(pipe, &rs_state); 260 if (!filter->rs_state) 261 goto error_rs_state; 262 263 memset(&blend, 0, sizeof blend); 264 blend.rt[0].rgb_func = PIPE_BLEND_ADD; 265 blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE; 266 blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE; 267 blend.rt[0].alpha_func = PIPE_BLEND_ADD; 268 blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE; 269 blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE; 270 blend.logicop_func = PIPE_LOGICOP_CLEAR; 271 blend.rt[0].colormask = PIPE_MASK_RGBA; 272 filter->blend = pipe->create_blend_state(pipe, &blend); 273 if (!filter->blend) 274 goto error_blend; 275 276 memset(&sampler, 0, sizeof(sampler)); 277 sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 278 sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 279 sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 280 sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST; 281 sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE; 282 sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST; 283 sampler.compare_mode = PIPE_TEX_COMPARE_NONE; 284 sampler.compare_func = PIPE_FUNC_ALWAYS; 285 sampler.normalized_coords = 1; 286 filter->sampler = pipe->create_sampler_state(pipe, &sampler); 287 if (!filter->sampler) 288 goto error_sampler; 289 290 filter->quad = vl_vb_upload_quads(pipe); 291 if(!filter->quad.buffer) 292 goto error_quad; 293 294 memset(&ve, 0, sizeof(ve)); 295 ve.src_offset = 0; 296 ve.instance_divisor = 0; 297 ve.vertex_buffer_index = 0; 298 ve.src_format = PIPE_FORMAT_R32G32_FLOAT; 299 filter->ves = pipe->create_vertex_elements_state(pipe, 1, &ve); 300 if (!filter->ves) 301 goto error_ves; 302 303 generate_offsets(shape, size, &offsets, &num_offsets); 304 if (!offsets) 305 goto error_offsets; 306 307 for (i = 0; i < num_offsets; ++i) { 308 offsets[i].x /= width; 309 offsets[i].y /= height; 310 } 311 312 filter->vs = create_vert_shader(filter); 313 if (!filter->vs) 314 goto error_vs; 315 316 filter->fs = create_frag_shader(filter, offsets, num_offsets); 317 if (!filter->fs) 318 goto error_fs; 319 320 FREE(offsets); 321 return true; 322 323 error_fs: 324 pipe->delete_vs_state(pipe, filter->vs); 325 326 error_vs: 327 FREE(offsets); 328 329 error_offsets: 330 pipe->delete_vertex_elements_state(pipe, filter->ves); 331 332 error_ves: 333 pipe_resource_reference(&filter->quad.buffer, NULL); 334 335 error_quad: 336 pipe->delete_sampler_state(pipe, filter->sampler); 337 338 error_sampler: 339 pipe->delete_blend_state(pipe, filter->blend); 340 341 error_blend: 342 pipe->delete_rasterizer_state(pipe, filter->rs_state); 343 344 error_rs_state: 345 return false; 346 } 347 348 void 349 vl_median_filter_cleanup(struct vl_median_filter *filter) 350 { 351 assert(filter); 352 353 filter->pipe->delete_sampler_state(filter->pipe, filter->sampler); 354 filter->pipe->delete_blend_state(filter->pipe, filter->blend); 355 filter->pipe->delete_rasterizer_state(filter->pipe, filter->rs_state); 356 filter->pipe->delete_vertex_elements_state(filter->pipe, filter->ves); 357 pipe_resource_reference(&filter->quad.buffer, NULL); 358 359 filter->pipe->delete_vs_state(filter->pipe, filter->vs); 360 filter->pipe->delete_fs_state(filter->pipe, filter->fs); 361 } 362 363 void 364 vl_median_filter_render(struct vl_median_filter *filter, 365 struct pipe_sampler_view *src, 366 struct pipe_surface *dst) 367 { 368 struct pipe_viewport_state viewport; 369 struct pipe_framebuffer_state fb_state; 370 371 assert(filter && src && dst); 372 373 memset(&viewport, 0, sizeof(viewport)); 374 viewport.scale[0] = dst->width; 375 viewport.scale[1] = dst->height; 376 viewport.scale[2] = 1; 377 viewport.scale[3] = 1; 378 379 memset(&fb_state, 0, sizeof(fb_state)); 380 fb_state.width = dst->width; 381 fb_state.height = dst->height; 382 fb_state.nr_cbufs = 1; 383 fb_state.cbufs[0] = dst; 384 385 filter->pipe->bind_rasterizer_state(filter->pipe, filter->rs_state); 386 filter->pipe->bind_blend_state(filter->pipe, filter->blend); 387 filter->pipe->bind_fragment_sampler_states(filter->pipe, 1, &filter->sampler); 388 filter->pipe->set_fragment_sampler_views(filter->pipe, 1, &src); 389 filter->pipe->bind_vs_state(filter->pipe, filter->vs); 390 filter->pipe->bind_fs_state(filter->pipe, filter->fs); 391 filter->pipe->set_framebuffer_state(filter->pipe, &fb_state); 392 filter->pipe->set_viewport_state(filter->pipe, &viewport); 393 filter->pipe->set_vertex_buffers(filter->pipe, 1, &filter->quad); 394 filter->pipe->bind_vertex_elements_state(filter->pipe, filter->ves); 395 396 util_draw_arrays(filter->pipe, PIPE_PRIM_QUADS, 0, 4); 397 } 398
