vl_idct.c revision 7408a6ab89e0bc87209b50334604fae93277fdc6
1/************************************************************************** 2 * 3 * Copyright 2010 Christian König 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 "vl_idct.h" 29#include "util/u_draw.h" 30#include <assert.h> 31#include <pipe/p_context.h> 32#include <pipe/p_screen.h> 33#include <util/u_inlines.h> 34#include <util/u_sampler.h> 35#include <util/u_format.h> 36#include <tgsi/tgsi_ureg.h> 37#include "vl_types.h" 38 39#define BLOCK_WIDTH 8 40#define BLOCK_HEIGHT 8 41 42#define SCALE_FACTOR_16_TO_9 (32768.0f / 256.0f) 43 44#define STAGE1_SCALE 4.0f 45#define STAGE2_SCALE (SCALE_FACTOR_16_TO_9 / STAGE1_SCALE) 46 47struct vertex_shader_consts 48{ 49 struct vertex4f norm; 50}; 51 52enum VS_INPUT 53{ 54 VS_I_RECT, 55 VS_I_VPOS, 56 57 NUM_VS_INPUTS 58}; 59 60enum VS_OUTPUT 61{ 62 VS_O_VPOS, 63 VS_O_BLOCK, 64 VS_O_TEX, 65 VS_O_START 66}; 67 68static const float const_matrix[8][8] = { 69 { 0.3535530f, 0.3535530f, 0.3535530f, 0.3535530f, 0.3535530f, 0.3535530f, 0.353553f, 0.3535530f }, 70 { 0.4903930f, 0.4157350f, 0.2777850f, 0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f }, 71 { 0.4619400f, 0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f, 0.191342f, 0.4619400f }, 72 { 0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f, 0.2777850f, 0.4903930f, 0.097545f, -0.4157350f }, 73 { 0.3535530f, -0.3535530f, -0.3535530f, 0.3535540f, 0.3535530f, -0.3535540f, -0.353553f, 0.3535530f }, 74 { 0.2777850f, -0.4903930f, 0.0975452f, 0.4157350f, -0.4157350f, -0.0975451f, 0.490393f, -0.2777850f }, 75 { 0.1913420f, -0.4619400f, 0.4619400f, -0.1913420f, -0.1913410f, 0.4619400f, -0.461940f, 0.1913420f }, 76 { 0.0975451f, -0.2777850f, 0.4157350f, -0.4903930f, 0.4903930f, -0.4157350f, 0.277786f, -0.0975458f } 77}; 78 79/* vertices for a quad covering a block */ 80static const struct vertex2f const_quad[4] = { 81 {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f} 82}; 83 84static void * 85create_vert_shader(struct vl_idct *idct) 86{ 87 struct ureg_program *shader; 88 struct ureg_src scale; 89 struct ureg_src vrect, vpos; 90 struct ureg_dst t_vpos; 91 struct ureg_dst o_vpos, o_block, o_tex, o_start; 92 93 shader = ureg_create(TGSI_PROCESSOR_VERTEX); 94 if (!shader) 95 return NULL; 96 97 scale = ureg_imm2f(shader, 98 (float)BLOCK_WIDTH / idct->destination->width0, 99 (float)BLOCK_HEIGHT / idct->destination->height0); 100 101 t_vpos = ureg_DECL_temporary(shader); 102 103 vrect = ureg_DECL_vs_input(shader, VS_I_RECT); 104 vpos = ureg_DECL_vs_input(shader, VS_I_VPOS); 105 106 o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS); 107 o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK); 108 o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX); 109 o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START); 110 111 /* 112 * t_vpos = vpos + vrect 113 * o_vpos.xy = t_vpos * scale 114 * o_vpos.zw = vpos 115 * 116 * o_block = vrect 117 * o_tex = t_pos 118 * o_start = vpos * scale 119 * 120 */ 121 ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect); 122 ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), scale); 123 ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos)); 124 ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos); 125 126 ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect); 127 ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos)); 128 ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, scale); 129 130 ureg_release_temporary(shader, t_vpos); 131 132 ureg_END(shader); 133 134 return ureg_create_shader_and_destroy(shader, idct->pipe); 135} 136 137static void 138matrix_mul(struct ureg_program *shader, struct ureg_dst dst, 139 struct ureg_src tc[2], struct ureg_src sampler[2], 140 struct ureg_src start[2], struct ureg_src step[2], 141 bool fetch4[2], float scale) 142{ 143 struct ureg_dst t_tc[2], m[2][2], tmp[2]; 144 unsigned side, i, j; 145 146 for(i = 0; i < 2; ++i) { 147 t_tc[i] = ureg_DECL_temporary(shader); 148 for(j = 0; j < 2; ++j) 149 m[i][j] = ureg_DECL_temporary(shader); 150 tmp[i] = ureg_DECL_temporary(shader); 151 } 152 153 /* 154 * m[0..1][0] = ? 155 * tmp[0..1] = dot4(m[0..1][0], m[0..1][1]) 156 * fragment = tmp[0] + tmp[1] 157 */ 158 ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_X), ureg_scalar(start[0], TGSI_SWIZZLE_X)); 159 ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_Y), ureg_scalar(tc[0], TGSI_SWIZZLE_Y)); 160 161 if(fetch4[1]) { 162 ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_X), ureg_scalar(start[1], TGSI_SWIZZLE_Y)); 163 ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), ureg_scalar(tc[1], TGSI_SWIZZLE_X)); 164 } else { 165 ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_X), ureg_scalar(tc[1], TGSI_SWIZZLE_X)); 166 ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), ureg_scalar(start[1], TGSI_SWIZZLE_Y)); 167 } 168 169 for(side = 0; side < 2; ++side) { 170 for(i = 0; i < 2; ++i) { 171 if(fetch4[side]) { 172 ureg_TEX(shader, m[i][side], TGSI_TEXTURE_2D, ureg_src(t_tc[side]), sampler[side]); 173 ureg_MOV(shader, ureg_writemask(t_tc[side], TGSI_WRITEMASK_X), step[side]); 174 175 } else for(j = 0; j < 4; ++j) { 176 /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */ 177 ureg_TEX(shader, tmp[side], TGSI_TEXTURE_2D, ureg_src(t_tc[side]), sampler[side]); 178 ureg_MOV(shader, ureg_writemask(m[i][side], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[side]), TGSI_SWIZZLE_X)); 179 180 ureg_ADD(shader, ureg_writemask(t_tc[side], TGSI_WRITEMASK_X << side), ureg_src(t_tc[side]), step[side]); 181 } 182 } 183 } 184 185 ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(m[0][0]), ureg_src(m[0][1])); 186 ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(m[1][0]), ureg_src(m[1][1])); 187 ureg_ADD(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(tmp[0]), ureg_src(tmp[1])); 188 ureg_MUL(shader, dst, ureg_src(tmp[0]), ureg_imm1f(shader, scale)); 189 190 for(i = 0; i < 2; ++i) { 191 ureg_release_temporary(shader, t_tc[i]); 192 for(j = 0; j < 2; ++j) 193 ureg_release_temporary(shader, m[i][j]); 194 ureg_release_temporary(shader, tmp[i]); 195 } 196} 197 198static void * 199create_transpose_frag_shader(struct vl_idct *idct) 200{ 201 struct ureg_program *shader; 202 struct ureg_src tc[2], sampler[2]; 203 struct ureg_src start[2], step[2]; 204 struct ureg_dst fragment; 205 bool fetch4[2]; 206 207 shader = ureg_create(TGSI_PROCESSOR_FRAGMENT); 208 if (!shader) 209 return NULL; 210 211 tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR); 212 tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR); 213 214 start[0] = ureg_imm1f(shader, 0.0f); 215 start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT); 216 217 step[0] = ureg_imm1f(shader, 4.0f / BLOCK_HEIGHT); 218 step[1] = ureg_imm1f(shader, 1.0f / idct->destination->height0); 219 220 sampler[0] = ureg_DECL_sampler(shader, 0); 221 sampler[1] = ureg_DECL_sampler(shader, 1); 222 223 fetch4[0] = true; 224 fetch4[1] = false; 225 226 fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0); 227 228 matrix_mul(shader, fragment, tc, sampler, start, step, fetch4, STAGE1_SCALE); 229 230 ureg_END(shader); 231 232 return ureg_create_shader_and_destroy(shader, idct->pipe); 233} 234 235static void * 236create_matrix_frag_shader(struct vl_idct *idct) 237{ 238 struct ureg_program *shader; 239 struct ureg_src tc[2], sampler[2]; 240 struct ureg_src start[2], step[2]; 241 struct ureg_dst fragment; 242 bool fetch4[2]; 243 244 shader = ureg_create(TGSI_PROCESSOR_FRAGMENT); 245 if (!shader) 246 return NULL; 247 248 tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR); 249 tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR); 250 251 start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT); 252 start[1] = ureg_imm1f(shader, 0.0f); 253 254 step[0] = ureg_imm1f(shader, 1.0f / idct->destination->width0); 255 step[1] = ureg_imm1f(shader, 4.0f / BLOCK_WIDTH); 256 257 sampler[0] = ureg_DECL_sampler(shader, 1); 258 sampler[1] = ureg_DECL_sampler(shader, 0); 259 260 fetch4[0] = false; 261 fetch4[1] = true; 262 263 fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0); 264 265 matrix_mul(shader, fragment, tc, sampler, start, step, fetch4, STAGE2_SCALE); 266 267 ureg_END(shader); 268 269 return ureg_create_shader_and_destroy(shader, idct->pipe); 270} 271 272static void * 273create_empty_block_frag_shader(struct vl_idct *idct) 274{ 275 struct ureg_program *shader; 276 struct ureg_dst fragment; 277 278 shader = ureg_create(TGSI_PROCESSOR_FRAGMENT); 279 if (!shader) 280 return NULL; 281 282 fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0); 283 284 ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f)); 285 286 ureg_END(shader); 287 288 return ureg_create_shader_and_destroy(shader, idct->pipe); 289} 290 291static void 292xfer_buffers_map(struct vl_idct *idct) 293{ 294 struct pipe_box rect = 295 { 296 0, 0, 0, 297 idct->destination->width0, 298 idct->destination->height0, 299 1 300 }; 301 302 idct->tex_transfer = idct->pipe->get_transfer 303 ( 304 idct->pipe, idct->textures.individual.source, 305 u_subresource(0, 0), 306 PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, 307 &rect 308 ); 309 310 idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer); 311 312 idct->vectors = pipe_buffer_map 313 ( 314 idct->pipe, 315 idct->vertex_bufs.individual.pos.buffer, 316 PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, 317 &idct->vec_transfer 318 ); 319} 320 321static void 322xfer_buffers_unmap(struct vl_idct *idct) 323{ 324 pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer); 325 326 idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer); 327 idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer); 328} 329 330static bool 331init_shaders(struct vl_idct *idct) 332{ 333 idct->vs = create_vert_shader(idct); 334 idct->transpose_fs = create_transpose_frag_shader(idct); 335 idct->matrix_fs = create_matrix_frag_shader(idct); 336 idct->eb_fs = create_empty_block_frag_shader(idct); 337 338 return 339 idct->vs != NULL && 340 idct->transpose_fs != NULL && 341 idct->matrix_fs != NULL && 342 idct->eb_fs != NULL; 343} 344 345static void 346cleanup_shaders(struct vl_idct *idct) 347{ 348 idct->pipe->delete_vs_state(idct->pipe, idct->vs); 349 idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs); 350 idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs); 351 idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs); 352} 353 354static bool 355init_buffers(struct vl_idct *idct) 356{ 357 struct pipe_resource template; 358 struct pipe_sampler_view sampler_view; 359 struct pipe_vertex_element vertex_elems[2]; 360 unsigned i; 361 362 idct->max_blocks = 363 align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH * 364 align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT * 365 idct->destination->depth0; 366 367 memset(&template, 0, sizeof(struct pipe_resource)); 368 template.target = PIPE_TEXTURE_2D; 369 template.format = PIPE_FORMAT_R32G32B32A32_FLOAT; 370 template.last_level = 0; 371 template.width0 = 2; 372 template.height0 = 8; 373 template.depth0 = 1; 374 template.usage = PIPE_USAGE_IMMUTABLE; 375 template.bind = PIPE_BIND_SAMPLER_VIEW; 376 template.flags = 0; 377 378 template.format = idct->destination->format; 379 template.width0 = idct->destination->width0; 380 template.height0 = idct->destination->height0; 381 template.depth0 = idct->destination->depth0; 382 template.usage = PIPE_USAGE_DYNAMIC; 383 idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template); 384 385 template.usage = PIPE_USAGE_STATIC; 386 idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template); 387 388 for (i = 0; i < 4; ++i) { 389 if(idct->textures.all[i] == NULL) 390 return false; /* a texture failed to allocate */ 391 392 u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format); 393 idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view); 394 } 395 396 idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f); 397 idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1; 398 idct->vertex_bufs.individual.quad.buffer_offset = 0; 399 idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create 400 ( 401 idct->pipe->screen, 402 PIPE_BIND_VERTEX_BUFFER, 403 sizeof(struct vertex2f) * 4 * idct->max_blocks 404 ); 405 406 if(idct->vertex_bufs.individual.quad.buffer == NULL) 407 return false; 408 409 idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f); 410 idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1; 411 idct->vertex_bufs.individual.pos.buffer_offset = 0; 412 idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create 413 ( 414 idct->pipe->screen, 415 PIPE_BIND_VERTEX_BUFFER, 416 sizeof(struct vertex2f) * 4 * idct->max_blocks 417 ); 418 419 if(idct->vertex_bufs.individual.pos.buffer == NULL) 420 return false; 421 422 /* Rect element */ 423 vertex_elems[0].src_offset = 0; 424 vertex_elems[0].instance_divisor = 0; 425 vertex_elems[0].vertex_buffer_index = 0; 426 vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT; 427 428 /* Pos element */ 429 vertex_elems[1].src_offset = 0; 430 vertex_elems[1].instance_divisor = 0; 431 vertex_elems[1].vertex_buffer_index = 1; 432 vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT; 433 434 idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems); 435 436 return true; 437} 438 439static void 440cleanup_buffers(struct vl_idct *idct) 441{ 442 unsigned i; 443 444 assert(idct); 445 446 for (i = 0; i < 4; ++i) { 447 pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL); 448 pipe_resource_reference(&idct->textures.all[i], NULL); 449 } 450 451 idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state); 452 pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL); 453 pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL); 454} 455 456static void 457init_constants(struct vl_idct *idct) 458{ 459 struct pipe_transfer *buf_transfer; 460 struct vertex2f *v; 461 462 unsigned i; 463 464 /* quad vectors */ 465 v = pipe_buffer_map 466 ( 467 idct->pipe, 468 idct->vertex_bufs.individual.quad.buffer, 469 PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, 470 &buf_transfer 471 ); 472 for ( i = 0; i < idct->max_blocks; ++i) 473 memcpy(v + i * 4, &const_quad, sizeof(const_quad)); 474 pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, buf_transfer); 475} 476 477static void 478init_state(struct vl_idct *idct) 479{ 480 struct pipe_sampler_state sampler; 481 unsigned i; 482 483 idct->num_blocks = 0; 484 idct->num_empty_blocks = 0; 485 486 idct->viewport.scale[0] = idct->destination->width0; 487 idct->viewport.scale[1] = idct->destination->height0; 488 idct->viewport.scale[2] = 1; 489 idct->viewport.scale[3] = 1; 490 idct->viewport.translate[0] = 0; 491 idct->viewport.translate[1] = 0; 492 idct->viewport.translate[2] = 0; 493 idct->viewport.translate[3] = 0; 494 495 idct->fb_state.width = idct->destination->width0; 496 idct->fb_state.height = idct->destination->height0; 497 idct->fb_state.nr_cbufs = 1; 498 idct->fb_state.zsbuf = NULL; 499 500 for (i = 0; i < 4; ++i) { 501 memset(&sampler, 0, sizeof(sampler)); 502 sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 503 sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 504 sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 505 sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST; 506 sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE; 507 sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST; 508 sampler.compare_mode = PIPE_TEX_COMPARE_NONE; 509 sampler.compare_func = PIPE_FUNC_ALWAYS; 510 sampler.normalized_coords = 1; 511 /*sampler.shadow_ambient = ; */ 512 /*sampler.lod_bias = ; */ 513 sampler.min_lod = 0; 514 /*sampler.max_lod = ; */ 515 /*sampler.border_color[0] = ; */ 516 /*sampler.max_anisotropy = ; */ 517 idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler); 518 } 519} 520 521static void 522cleanup_state(struct vl_idct *idct) 523{ 524 unsigned i; 525 526 for (i = 0; i < 4; ++i) 527 idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]); 528} 529 530struct pipe_resource * 531vl_idct_upload_matrix(struct pipe_context *pipe) 532{ 533 struct pipe_resource template, *matrix; 534 struct pipe_transfer *buf_transfer; 535 unsigned i, j, pitch; 536 float *f; 537 538 struct pipe_box rect = 539 { 540 0, 0, 0, 541 BLOCK_WIDTH, 542 BLOCK_HEIGHT, 543 1 544 }; 545 546 memset(&template, 0, sizeof(struct pipe_resource)); 547 template.target = PIPE_TEXTURE_2D; 548 template.format = PIPE_FORMAT_R32G32B32A32_FLOAT; 549 template.last_level = 0; 550 template.width0 = 2; 551 template.height0 = 8; 552 template.depth0 = 1; 553 template.usage = PIPE_USAGE_IMMUTABLE; 554 template.bind = PIPE_BIND_SAMPLER_VIEW; 555 template.flags = 0; 556 557 matrix = pipe->screen->resource_create(pipe->screen, &template); 558 559 /* matrix */ 560 buf_transfer = pipe->get_transfer 561 ( 562 pipe, matrix, 563 u_subresource(0, 0), 564 PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, 565 &rect 566 ); 567 pitch = buf_transfer->stride / util_format_get_blocksize(buf_transfer->resource->format); 568 569 f = pipe->transfer_map(pipe, buf_transfer); 570 for(i = 0; i < BLOCK_HEIGHT; ++i) 571 for(j = 0; j < BLOCK_WIDTH; ++j) 572 f[i * pitch * 4 + j] = const_matrix[j][i]; // transpose 573 574 pipe->transfer_unmap(pipe, buf_transfer); 575 pipe->transfer_destroy(pipe, buf_transfer); 576 577 return matrix; 578} 579 580bool 581vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix) 582{ 583 assert(idct && pipe && dst); 584 585 idct->pipe = pipe; 586 pipe_resource_reference(&idct->textures.individual.matrix, matrix); 587 pipe_resource_reference(&idct->textures.individual.transpose, matrix); 588 pipe_resource_reference(&idct->destination, dst); 589 590 init_state(idct); 591 592 if(!init_shaders(idct)) 593 return false; 594 595 if(!init_buffers(idct)) { 596 cleanup_shaders(idct); 597 return false; 598 } 599 600 idct->surfaces.intermediate = idct->pipe->screen->get_tex_surface( 601 idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, 0, 602 PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET); 603 604 idct->surfaces.destination = idct->pipe->screen->get_tex_surface( 605 idct->pipe->screen, idct->destination, 0, 0, 0, 606 PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET); 607 608 init_constants(idct); 609 xfer_buffers_map(idct); 610 611 return true; 612} 613 614void 615vl_idct_cleanup(struct vl_idct *idct) 616{ 617 idct->pipe->screen->tex_surface_destroy(idct->surfaces.destination); 618 idct->pipe->screen->tex_surface_destroy(idct->surfaces.intermediate); 619 620 cleanup_shaders(idct); 621 cleanup_buffers(idct); 622 623 cleanup_state(idct); 624 625 pipe_resource_reference(&idct->destination, NULL); 626} 627 628void 629vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block) 630{ 631 struct vertex2f v, *v_dst; 632 633 unsigned tex_pitch; 634 short *texels; 635 636 unsigned i; 637 638 assert(idct); 639 640 if(block) { 641 tex_pitch = idct->tex_transfer->stride / util_format_get_blocksize(idct->tex_transfer->resource->format); 642 texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH; 643 644 for (i = 0; i < BLOCK_HEIGHT; ++i) 645 memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * 2); 646 647 /* non empty blocks fills the vector buffer from left to right */ 648 v_dst = idct->vectors + idct->num_blocks * 4; 649 650 idct->num_blocks++; 651 652 } else { 653 654 /* while empty blocks fills the vector buffer from right to left */ 655 v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4; 656 657 idct->num_empty_blocks++; 658 } 659 660 v.x = x; 661 v.y = y; 662 663 for (i = 0; i < 4; ++i) { 664 v_dst[i] = v; 665 } 666} 667 668void 669vl_idct_flush(struct vl_idct *idct) 670{ 671 xfer_buffers_unmap(idct); 672 673 if(idct->num_blocks > 0) { 674 675 /* first stage */ 676 idct->fb_state.cbufs[0] = idct->surfaces.intermediate; 677 idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state); 678 idct->pipe->set_viewport_state(idct->pipe, &idct->viewport); 679 680 idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all); 681 idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state); 682 idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]); 683 idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]); 684 idct->pipe->bind_vs_state(idct->pipe, idct->vs); 685 idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs); 686 687 util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4); 688 689 /* second stage */ 690 idct->fb_state.cbufs[0] = idct->surfaces.destination; 691 idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state); 692 idct->pipe->set_viewport_state(idct->pipe, &idct->viewport); 693 694 idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all); 695 idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state); 696 idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]); 697 idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]); 698 idct->pipe->bind_vs_state(idct->pipe, idct->vs); 699 idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs); 700 701 util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4); 702 } 703 704 if(idct->num_empty_blocks > 0) { 705 706 /* empty block handling */ 707 idct->fb_state.cbufs[0] = idct->surfaces.destination; 708 idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state); 709 idct->pipe->set_viewport_state(idct->pipe, &idct->viewport); 710 711 idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all); 712 idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state); 713 idct->pipe->set_fragment_sampler_views(idct->pipe, 4, idct->sampler_views.all); 714 idct->pipe->bind_fragment_sampler_states(idct->pipe, 4, idct->samplers.all); 715 idct->pipe->bind_vs_state(idct->pipe, idct->vs); 716 idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs); 717 718 util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 719 (idct->max_blocks - idct->num_empty_blocks) * 4, 720 idct->num_empty_blocks * 4); 721 } 722 723 idct->num_blocks = 0; 724 idct->num_empty_blocks = 0; 725 xfer_buffers_map(idct); 726} 727