draw_context.c revision 57f2f84f0e2a06313d30b3d9341229f6643683a1
1/************************************************************************** 2 * 3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. 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 /* 29 * Authors: 30 * Keith Whitwell <keith@tungstengraphics.com> 31 */ 32 33 34#include "pipe/p_context.h" 35#include "util/u_memory.h" 36#include "util/u_math.h" 37#include "util/u_cpu_detect.h" 38#include "util/u_inlines.h" 39#include "draw_context.h" 40#include "draw_vs.h" 41#include "draw_gs.h" 42 43#if HAVE_LLVM 44#include "gallivm/lp_bld_init.h" 45#include "draw_llvm.h" 46 47static boolean 48draw_get_option_use_llvm(void) 49{ 50 static boolean first = TRUE; 51 static boolean value; 52 if (first) { 53 first = FALSE; 54 value = debug_get_bool_option("DRAW_USE_LLVM", TRUE); 55 56#ifdef PIPE_ARCH_X86 57 util_cpu_detect(); 58 /* require SSE2 due to LLVM PR6960. */ 59 if (!util_cpu_caps.has_sse2) 60 value = FALSE; 61#endif 62 } 63 return value; 64} 65#endif 66 67 68 69/** 70 * Create new draw module context. 71 */ 72struct draw_context * 73draw_create(struct pipe_context *pipe) 74{ 75 return draw_create_gallivm(pipe, NULL); 76} 77 78 79 80/** 81 * Create new draw module context with gallivm state for LLVM JIT. 82 */ 83struct draw_context * 84draw_create_gallivm(struct pipe_context *pipe, struct gallivm_state *gallivm) 85{ 86 struct draw_context *draw = CALLOC_STRUCT( draw_context ); 87 if (draw == NULL) 88 goto fail; 89 90#if HAVE_LLVM 91 if (draw_get_option_use_llvm()) { 92 if (!gallivm) { 93 gallivm = gallivm_create(); 94 draw->own_gallivm = gallivm; 95 } 96 97 if (gallivm) 98 draw->llvm = draw_llvm_create(draw, gallivm); 99 } 100#endif 101 102 if (!draw_init(draw)) 103 goto fail; 104 105 draw->pipe = pipe; 106 107 return draw; 108 109fail: 110 draw_destroy( draw ); 111 return NULL; 112} 113 114 115 116boolean draw_init(struct draw_context *draw) 117{ 118 /* 119 * Note that several functions compute the clipmask of the predefined 120 * formats with hardcoded formulas instead of using these. So modifications 121 * here must be reflected there too. 122 */ 123 124 ASSIGN_4V( draw->plane[0], -1, 0, 0, 1 ); 125 ASSIGN_4V( draw->plane[1], 1, 0, 0, 1 ); 126 ASSIGN_4V( draw->plane[2], 0, -1, 0, 1 ); 127 ASSIGN_4V( draw->plane[3], 0, 1, 0, 1 ); 128 ASSIGN_4V( draw->plane[4], 0, 0, 1, 1 ); /* yes these are correct */ 129 ASSIGN_4V( draw->plane[5], 0, 0, -1, 1 ); /* mesa's a bit wonky */ 130 draw->nr_planes = 6; 131 draw->clip_xy = TRUE; 132 draw->clip_z = TRUE; 133 134 135 draw->reduced_prim = ~0; /* != any of PIPE_PRIM_x */ 136 137 138 if (!draw_pipeline_init( draw )) 139 return FALSE; 140 141 if (!draw_pt_init( draw )) 142 return FALSE; 143 144 if (!draw_vs_init( draw )) 145 return FALSE; 146 147 if (!draw_gs_init( draw )) 148 return FALSE; 149 150 return TRUE; 151} 152 153 154void draw_destroy( struct draw_context *draw ) 155{ 156 struct pipe_context *pipe; 157 int i, j; 158 159 if (!draw) 160 return; 161 162 pipe = draw->pipe; 163 164 /* free any rasterizer CSOs that we may have created. 165 */ 166 for (i = 0; i < 2; i++) { 167 for (j = 0; j < 2; j++) { 168 if (draw->rasterizer_no_cull[i][j]) { 169 pipe->delete_rasterizer_state(pipe, draw->rasterizer_no_cull[i][j]); 170 } 171 } 172 } 173 174 for (i = 0; i < draw->pt.nr_vertex_buffers; i++) { 175 pipe_resource_reference(&draw->pt.vertex_buffer[i].buffer, NULL); 176 } 177 178 /* Not so fast -- we're just borrowing this at the moment. 179 * 180 if (draw->render) 181 draw->render->destroy( draw->render ); 182 */ 183 184 draw_pipeline_destroy( draw ); 185 draw_pt_destroy( draw ); 186 draw_vs_destroy( draw ); 187 draw_gs_destroy( draw ); 188#ifdef HAVE_LLVM 189 if (draw->llvm) 190 draw_llvm_destroy( draw->llvm ); 191 192 if (draw->own_gallivm) 193 gallivm_destroy(draw->own_gallivm); 194#endif 195 196 FREE( draw ); 197} 198 199 200 201void draw_flush( struct draw_context *draw ) 202{ 203 draw_do_flush( draw, DRAW_FLUSH_BACKEND ); 204} 205 206 207/** 208 * Specify the Minimum Resolvable Depth factor for polygon offset. 209 * This factor potentially depends on the number of Z buffer bits, 210 * the rasterization algorithm and the arithmetic performed on Z 211 * values between vertex shading and rasterization. It will vary 212 * from one driver to another. 213 */ 214void draw_set_mrd(struct draw_context *draw, double mrd) 215{ 216 draw->mrd = mrd; 217} 218 219 220static void update_clip_flags( struct draw_context *draw ) 221{ 222 draw->clip_xy = !draw->driver.bypass_clip_xy; 223 draw->clip_z = (!draw->driver.bypass_clip_z && 224 !draw->depth_clamp); 225 draw->clip_user = (draw->nr_planes > 6); 226} 227 228/** 229 * Register new primitive rasterization/rendering state. 230 * This causes the drawing pipeline to be rebuilt. 231 */ 232void draw_set_rasterizer_state( struct draw_context *draw, 233 const struct pipe_rasterizer_state *raster, 234 void *rast_handle ) 235{ 236 if (!draw->suspend_flushing) { 237 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 238 239 draw->rasterizer = raster; 240 draw->rast_handle = rast_handle; 241 242 } 243} 244 245/* With a little more work, llvmpipe will be able to turn this off and 246 * do its own x/y clipping. 247 * 248 * Some hardware can turn off clipping altogether - in particular any 249 * hardware with a TNL unit can do its own clipping, even if it is 250 * relying on the draw module for some other reason. 251 */ 252void draw_set_driver_clipping( struct draw_context *draw, 253 boolean bypass_clip_xy, 254 boolean bypass_clip_z ) 255{ 256 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 257 258 draw->driver.bypass_clip_xy = bypass_clip_xy; 259 draw->driver.bypass_clip_z = bypass_clip_z; 260 update_clip_flags(draw); 261} 262 263 264/** 265 * Plug in the primitive rendering/rasterization stage (which is the last 266 * stage in the drawing pipeline). 267 * This is provided by the device driver. 268 */ 269void draw_set_rasterize_stage( struct draw_context *draw, 270 struct draw_stage *stage ) 271{ 272 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 273 274 draw->pipeline.rasterize = stage; 275} 276 277 278/** 279 * Set the draw module's clipping state. 280 */ 281void draw_set_clip_state( struct draw_context *draw, 282 const struct pipe_clip_state *clip ) 283{ 284 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 285 286 assert(clip->nr <= PIPE_MAX_CLIP_PLANES); 287 memcpy(&draw->plane[6], clip->ucp, clip->nr * sizeof(clip->ucp[0])); 288 draw->nr_planes = 6 + clip->nr; 289 draw->depth_clamp = clip->depth_clamp; 290 291 update_clip_flags(draw); 292} 293 294 295/** 296 * Set the draw module's viewport state. 297 */ 298void draw_set_viewport_state( struct draw_context *draw, 299 const struct pipe_viewport_state *viewport ) 300{ 301 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 302 draw->viewport = *viewport; /* struct copy */ 303 draw->identity_viewport = (viewport->scale[0] == 1.0f && 304 viewport->scale[1] == 1.0f && 305 viewport->scale[2] == 1.0f && 306 viewport->scale[3] == 1.0f && 307 viewport->translate[0] == 0.0f && 308 viewport->translate[1] == 0.0f && 309 viewport->translate[2] == 0.0f && 310 viewport->translate[3] == 0.0f); 311 312 draw_vs_set_viewport( draw, viewport ); 313} 314 315 316 317void 318draw_set_vertex_buffers(struct draw_context *draw, 319 unsigned count, 320 const struct pipe_vertex_buffer *buffers) 321{ 322 assert(count <= PIPE_MAX_ATTRIBS); 323 324 util_copy_vertex_buffers(draw->pt.vertex_buffer, 325 &draw->pt.nr_vertex_buffers, 326 buffers, count); 327} 328 329 330void 331draw_set_vertex_elements(struct draw_context *draw, 332 unsigned count, 333 const struct pipe_vertex_element *elements) 334{ 335 assert(count <= PIPE_MAX_ATTRIBS); 336 337 memcpy(draw->pt.vertex_element, elements, count * sizeof(elements[0])); 338 draw->pt.nr_vertex_elements = count; 339} 340 341 342/** 343 * Tell drawing context where to find mapped vertex buffers. 344 */ 345void 346draw_set_mapped_vertex_buffer(struct draw_context *draw, 347 unsigned attr, const void *buffer) 348{ 349 draw->pt.user.vbuffer[attr] = buffer; 350} 351 352 353void 354draw_set_mapped_constant_buffer(struct draw_context *draw, 355 unsigned shader_type, 356 unsigned slot, 357 const void *buffer, 358 unsigned size ) 359{ 360 debug_assert(shader_type == PIPE_SHADER_VERTEX || 361 shader_type == PIPE_SHADER_GEOMETRY); 362 debug_assert(slot < PIPE_MAX_CONSTANT_BUFFERS); 363 364 switch (shader_type) { 365 case PIPE_SHADER_VERTEX: 366 draw->pt.user.vs_constants[slot] = buffer; 367 draw->pt.user.vs_constants_size[slot] = size; 368 draw->pt.user.planes = (float (*) [12][4]) &(draw->plane[0]); 369 draw_vs_set_constants(draw, slot, buffer, size); 370 break; 371 case PIPE_SHADER_GEOMETRY: 372 draw->pt.user.gs_constants[slot] = buffer; 373 draw->pt.user.gs_constants_size[slot] = size; 374 draw_gs_set_constants(draw, slot, buffer, size); 375 break; 376 default: 377 assert(0 && "invalid shader type in draw_set_mapped_constant_buffer"); 378 } 379} 380 381 382/** 383 * Tells the draw module to draw points with triangles if their size 384 * is greater than this threshold. 385 */ 386void 387draw_wide_point_threshold(struct draw_context *draw, float threshold) 388{ 389 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 390 draw->pipeline.wide_point_threshold = threshold; 391} 392 393 394/** 395 * Should the draw module handle point->quad conversion for drawing sprites? 396 */ 397void 398draw_wide_point_sprites(struct draw_context *draw, boolean draw_sprite) 399{ 400 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 401 draw->pipeline.wide_point_sprites = draw_sprite; 402} 403 404 405/** 406 * Tells the draw module to draw lines with triangles if their width 407 * is greater than this threshold. 408 */ 409void 410draw_wide_line_threshold(struct draw_context *draw, float threshold) 411{ 412 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 413 draw->pipeline.wide_line_threshold = roundf(threshold); 414} 415 416 417/** 418 * Tells the draw module whether or not to implement line stipple. 419 */ 420void 421draw_enable_line_stipple(struct draw_context *draw, boolean enable) 422{ 423 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 424 draw->pipeline.line_stipple = enable; 425} 426 427 428/** 429 * Tells draw module whether to convert points to quads for sprite mode. 430 */ 431void 432draw_enable_point_sprites(struct draw_context *draw, boolean enable) 433{ 434 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 435 draw->pipeline.point_sprite = enable; 436} 437 438 439void 440draw_set_force_passthrough( struct draw_context *draw, boolean enable ) 441{ 442 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); 443 draw->force_passthrough = enable; 444} 445 446 447 448/** 449 * Allocate an extra vertex/geometry shader vertex attribute. 450 * This is used by some of the optional draw module stages such 451 * as wide_point which may need to allocate additional generic/texcoord 452 * attributes. 453 */ 454int 455draw_alloc_extra_vertex_attrib(struct draw_context *draw, 456 uint semantic_name, uint semantic_index) 457{ 458 const int num_outputs = draw_current_shader_outputs(draw); 459 const int n = draw->extra_shader_outputs.num; 460 461 assert(n < Elements(draw->extra_shader_outputs.semantic_name)); 462 463 draw->extra_shader_outputs.semantic_name[n] = semantic_name; 464 draw->extra_shader_outputs.semantic_index[n] = semantic_index; 465 draw->extra_shader_outputs.slot[n] = num_outputs + n; 466 draw->extra_shader_outputs.num++; 467 468 return draw->extra_shader_outputs.slot[n]; 469} 470 471 472/** 473 * Remove all extra vertex attributes that were allocated with 474 * draw_alloc_extra_vertex_attrib(). 475 */ 476void 477draw_remove_extra_vertex_attribs(struct draw_context *draw) 478{ 479 draw->extra_shader_outputs.num = 0; 480} 481 482 483/** 484 * Ask the draw module for the location/slot of the given vertex attribute in 485 * a post-transformed vertex. 486 * 487 * With this function, drivers that use the draw module should have no reason 488 * to track the current vertex/geometry shader. 489 * 490 * Note that the draw module may sometimes generate vertices with extra 491 * attributes (such as texcoords for AA lines). The driver can call this 492 * function to find those attributes. 493 * 494 * Zero is returned if the attribute is not found since this is 495 * a don't care / undefined situtation. Returning -1 would be a bit more 496 * work for the drivers. 497 */ 498int 499draw_find_shader_output(const struct draw_context *draw, 500 uint semantic_name, uint semantic_index) 501{ 502 const struct draw_vertex_shader *vs = draw->vs.vertex_shader; 503 const struct draw_geometry_shader *gs = draw->gs.geometry_shader; 504 uint i; 505 const struct tgsi_shader_info *info = &vs->info; 506 507 if (gs) 508 info = &gs->info; 509 510 for (i = 0; i < info->num_outputs; i++) { 511 if (info->output_semantic_name[i] == semantic_name && 512 info->output_semantic_index[i] == semantic_index) 513 return i; 514 } 515 516 /* Search the extra vertex attributes */ 517 for (i = 0; i < draw->extra_shader_outputs.num; i++) { 518 if (draw->extra_shader_outputs.semantic_name[i] == semantic_name && 519 draw->extra_shader_outputs.semantic_index[i] == semantic_index) { 520 return draw->extra_shader_outputs.slot[i]; 521 } 522 } 523 524 return 0; 525} 526 527 528/** 529 * Return total number of the shader outputs. This function is similar to 530 * draw_current_shader_outputs() but this function also counts any extra 531 * vertex/geometry output attributes that may be filled in by some draw 532 * stages (such as AA point, AA line). 533 * 534 * If geometry shader is present, its output will be returned, 535 * if not vertex shader is used. 536 */ 537uint 538draw_num_shader_outputs(const struct draw_context *draw) 539{ 540 uint count; 541 542 /* If a geometry shader is present, its outputs go to the 543 * driver, else the vertex shader's outputs. 544 */ 545 if (draw->gs.geometry_shader) 546 count = draw->gs.geometry_shader->info.num_outputs; 547 else 548 count = draw->vs.vertex_shader->info.num_outputs; 549 550 count += draw->extra_shader_outputs.num; 551 552 return count; 553} 554 555 556/** 557 * Provide TGSI sampler objects for vertex/geometry shaders that use 558 * texture fetches. 559 * This might only be used by software drivers for the time being. 560 */ 561void 562draw_texture_samplers(struct draw_context *draw, 563 uint shader, 564 uint num_samplers, 565 struct tgsi_sampler **samplers) 566{ 567 if (shader == PIPE_SHADER_VERTEX) { 568 draw->vs.num_samplers = num_samplers; 569 draw->vs.samplers = samplers; 570 } else { 571 debug_assert(shader == PIPE_SHADER_GEOMETRY); 572 draw->gs.num_samplers = num_samplers; 573 draw->gs.samplers = samplers; 574 } 575} 576 577 578 579 580void draw_set_render( struct draw_context *draw, 581 struct vbuf_render *render ) 582{ 583 draw->render = render; 584} 585 586 587void 588draw_set_index_buffer(struct draw_context *draw, 589 const struct pipe_index_buffer *ib) 590{ 591 if (ib) 592 memcpy(&draw->pt.index_buffer, ib, sizeof(draw->pt.index_buffer)); 593 else 594 memset(&draw->pt.index_buffer, 0, sizeof(draw->pt.index_buffer)); 595} 596 597 598/** 599 * Tell drawing context where to find mapped index/element buffer. 600 */ 601void 602draw_set_mapped_index_buffer(struct draw_context *draw, 603 const void *elements) 604{ 605 draw->pt.user.elts = elements; 606} 607 608 609/* Revamp me please: 610 */ 611void draw_do_flush( struct draw_context *draw, unsigned flags ) 612{ 613 if (!draw->suspend_flushing) 614 { 615 assert(!draw->flushing); /* catch inadvertant recursion */ 616 617 draw->flushing = TRUE; 618 619 draw_pipeline_flush( draw, flags ); 620 621 draw->reduced_prim = ~0; /* is reduced_prim needed any more? */ 622 623 draw->flushing = FALSE; 624 } 625} 626 627 628/** 629 * Return the number of output attributes produced by the geometry 630 * shader, if present. If no geometry shader, return the number of 631 * outputs from the vertex shader. 632 * \sa draw_num_shader_outputs 633 */ 634uint 635draw_current_shader_outputs(const struct draw_context *draw) 636{ 637 if (draw->gs.geometry_shader) 638 return draw->gs.num_gs_outputs; 639 return draw->vs.num_vs_outputs; 640} 641 642 643/** 644 * Return the index of the shader output which will contain the 645 * vertex position. 646 */ 647uint 648draw_current_shader_position_output(const struct draw_context *draw) 649{ 650 if (draw->gs.geometry_shader) 651 return draw->gs.position_output; 652 return draw->vs.position_output; 653} 654 655 656/** 657 * Return a pointer/handle for a driver/CSO rasterizer object which 658 * disabled culling, stippling, unfilled tris, etc. 659 * This is used by some pipeline stages (such as wide_point, aa_line 660 * and aa_point) which convert points/lines into triangles. In those 661 * cases we don't want to accidentally cull the triangles. 662 * 663 * \param scissor should the rasterizer state enable scissoring? 664 * \param flatshade should the rasterizer state use flat shading? 665 * \return rasterizer CSO handle 666 */ 667void * 668draw_get_rasterizer_no_cull( struct draw_context *draw, 669 boolean scissor, 670 boolean flatshade ) 671{ 672 if (!draw->rasterizer_no_cull[scissor][flatshade]) { 673 /* create now */ 674 struct pipe_context *pipe = draw->pipe; 675 struct pipe_rasterizer_state rast; 676 677 memset(&rast, 0, sizeof(rast)); 678 rast.scissor = scissor; 679 rast.flatshade = flatshade; 680 rast.front_ccw = 1; 681 rast.gl_rasterization_rules = draw->rasterizer->gl_rasterization_rules; 682 683 draw->rasterizer_no_cull[scissor][flatshade] = 684 pipe->create_rasterizer_state(pipe, &rast); 685 } 686 return draw->rasterizer_no_cull[scissor][flatshade]; 687} 688 689void 690draw_set_mapped_so_buffers(struct draw_context *draw, 691 void *buffers[PIPE_MAX_SO_BUFFERS], 692 unsigned num_buffers) 693{ 694 int i; 695 696 for (i = 0; i < num_buffers; ++i) { 697 draw->so.buffers[i] = buffers[i]; 698 } 699 draw->so.num_buffers = num_buffers; 700} 701 702void 703draw_set_so_state(struct draw_context *draw, 704 struct pipe_stream_output_state *state) 705{ 706 memcpy(&draw->so.state, 707 state, 708 sizeof(struct pipe_stream_output_state)); 709} 710 711void 712draw_set_sampler_views(struct draw_context *draw, 713 struct pipe_sampler_view **views, 714 unsigned num) 715{ 716 unsigned i; 717 718 debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS); 719 720 for (i = 0; i < num; ++i) 721 draw->sampler_views[i] = views[i]; 722 for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i) 723 draw->sampler_views[i] = NULL; 724 725 draw->num_sampler_views = num; 726} 727 728void 729draw_set_samplers(struct draw_context *draw, 730 struct pipe_sampler_state **samplers, 731 unsigned num) 732{ 733 unsigned i; 734 735 debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS); 736 737 for (i = 0; i < num; ++i) 738 draw->samplers[i] = samplers[i]; 739 for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i) 740 draw->samplers[i] = NULL; 741 742 draw->num_samplers = num; 743 744#ifdef HAVE_LLVM 745 if (draw->llvm) 746 draw_llvm_set_sampler_state(draw); 747#endif 748} 749 750void 751draw_set_mapped_texture(struct draw_context *draw, 752 unsigned sampler_idx, 753 uint32_t width, uint32_t height, uint32_t depth, 754 uint32_t first_level, uint32_t last_level, 755 uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS], 756 uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS], 757 const void *data[PIPE_MAX_TEXTURE_LEVELS]) 758{ 759#ifdef HAVE_LLVM 760 if(draw->llvm) 761 draw_llvm_set_mapped_texture(draw, 762 sampler_idx, 763 width, height, depth, first_level, last_level, 764 row_stride, img_stride, data); 765#endif 766} 767