1 /* 2 * Copyright 2003 VMware, Inc. 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the "Software"), 7 * to deal in the Software without restriction, including without limitation 8 * on the rights to use, copy, modify, merge, publish, distribute, sub 9 * license, and/or sell copies of the Software, and to permit persons to whom 10 * the Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 19 * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 22 * USE OR OTHER DEALINGS IN THE SOFTWARE. 23 * 24 * Authors: 25 * Keith Whitwell <keithw (at) vmware.com> 26 */ 27 28 #include <stdio.h> 29 #include "main/glheader.h" 30 #include "main/context.h" 31 #include "main/execmem.h" 32 #include "swrast/s_chan.h" 33 #include "t_context.h" 34 #include "t_vertex.h" 35 36 #define DBG 0 37 38 /* Build and manage clipspace/ndc/window vertices. 39 */ 40 41 static GLboolean match_fastpath( struct tnl_clipspace *vtx, 42 const struct tnl_clipspace_fastpath *fp) 43 { 44 GLuint j; 45 46 if (vtx->attr_count != fp->attr_count) 47 return GL_FALSE; 48 49 for (j = 0; j < vtx->attr_count; j++) 50 if (vtx->attr[j].format != fp->attr[j].format || 51 vtx->attr[j].inputsize != fp->attr[j].size || 52 vtx->attr[j].vertoffset != fp->attr[j].offset) 53 return GL_FALSE; 54 55 if (fp->match_strides) { 56 if (vtx->vertex_size != fp->vertex_size) 57 return GL_FALSE; 58 59 for (j = 0; j < vtx->attr_count; j++) 60 if (vtx->attr[j].inputstride != fp->attr[j].stride) 61 return GL_FALSE; 62 } 63 64 return GL_TRUE; 65 } 66 67 static GLboolean search_fastpath_emit( struct tnl_clipspace *vtx ) 68 { 69 struct tnl_clipspace_fastpath *fp = vtx->fastpath; 70 71 for ( ; fp ; fp = fp->next) { 72 if (match_fastpath(vtx, fp)) { 73 vtx->emit = fp->func; 74 return GL_TRUE; 75 } 76 } 77 78 return GL_FALSE; 79 } 80 81 void _tnl_register_fastpath( struct tnl_clipspace *vtx, 82 GLboolean match_strides ) 83 { 84 struct tnl_clipspace_fastpath *fastpath = CALLOC_STRUCT(tnl_clipspace_fastpath); 85 GLuint i; 86 87 if (fastpath == NULL) { 88 _mesa_error_no_memory(__func__); 89 return; 90 } 91 92 fastpath->vertex_size = vtx->vertex_size; 93 fastpath->attr_count = vtx->attr_count; 94 fastpath->match_strides = match_strides; 95 fastpath->func = vtx->emit; 96 fastpath->attr = malloc(vtx->attr_count * sizeof(fastpath->attr[0])); 97 98 if (fastpath->attr == NULL) { 99 free(fastpath); 100 _mesa_error_no_memory(__func__); 101 return; 102 } 103 104 for (i = 0; i < vtx->attr_count; i++) { 105 fastpath->attr[i].format = vtx->attr[i].format; 106 fastpath->attr[i].stride = vtx->attr[i].inputstride; 107 fastpath->attr[i].size = vtx->attr[i].inputsize; 108 fastpath->attr[i].offset = vtx->attr[i].vertoffset; 109 } 110 111 fastpath->next = vtx->fastpath; 112 vtx->fastpath = fastpath; 113 } 114 115 116 117 /*********************************************************************** 118 * Build codegen functions or return generic ones: 119 */ 120 static void choose_emit_func( struct gl_context *ctx, GLuint count, GLubyte *dest) 121 { 122 struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb; 123 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 124 struct tnl_clipspace_attr *a = vtx->attr; 125 const GLuint attr_count = vtx->attr_count; 126 GLuint j; 127 128 for (j = 0; j < attr_count; j++) { 129 GLvector4f *vptr = VB->AttribPtr[a[j].attrib]; 130 a[j].inputstride = vptr->stride; 131 a[j].inputsize = vptr->size; 132 a[j].emit = a[j].insert[vptr->size - 1]; /* not always used */ 133 } 134 135 vtx->emit = NULL; 136 137 /* Does this match an existing (hardwired, codegen or known-bad) 138 * fastpath? 139 */ 140 if (search_fastpath_emit(vtx)) { 141 /* Use this result. If it is null, then it is already known 142 * that the current state will fail for codegen and there is no 143 * point trying again. 144 */ 145 } 146 else if (vtx->codegen_emit) { 147 vtx->codegen_emit(ctx); 148 } 149 150 if (!vtx->emit) { 151 _tnl_generate_hardwired_emit(ctx); 152 } 153 154 /* Otherwise use the generic version: 155 */ 156 if (!vtx->emit) 157 vtx->emit = _tnl_generic_emit; 158 159 vtx->emit( ctx, count, dest ); 160 } 161 162 163 164 static void choose_interp_func( struct gl_context *ctx, 165 GLfloat t, 166 GLuint edst, GLuint eout, GLuint ein, 167 GLboolean force_boundary ) 168 { 169 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 170 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL || 171 ctx->Polygon.BackMode != GL_FILL); 172 GLboolean twosided = ctx->Light.Enabled && ctx->Light.Model.TwoSide; 173 174 if (vtx->need_extras && (twosided || unfilled)) { 175 vtx->interp = _tnl_generic_interp_extras; 176 } else { 177 vtx->interp = _tnl_generic_interp; 178 } 179 180 vtx->interp( ctx, t, edst, eout, ein, force_boundary ); 181 } 182 183 184 static void choose_copy_pv_func( struct gl_context *ctx, GLuint edst, GLuint esrc ) 185 { 186 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 187 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL || 188 ctx->Polygon.BackMode != GL_FILL); 189 190 GLboolean twosided = ctx->Light.Enabled && ctx->Light.Model.TwoSide; 191 192 if (vtx->need_extras && (twosided || unfilled)) { 193 vtx->copy_pv = _tnl_generic_copy_pv_extras; 194 } else { 195 vtx->copy_pv = _tnl_generic_copy_pv; 196 } 197 198 vtx->copy_pv( ctx, edst, esrc ); 199 } 200 201 202 /*********************************************************************** 203 * Public entrypoints, mostly dispatch to the above: 204 */ 205 206 207 /* Interpolate between two vertices to produce a third: 208 */ 209 void _tnl_interp( struct gl_context *ctx, 210 GLfloat t, 211 GLuint edst, GLuint eout, GLuint ein, 212 GLboolean force_boundary ) 213 { 214 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 215 vtx->interp( ctx, t, edst, eout, ein, force_boundary ); 216 } 217 218 /* Copy colors from one vertex to another: 219 */ 220 void _tnl_copy_pv( struct gl_context *ctx, GLuint edst, GLuint esrc ) 221 { 222 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 223 vtx->copy_pv( ctx, edst, esrc ); 224 } 225 226 227 /* Extract a named attribute from a hardware vertex. Will have to 228 * reverse any viewport transformation, swizzling or other conversions 229 * which may have been applied: 230 */ 231 void _tnl_get_attr( struct gl_context *ctx, const void *vin, 232 GLenum attr, GLfloat *dest ) 233 { 234 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 235 const struct tnl_clipspace_attr *a = vtx->attr; 236 const GLuint attr_count = vtx->attr_count; 237 GLuint j; 238 239 for (j = 0; j < attr_count; j++) { 240 if (a[j].attrib == attr) { 241 a[j].extract( &a[j], dest, (GLubyte *)vin + a[j].vertoffset ); 242 return; 243 } 244 } 245 246 /* Else return the value from ctx->Current. 247 */ 248 if (attr == _TNL_ATTRIB_POINTSIZE) { 249 /* If the hardware vertex doesn't have point size then use size from 250 * struct gl_context. XXX this will be wrong if drawing attenuated points! 251 */ 252 dest[0] = ctx->Point.Size; 253 } 254 else { 255 memcpy( dest, ctx->Current.Attrib[attr], 4*sizeof(GLfloat)); 256 } 257 } 258 259 260 /* Complementary operation to the above. 261 */ 262 void _tnl_set_attr( struct gl_context *ctx, void *vout, 263 GLenum attr, const GLfloat *src ) 264 { 265 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 266 const struct tnl_clipspace_attr *a = vtx->attr; 267 const GLuint attr_count = vtx->attr_count; 268 GLuint j; 269 270 for (j = 0; j < attr_count; j++) { 271 if (a[j].attrib == attr) { 272 a[j].insert[4-1]( &a[j], (GLubyte *)vout + a[j].vertoffset, src ); 273 return; 274 } 275 } 276 } 277 278 279 void *_tnl_get_vertex( struct gl_context *ctx, GLuint nr ) 280 { 281 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 282 283 return vtx->vertex_buf + nr * vtx->vertex_size; 284 } 285 286 void _tnl_invalidate_vertex_state( struct gl_context *ctx, GLuint new_state ) 287 { 288 /* if two-sided lighting changes or filled/unfilled polygon state changes */ 289 if (new_state & (_NEW_LIGHT | _NEW_POLYGON) ) { 290 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 291 vtx->new_inputs = ~0; 292 vtx->interp = choose_interp_func; 293 vtx->copy_pv = choose_copy_pv_func; 294 } 295 } 296 297 static void invalidate_funcs( struct tnl_clipspace *vtx ) 298 { 299 vtx->emit = choose_emit_func; 300 vtx->interp = choose_interp_func; 301 vtx->copy_pv = choose_copy_pv_func; 302 vtx->new_inputs = ~0; 303 } 304 305 GLuint _tnl_install_attrs( struct gl_context *ctx, const struct tnl_attr_map *map, 306 GLuint nr, const GLfloat *vp, 307 GLuint unpacked_size ) 308 { 309 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 310 GLuint offset = 0; 311 GLuint i, j; 312 313 assert(nr < _TNL_ATTRIB_MAX); 314 assert(nr == 0 || map[0].attrib == VERT_ATTRIB_POS); 315 316 vtx->new_inputs = ~0; 317 vtx->need_viewport = GL_FALSE; 318 319 if (vp) { 320 vtx->need_viewport = GL_TRUE; 321 } 322 323 for (j = 0, i = 0; i < nr; i++) { 324 const GLuint format = map[i].format; 325 if (format == EMIT_PAD) { 326 if (DBG) 327 printf("%d: pad %d, offset %d\n", i, 328 map[i].offset, offset); 329 330 offset += map[i].offset; 331 332 } 333 else { 334 GLuint tmpoffset; 335 336 if (unpacked_size) 337 tmpoffset = map[i].offset; 338 else 339 tmpoffset = offset; 340 341 if (vtx->attr_count != j || 342 vtx->attr[j].attrib != map[i].attrib || 343 vtx->attr[j].format != format || 344 vtx->attr[j].vertoffset != tmpoffset) { 345 invalidate_funcs(vtx); 346 347 vtx->attr[j].attrib = map[i].attrib; 348 vtx->attr[j].format = format; 349 vtx->attr[j].vp = vp; 350 vtx->attr[j].insert = _tnl_format_info[format].insert; 351 vtx->attr[j].extract = _tnl_format_info[format].extract; 352 vtx->attr[j].vertattrsize = _tnl_format_info[format].attrsize; 353 vtx->attr[j].vertoffset = tmpoffset; 354 } 355 356 357 if (DBG) 358 printf("%d: %s, vp %p, offset %d\n", i, 359 _tnl_format_info[format].name, (void *)vp, 360 vtx->attr[j].vertoffset); 361 362 offset += _tnl_format_info[format].attrsize; 363 j++; 364 } 365 } 366 367 vtx->attr_count = j; 368 369 if (unpacked_size) 370 vtx->vertex_size = unpacked_size; 371 else 372 vtx->vertex_size = offset; 373 374 assert(vtx->vertex_size <= vtx->max_vertex_size); 375 return vtx->vertex_size; 376 } 377 378 379 380 void _tnl_invalidate_vertices( struct gl_context *ctx, GLuint newinputs ) 381 { 382 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 383 vtx->new_inputs |= newinputs; 384 } 385 386 387 /* This event has broader use beyond this file - will move elsewhere 388 * and probably invoke a driver callback. 389 */ 390 void _tnl_notify_pipeline_output_change( struct gl_context *ctx ) 391 { 392 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 393 invalidate_funcs(vtx); 394 } 395 396 397 static void adjust_input_ptrs( struct gl_context *ctx, GLint diff) 398 { 399 struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb; 400 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 401 struct tnl_clipspace_attr *a = vtx->attr; 402 const GLuint count = vtx->attr_count; 403 GLuint j; 404 405 diff -= 1; 406 for (j=0; j<count; ++j) { 407 register GLvector4f *vptr = VB->AttribPtr[a->attrib]; 408 (a++)->inputptr += diff*vptr->stride; 409 } 410 } 411 412 static void update_input_ptrs( struct gl_context *ctx, GLuint start ) 413 { 414 struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb; 415 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 416 struct tnl_clipspace_attr *a = vtx->attr; 417 const GLuint count = vtx->attr_count; 418 GLuint j; 419 420 for (j = 0; j < count; j++) { 421 GLvector4f *vptr = VB->AttribPtr[a[j].attrib]; 422 423 if (vtx->emit != choose_emit_func) { 424 assert(a[j].inputstride == vptr->stride); 425 assert(a[j].inputsize == vptr->size); 426 } 427 428 a[j].inputptr = ((GLubyte *)vptr->data) + start * vptr->stride; 429 } 430 431 if (a->vp) { 432 vtx->vp_scale[0] = a->vp[MAT_SX]; 433 vtx->vp_scale[1] = a->vp[MAT_SY]; 434 vtx->vp_scale[2] = a->vp[MAT_SZ]; 435 vtx->vp_scale[3] = 1.0; 436 vtx->vp_xlate[0] = a->vp[MAT_TX]; 437 vtx->vp_xlate[1] = a->vp[MAT_TY]; 438 vtx->vp_xlate[2] = a->vp[MAT_TZ]; 439 vtx->vp_xlate[3] = 0.0; 440 } 441 } 442 443 444 void _tnl_build_vertices( struct gl_context *ctx, 445 GLuint start, 446 GLuint end, 447 GLuint newinputs ) 448 { 449 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 450 update_input_ptrs( ctx, start ); 451 vtx->emit( ctx, end - start, 452 (GLubyte *)(vtx->vertex_buf + 453 start * vtx->vertex_size)); 454 } 455 456 /* Emit VB vertices start..end to dest. Note that VB vertex at 457 * postion start will be emitted to dest at position zero. 458 */ 459 void *_tnl_emit_vertices_to_buffer( struct gl_context *ctx, 460 GLuint start, 461 GLuint end, 462 void *dest ) 463 { 464 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 465 466 update_input_ptrs(ctx, start); 467 /* Note: dest should not be adjusted for non-zero 'start' values: 468 */ 469 vtx->emit( ctx, end - start, (GLubyte*) dest ); 470 return (void *)((GLubyte *)dest + vtx->vertex_size * (end - start)); 471 } 472 473 /* Emit indexed VB vertices start..end to dest. Note that VB vertex at 474 * postion start will be emitted to dest at position zero. 475 */ 476 477 void *_tnl_emit_indexed_vertices_to_buffer( struct gl_context *ctx, 478 const GLuint *elts, 479 GLuint start, 480 GLuint end, 481 void *dest ) 482 { 483 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 484 GLuint oldIndex; 485 GLubyte *cdest = dest; 486 487 update_input_ptrs(ctx, oldIndex = elts[start++]); 488 vtx->emit( ctx, 1, cdest ); 489 cdest += vtx->vertex_size; 490 491 for (; start < end; ++start) { 492 adjust_input_ptrs(ctx, elts[start] - oldIndex); 493 oldIndex = elts[start]; 494 vtx->emit( ctx, 1, cdest); 495 cdest += vtx->vertex_size; 496 } 497 498 return (void *) cdest; 499 } 500 501 502 void _tnl_init_vertices( struct gl_context *ctx, 503 GLuint vb_size, 504 GLuint max_vertex_size ) 505 { 506 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 507 508 _tnl_install_attrs( ctx, NULL, 0, NULL, 0 ); 509 510 vtx->need_extras = GL_TRUE; 511 if (max_vertex_size > vtx->max_vertex_size) { 512 _tnl_free_vertices( ctx ); 513 vtx->max_vertex_size = max_vertex_size; 514 vtx->vertex_buf = _mesa_align_calloc(vb_size * max_vertex_size, 32 ); 515 invalidate_funcs(vtx); 516 } 517 518 switch(CHAN_TYPE) { 519 case GL_UNSIGNED_BYTE: 520 vtx->chan_scale[0] = 255.0; 521 vtx->chan_scale[1] = 255.0; 522 vtx->chan_scale[2] = 255.0; 523 vtx->chan_scale[3] = 255.0; 524 break; 525 case GL_UNSIGNED_SHORT: 526 vtx->chan_scale[0] = 65535.0; 527 vtx->chan_scale[1] = 65535.0; 528 vtx->chan_scale[2] = 65535.0; 529 vtx->chan_scale[3] = 65535.0; 530 break; 531 default: 532 vtx->chan_scale[0] = 1.0; 533 vtx->chan_scale[1] = 1.0; 534 vtx->chan_scale[2] = 1.0; 535 vtx->chan_scale[3] = 1.0; 536 break; 537 } 538 539 vtx->identity[0] = 0.0; 540 vtx->identity[1] = 0.0; 541 vtx->identity[2] = 0.0; 542 vtx->identity[3] = 1.0; 543 544 vtx->codegen_emit = NULL; 545 546 #ifdef USE_SSE_ASM 547 if (!getenv("MESA_NO_CODEGEN")) 548 vtx->codegen_emit = _tnl_generate_sse_emit; 549 #endif 550 } 551 552 553 void _tnl_free_vertices( struct gl_context *ctx ) 554 { 555 TNLcontext *tnl = TNL_CONTEXT(ctx); 556 if (tnl) { 557 struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx); 558 struct tnl_clipspace_fastpath *fp, *tmp; 559 560 _mesa_align_free(vtx->vertex_buf); 561 vtx->vertex_buf = NULL; 562 563 for (fp = vtx->fastpath ; fp ; fp = tmp) { 564 tmp = fp->next; 565 free(fp->attr); 566 567 /* KW: At the moment, fp->func is constrained to be allocated by 568 * _mesa_exec_alloc(), as the hardwired fastpaths in 569 * t_vertex_generic.c are handled specially. It would be nice 570 * to unify them, but this probably won't change until this 571 * module gets another overhaul. 572 */ 573 _mesa_exec_free((void *) fp->func); 574 free(fp); 575 } 576 577 vtx->fastpath = NULL; 578 } 579 } 580