t_vp_build.c revision 501ee87180047dd04afc69103c31e1eae5374bf1
1/* 2 * Mesa 3-D graphics library 3 * Version: 6.5 4 * 5 * Copyright (C) 2006 Tungsten Graphics All Rights Reserved. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the "Software"), 9 * to deal in the Software without restriction, including without limitation 10 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 11 * and/or sell copies of the Software, and to permit persons to whom the 12 * Software is furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included 15 * in all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * TUNGSTEN GRAPHICS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, 21 * WHETHER IN 22 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 23 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 24 */ 25 26/** 27 * \file t_vp_build.c 28 * Create a vertex program to execute the current fixed function T&L pipeline. 29 * \author Keith Whitwell 30 */ 31 32 33#include "glheader.h" 34#include "macros.h" 35#include "enums.h" 36#include "program.h" 37#include "prog_instruction.h" 38#include "prog_parameter.h" 39#include "prog_print.h" 40#include "prog_statevars.h" 41#include "t_context.h" /* NOTE: very light dependency on this */ 42#include "t_vp_build.h" 43 44 45struct state_key { 46 unsigned light_global_enabled:1; 47 unsigned light_local_viewer:1; 48 unsigned light_twoside:1; 49 unsigned light_color_material:1; 50 unsigned light_color_material_mask:12; 51 unsigned light_material_mask:12; 52 53 unsigned normalize:1; 54 unsigned rescale_normals:1; 55 unsigned fog_source_is_depth:1; 56 unsigned tnl_do_vertex_fog:1; 57 unsigned separate_specular:1; 58 unsigned fog_mode:2; 59 unsigned point_attenuated:1; 60 unsigned texture_enabled_global:1; 61 unsigned fragprog_inputs_read:12; 62 63 struct { 64 unsigned light_enabled:1; 65 unsigned light_eyepos3_is_zero:1; 66 unsigned light_spotcutoff_is_180:1; 67 unsigned light_attenuated:1; 68 unsigned texunit_really_enabled:1; 69 unsigned texmat_enabled:1; 70 unsigned texgen_enabled:4; 71 unsigned texgen_mode0:4; 72 unsigned texgen_mode1:4; 73 unsigned texgen_mode2:4; 74 unsigned texgen_mode3:4; 75 } unit[8]; 76}; 77 78 79 80#define FOG_NONE 0 81#define FOG_LINEAR 1 82#define FOG_EXP 2 83#define FOG_EXP2 3 84 85static GLuint translate_fog_mode( GLenum mode ) 86{ 87 switch (mode) { 88 case GL_LINEAR: return FOG_LINEAR; 89 case GL_EXP: return FOG_EXP; 90 case GL_EXP2: return FOG_EXP2; 91 default: return FOG_NONE; 92 } 93} 94 95#define TXG_NONE 0 96#define TXG_OBJ_LINEAR 1 97#define TXG_EYE_LINEAR 2 98#define TXG_SPHERE_MAP 3 99#define TXG_REFLECTION_MAP 4 100#define TXG_NORMAL_MAP 5 101 102static GLuint translate_texgen( GLboolean enabled, GLenum mode ) 103{ 104 if (!enabled) 105 return TXG_NONE; 106 107 switch (mode) { 108 case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR; 109 case GL_EYE_LINEAR: return TXG_EYE_LINEAR; 110 case GL_SPHERE_MAP: return TXG_SPHERE_MAP; 111 case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP; 112 case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP; 113 default: return TXG_NONE; 114 } 115} 116 117static struct state_key *make_state_key( GLcontext *ctx ) 118{ 119 TNLcontext *tnl = TNL_CONTEXT(ctx); 120 struct vertex_buffer *VB = &tnl->vb; 121 const struct gl_fragment_program *fp = ctx->FragmentProgram._Current; 122 struct state_key *key = CALLOC_STRUCT(state_key); 123 GLuint i; 124 125 /* This now relies on texenvprogram.c being active: 126 */ 127 assert(fp); 128 129 key->fragprog_inputs_read = fp->Base.InputsRead; 130 131 key->separate_specular = (ctx->Light.Model.ColorControl == 132 GL_SEPARATE_SPECULAR_COLOR); 133 134 if (ctx->Light.Enabled) { 135 key->light_global_enabled = 1; 136 137 if (ctx->Light.Model.LocalViewer) 138 key->light_local_viewer = 1; 139 140 if (ctx->Light.Model.TwoSide) 141 key->light_twoside = 1; 142 143 if (ctx->Light.ColorMaterialEnabled) { 144 key->light_color_material = 1; 145 key->light_color_material_mask = ctx->Light.ColorMaterialBitmask; 146 } 147 148 for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) 149 if (VB->AttribPtr[i]->stride) 150 key->light_material_mask |= 1<<(i-_TNL_ATTRIB_MAT_FRONT_AMBIENT); 151 152 for (i = 0; i < MAX_LIGHTS; i++) { 153 struct gl_light *light = &ctx->Light.Light[i]; 154 155 if (light->Enabled) { 156 key->unit[i].light_enabled = 1; 157 158 if (light->EyePosition[3] == 0.0) 159 key->unit[i].light_eyepos3_is_zero = 1; 160 161 if (light->SpotCutoff == 180.0) 162 key->unit[i].light_spotcutoff_is_180 = 1; 163 164 if (light->ConstantAttenuation != 1.0 || 165 light->LinearAttenuation != 0.0 || 166 light->QuadraticAttenuation != 0.0) 167 key->unit[i].light_attenuated = 1; 168 } 169 } 170 } 171 172 if (ctx->Transform.Normalize) 173 key->normalize = 1; 174 175 if (ctx->Transform.RescaleNormals) 176 key->rescale_normals = 1; 177 178 key->fog_mode = translate_fog_mode(fp->FogOption); 179 180 if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT) 181 key->fog_source_is_depth = 1; 182 183 if (tnl->_DoVertexFog) 184 key->tnl_do_vertex_fog = 1; 185 186 if (ctx->Point._Attenuated) 187 key->point_attenuated = 1; 188 189 if (ctx->Texture._TexGenEnabled || 190 ctx->Texture._TexMatEnabled || 191 ctx->Texture._EnabledUnits) 192 key->texture_enabled_global = 1; 193 194 for (i = 0; i < MAX_TEXTURE_UNITS; i++) { 195 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i]; 196 197 if (texUnit->_ReallyEnabled) 198 key->unit[i].texunit_really_enabled = 1; 199 200 if (ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i)) 201 key->unit[i].texmat_enabled = 1; 202 203 if (texUnit->TexGenEnabled) { 204 key->unit[i].texgen_enabled = 1; 205 206 key->unit[i].texgen_mode0 = 207 translate_texgen( texUnit->TexGenEnabled & (1<<0), 208 texUnit->GenModeS ); 209 key->unit[i].texgen_mode1 = 210 translate_texgen( texUnit->TexGenEnabled & (1<<1), 211 texUnit->GenModeT ); 212 key->unit[i].texgen_mode2 = 213 translate_texgen( texUnit->TexGenEnabled & (1<<2), 214 texUnit->GenModeR ); 215 key->unit[i].texgen_mode3 = 216 translate_texgen( texUnit->TexGenEnabled & (1<<3), 217 texUnit->GenModeQ ); 218 } 219 } 220 221 return key; 222} 223 224 225 226/* Very useful debugging tool - produces annotated listing of 227 * generated program with line/function references for each 228 * instruction back into this file: 229 */ 230#define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM) 231 232/* Should be tunable by the driver - do we want to do matrix 233 * multiplications with DP4's or with MUL/MAD's? SSE works better 234 * with the latter, drivers may differ. 235 */ 236#define PREFER_DP4 0 237 238#define MAX_INSN 256 239 240/* Use uregs to represent registers internally, translate to Mesa's 241 * expected formats on emit. 242 * 243 * NOTE: These are passed by value extensively in this file rather 244 * than as usual by pointer reference. If this disturbs you, try 245 * remembering they are just 32bits in size. 246 * 247 * GCC is smart enough to deal with these dword-sized structures in 248 * much the same way as if I had defined them as dwords and was using 249 * macros to access and set the fields. This is much nicer and easier 250 * to evolve. 251 */ 252struct ureg { 253 GLuint file:4; 254 GLint idx:8; /* relative addressing may be negative */ 255 GLuint negate:1; 256 GLuint swz:12; 257 GLuint pad:7; 258}; 259 260 261struct tnl_program { 262 const struct state_key *state; 263 struct gl_vertex_program *program; 264 265 GLuint temp_in_use; 266 GLuint temp_reserved; 267 268 struct ureg eye_position; 269 struct ureg eye_position_normalized; 270 struct ureg eye_normal; 271 struct ureg identity; 272 273 GLuint materials; 274 GLuint color_materials; 275}; 276 277 278static const struct ureg undef = { 279 PROGRAM_UNDEFINED, 280 ~0, 281 0, 282 0, 283 0 284}; 285 286/* Local shorthand: 287 */ 288#define X SWIZZLE_X 289#define Y SWIZZLE_Y 290#define Z SWIZZLE_Z 291#define W SWIZZLE_W 292 293 294/* Construct a ureg: 295 */ 296static struct ureg make_ureg(GLuint file, GLint idx) 297{ 298 struct ureg reg; 299 reg.file = file; 300 reg.idx = idx; 301 reg.negate = 0; 302 reg.swz = SWIZZLE_NOOP; 303 reg.pad = 0; 304 return reg; 305} 306 307 308 309static struct ureg negate( struct ureg reg ) 310{ 311 reg.negate ^= 1; 312 return reg; 313} 314 315 316static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w ) 317{ 318 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x), 319 GET_SWZ(reg.swz, y), 320 GET_SWZ(reg.swz, z), 321 GET_SWZ(reg.swz, w)); 322 323 return reg; 324} 325 326static struct ureg swizzle1( struct ureg reg, int x ) 327{ 328 return swizzle(reg, x, x, x, x); 329} 330 331static struct ureg get_temp( struct tnl_program *p ) 332{ 333 int bit = _mesa_ffs( ~p->temp_in_use ); 334 if (!bit) { 335 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__); 336 _mesa_exit(1); 337 } 338 339 if ((GLuint) bit > p->program->Base.NumTemporaries) 340 p->program->Base.NumTemporaries = bit; 341 342 p->temp_in_use |= 1<<(bit-1); 343 return make_ureg(PROGRAM_TEMPORARY, bit-1); 344} 345 346static struct ureg reserve_temp( struct tnl_program *p ) 347{ 348 struct ureg temp = get_temp( p ); 349 p->temp_reserved |= 1<<temp.idx; 350 return temp; 351} 352 353static void release_temp( struct tnl_program *p, struct ureg reg ) 354{ 355 if (reg.file == PROGRAM_TEMPORARY) { 356 p->temp_in_use &= ~(1<<reg.idx); 357 p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */ 358 } 359} 360 361static void release_temps( struct tnl_program *p ) 362{ 363 p->temp_in_use = p->temp_reserved; 364} 365 366 367 368static struct ureg register_input( struct tnl_program *p, GLuint input ) 369{ 370 p->program->Base.InputsRead |= (1<<input); 371 return make_ureg(PROGRAM_INPUT, input); 372} 373 374static struct ureg register_output( struct tnl_program *p, GLuint output ) 375{ 376 p->program->Base.OutputsWritten |= (1<<output); 377 return make_ureg(PROGRAM_OUTPUT, output); 378} 379 380static struct ureg register_const4f( struct tnl_program *p, 381 GLfloat s0, 382 GLfloat s1, 383 GLfloat s2, 384 GLfloat s3) 385{ 386 GLfloat values[4]; 387 GLint idx; 388 GLuint swizzle; 389 values[0] = s0; 390 values[1] = s1; 391 values[2] = s2; 392 values[3] = s3; 393 idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4, 394 &swizzle ); 395 ASSERT(swizzle == SWIZZLE_NOOP); 396 return make_ureg(PROGRAM_STATE_VAR, idx); 397} 398 399#define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1) 400#define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0) 401#define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1) 402#define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1) 403 404static GLboolean is_undef( struct ureg reg ) 405{ 406 return reg.file == PROGRAM_UNDEFINED; 407} 408 409static struct ureg get_identity_param( struct tnl_program *p ) 410{ 411 if (is_undef(p->identity)) 412 p->identity = register_const4f(p, 0,0,0,1); 413 414 return p->identity; 415} 416 417static struct ureg register_param6( struct tnl_program *p, 418 GLint s0, 419 GLint s1, 420 GLint s2, 421 GLint s3, 422 GLint s4, 423 GLint s5) 424{ 425 GLint tokens[6]; 426 GLint idx; 427 tokens[0] = s0; 428 tokens[1] = s1; 429 tokens[2] = s2; 430 tokens[3] = s3; 431 tokens[4] = s4; 432 tokens[5] = s5; 433 idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens ); 434 return make_ureg(PROGRAM_STATE_VAR, idx); 435} 436 437 438#define register_param1(p,s0) register_param6(p,s0,0,0,0,0,0) 439#define register_param2(p,s0,s1) register_param6(p,s0,s1,0,0,0,0) 440#define register_param3(p,s0,s1,s2) register_param6(p,s0,s1,s2,0,0,0) 441#define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0) 442 443 444static void register_matrix_param6( struct tnl_program *p, 445 GLint s0, 446 GLint s1, 447 GLint s2, 448 GLint s3, 449 GLint s4, 450 GLint s5, 451 struct ureg *matrix ) 452{ 453 GLint i; 454 455 /* This is a bit sad as the support is there to pull the whole 456 * matrix out in one go: 457 */ 458 for (i = 0; i <= s4 - s3; i++) 459 matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 ); 460} 461 462 463static void emit_arg( struct prog_src_register *src, 464 struct ureg reg ) 465{ 466 src->File = reg.file; 467 src->Index = reg.idx; 468 src->Swizzle = reg.swz; 469 src->NegateBase = reg.negate ? NEGATE_XYZW : 0; 470 src->Abs = 0; 471 src->NegateAbs = 0; 472 src->RelAddr = 0; 473} 474 475static void emit_dst( struct prog_dst_register *dst, 476 struct ureg reg, GLuint mask ) 477{ 478 dst->File = reg.file; 479 dst->Index = reg.idx; 480 /* allow zero as a shorthand for xyzw */ 481 dst->WriteMask = mask ? mask : WRITEMASK_XYZW; 482 dst->CondMask = COND_TR; 483 dst->CondSwizzle = 0; 484 dst->CondSrc = 0; 485 dst->pad = 0; 486} 487 488static void debug_insn( struct prog_instruction *inst, const char *fn, 489 GLuint line ) 490{ 491 if (DISASSEM) { 492 static const char *last_fn; 493 494 if (fn != last_fn) { 495 last_fn = fn; 496 _mesa_printf("%s:\n", fn); 497 } 498 499 _mesa_printf("%d:\t", line); 500 _mesa_print_instruction(inst); 501 } 502} 503 504 505static void emit_op3fn(struct tnl_program *p, 506 GLuint op, 507 struct ureg dest, 508 GLuint mask, 509 struct ureg src0, 510 struct ureg src1, 511 struct ureg src2, 512 const char *fn, 513 GLuint line) 514{ 515 GLuint nr = p->program->Base.NumInstructions++; 516 struct prog_instruction *inst = &p->program->Base.Instructions[nr]; 517 518 if (p->program->Base.NumInstructions > MAX_INSN) { 519 _mesa_problem(0, "Out of instructions in emit_op3fn\n"); 520 return; 521 } 522 523 inst->Opcode = (enum prog_opcode) op; 524 inst->StringPos = 0; 525 inst->Data = 0; 526 527 emit_arg( &inst->SrcReg[0], src0 ); 528 emit_arg( &inst->SrcReg[1], src1 ); 529 emit_arg( &inst->SrcReg[2], src2 ); 530 531 emit_dst( &inst->DstReg, dest, mask ); 532 533 debug_insn(inst, fn, line); 534} 535 536 537#define emit_op3(p, op, dst, mask, src0, src1, src2) \ 538 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__) 539 540#define emit_op2(p, op, dst, mask, src0, src1) \ 541 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__) 542 543#define emit_op1(p, op, dst, mask, src0) \ 544 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__) 545 546 547static struct ureg make_temp( struct tnl_program *p, struct ureg reg ) 548{ 549 if (reg.file == PROGRAM_TEMPORARY && 550 !(p->temp_reserved & (1<<reg.idx))) 551 return reg; 552 else { 553 struct ureg temp = get_temp(p); 554 emit_op1(p, OPCODE_MOV, temp, 0, reg); 555 return temp; 556 } 557} 558 559 560/* Currently no tracking performed of input/output/register size or 561 * active elements. Could be used to reduce these operations, as 562 * could the matrix type. 563 */ 564static void emit_matrix_transform_vec4( struct tnl_program *p, 565 struct ureg dest, 566 const struct ureg *mat, 567 struct ureg src) 568{ 569 emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]); 570 emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]); 571 emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]); 572 emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]); 573} 574 575/* This version is much easier to implement if writemasks are not 576 * supported natively on the target or (like SSE), the target doesn't 577 * have a clean/obvious dotproduct implementation. 578 */ 579static void emit_transpose_matrix_transform_vec4( struct tnl_program *p, 580 struct ureg dest, 581 const struct ureg *mat, 582 struct ureg src) 583{ 584 struct ureg tmp; 585 586 if (dest.file != PROGRAM_TEMPORARY) 587 tmp = get_temp(p); 588 else 589 tmp = dest; 590 591 emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]); 592 emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp); 593 emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp); 594 emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp); 595 596 if (dest.file != PROGRAM_TEMPORARY) 597 release_temp(p, tmp); 598} 599 600static void emit_matrix_transform_vec3( struct tnl_program *p, 601 struct ureg dest, 602 const struct ureg *mat, 603 struct ureg src) 604{ 605 emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]); 606 emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]); 607 emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]); 608} 609 610 611static void emit_normalize_vec3( struct tnl_program *p, 612 struct ureg dest, 613 struct ureg src ) 614{ 615 struct ureg tmp = get_temp(p); 616 emit_op2(p, OPCODE_DP3, tmp, 0, src, src); 617 emit_op1(p, OPCODE_RSQ, tmp, 0, tmp); 618 emit_op2(p, OPCODE_MUL, dest, 0, src, tmp); 619 release_temp(p, tmp); 620} 621 622static void emit_passthrough( struct tnl_program *p, 623 GLuint input, 624 GLuint output ) 625{ 626 struct ureg out = register_output(p, output); 627 emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input)); 628} 629 630static struct ureg get_eye_position( struct tnl_program *p ) 631{ 632 if (is_undef(p->eye_position)) { 633 struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 634 struct ureg modelview[4]; 635 636 p->eye_position = reserve_temp(p); 637 638 if (PREFER_DP4) { 639 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3, 640 STATE_MATRIX, modelview ); 641 642 emit_matrix_transform_vec4(p, p->eye_position, modelview, pos); 643 } 644 else { 645 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3, 646 STATE_MATRIX_TRANSPOSE, modelview ); 647 648 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos); 649 } 650 } 651 652 return p->eye_position; 653} 654 655 656static struct ureg get_eye_position_normalized( struct tnl_program *p ) 657{ 658 if (is_undef(p->eye_position_normalized)) { 659 struct ureg eye = get_eye_position(p); 660 p->eye_position_normalized = reserve_temp(p); 661 emit_normalize_vec3(p, p->eye_position_normalized, eye); 662 } 663 664 return p->eye_position_normalized; 665} 666 667 668static struct ureg get_eye_normal( struct tnl_program *p ) 669{ 670 if (is_undef(p->eye_normal)) { 671 struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL ); 672 struct ureg mvinv[3]; 673 674 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 2, 675 STATE_MATRIX_INVTRANS, mvinv ); 676 677 p->eye_normal = reserve_temp(p); 678 679 /* Transform to eye space: 680 */ 681 emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal ); 682 683 /* Normalize/Rescale: 684 */ 685 if (p->state->normalize) { 686 emit_normalize_vec3( p, p->eye_normal, p->eye_normal ); 687 } 688 else if (p->state->rescale_normals) { 689 struct ureg rescale = register_param2(p, STATE_INTERNAL, 690 STATE_NORMAL_SCALE); 691 692 emit_op2( p, OPCODE_MUL, p->eye_normal, 0, normal, 693 swizzle1(rescale, X)); 694 } 695 } 696 697 return p->eye_normal; 698} 699 700 701 702static void build_hpos( struct tnl_program *p ) 703{ 704 struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 705 struct ureg hpos = register_output( p, VERT_RESULT_HPOS ); 706 struct ureg mvp[4]; 707 708 if (PREFER_DP4) { 709 register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3, 710 STATE_MATRIX, mvp ); 711 emit_matrix_transform_vec4( p, hpos, mvp, pos ); 712 } 713 else { 714 register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3, 715 STATE_MATRIX_TRANSPOSE, mvp ); 716 emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos ); 717 } 718} 719 720 721static GLuint material_attrib( GLuint side, GLuint property ) 722{ 723 return ((property - STATE_AMBIENT) * 2 + 724 side); 725} 726 727/* Get a bitmask of which material values vary on a per-vertex basis. 728 */ 729static void set_material_flags( struct tnl_program *p ) 730{ 731 p->color_materials = 0; 732 p->materials = 0; 733 734 if (p->state->light_color_material) { 735 p->materials = 736 p->color_materials = p->state->light_color_material_mask; 737 } 738 739 p->materials |= p->state->light_material_mask; 740} 741 742 743static struct ureg get_material( struct tnl_program *p, GLuint side, 744 GLuint property ) 745{ 746 GLuint attrib = material_attrib(side, property); 747 748 if (p->color_materials & (1<<attrib)) 749 return register_input(p, VERT_ATTRIB_COLOR0); 750 else if (p->materials & (1<<attrib)) 751 return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT ); 752 else 753 return register_param3( p, STATE_MATERIAL, side, property ); 754} 755 756#define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \ 757 MAT_BIT_FRONT_AMBIENT | \ 758 MAT_BIT_FRONT_DIFFUSE) << (side)) 759 760/* Either return a precalculated constant value or emit code to 761 * calculate these values dynamically in the case where material calls 762 * are present between begin/end pairs. 763 * 764 * Probably want to shift this to the program compilation phase - if 765 * we always emitted the calculation here, a smart compiler could 766 * detect that it was constant (given a certain set of inputs), and 767 * lift it out of the main loop. That way the programs created here 768 * would be independent of the vertex_buffer details. 769 */ 770static struct ureg get_scenecolor( struct tnl_program *p, GLuint side ) 771{ 772 if (p->materials & SCENE_COLOR_BITS(side)) { 773 struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT); 774 struct ureg material_emission = get_material(p, side, STATE_EMISSION); 775 struct ureg material_ambient = get_material(p, side, STATE_AMBIENT); 776 struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE); 777 struct ureg tmp = make_temp(p, material_diffuse); 778 emit_op3(p, OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient, 779 material_ambient, material_emission); 780 return tmp; 781 } 782 else 783 return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side ); 784} 785 786 787static struct ureg get_lightprod( struct tnl_program *p, GLuint light, 788 GLuint side, GLuint property ) 789{ 790 GLuint attrib = material_attrib(side, property); 791 if (p->materials & (1<<attrib)) { 792 struct ureg light_value = 793 register_param3(p, STATE_LIGHT, light, property); 794 struct ureg material_value = get_material(p, side, property); 795 struct ureg tmp = get_temp(p); 796 emit_op2(p, OPCODE_MUL, tmp, 0, light_value, material_value); 797 return tmp; 798 } 799 else 800 return register_param4(p, STATE_LIGHTPROD, light, side, property); 801} 802 803static struct ureg calculate_light_attenuation( struct tnl_program *p, 804 GLuint i, 805 struct ureg VPpli, 806 struct ureg dist ) 807{ 808 struct ureg attenuation = register_param3(p, STATE_LIGHT, i, 809 STATE_ATTENUATION); 810 struct ureg att = get_temp(p); 811 812 /* Calculate spot attenuation: 813 */ 814 if (!p->state->unit[i].light_spotcutoff_is_180) { 815 struct ureg spot_dir = register_param3(p, STATE_LIGHT, i, 816 STATE_SPOT_DIRECTION); 817 struct ureg spot = get_temp(p); 818 struct ureg slt = get_temp(p); 819 820 emit_normalize_vec3( p, spot, spot_dir ); /* XXX: precompute! */ 821 emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot); 822 emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir,W), spot); 823 emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W)); 824 emit_op2(p, OPCODE_MUL, att, 0, slt, spot); 825 826 release_temp(p, spot); 827 release_temp(p, slt); 828 } 829 830 /* Calculate distance attenuation: 831 */ 832 if (p->state->unit[i].light_attenuated) { 833 834 /* 1/d,d,d,1/d */ 835 emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist); 836 /* 1,d,d*d,1/d */ 837 emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y)); 838 /* 1/dist-atten */ 839 emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist); 840 841 if (!p->state->unit[i].light_spotcutoff_is_180) { 842 /* dist-atten */ 843 emit_op1(p, OPCODE_RCP, dist, 0, dist); 844 /* spot-atten * dist-atten */ 845 emit_op2(p, OPCODE_MUL, att, 0, dist, att); 846 } else { 847 /* dist-atten */ 848 emit_op1(p, OPCODE_RCP, att, 0, dist); 849 } 850 } 851 852 return att; 853} 854 855 856 857 858 859/* Need to add some addtional parameters to allow lighting in object 860 * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye 861 * space lighting. 862 */ 863static void build_lighting( struct tnl_program *p ) 864{ 865 const GLboolean twoside = p->state->light_twoside; 866 const GLboolean separate = p->state->separate_specular; 867 GLuint nr_lights = 0, count = 0; 868 struct ureg normal = get_eye_normal(p); 869 struct ureg lit = get_temp(p); 870 struct ureg dots = get_temp(p); 871 struct ureg _col0 = undef, _col1 = undef; 872 struct ureg _bfc0 = undef, _bfc1 = undef; 873 GLuint i; 874 875 for (i = 0; i < MAX_LIGHTS; i++) 876 if (p->state->unit[i].light_enabled) 877 nr_lights++; 878 879 set_material_flags(p); 880 881 { 882 struct ureg shininess = get_material(p, 0, STATE_SHININESS); 883 emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X)); 884 release_temp(p, shininess); 885 886 _col0 = make_temp(p, get_scenecolor(p, 0)); 887 if (separate) 888 _col1 = make_temp(p, get_identity_param(p)); 889 else 890 _col1 = _col0; 891 892 } 893 894 if (twoside) { 895 struct ureg shininess = get_material(p, 1, STATE_SHININESS); 896 emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z, 897 negate(swizzle1(shininess,X))); 898 release_temp(p, shininess); 899 900 _bfc0 = make_temp(p, get_scenecolor(p, 1)); 901 if (separate) 902 _bfc1 = make_temp(p, get_identity_param(p)); 903 else 904 _bfc1 = _bfc0; 905 } 906 907 908 /* If no lights, still need to emit the scenecolor. 909 */ 910 { 911 struct ureg res0 = register_output( p, VERT_RESULT_COL0 ); 912 emit_op1(p, OPCODE_MOV, res0, 0, _col0); 913 } 914 915 if (separate) { 916 struct ureg res1 = register_output( p, VERT_RESULT_COL1 ); 917 emit_op1(p, OPCODE_MOV, res1, 0, _col1); 918 } 919 920 if (twoside) { 921 struct ureg res0 = register_output( p, VERT_RESULT_BFC0 ); 922 emit_op1(p, OPCODE_MOV, res0, 0, _bfc0); 923 } 924 925 if (twoside && separate) { 926 struct ureg res1 = register_output( p, VERT_RESULT_BFC1 ); 927 emit_op1(p, OPCODE_MOV, res1, 0, _bfc1); 928 } 929 930 if (nr_lights == 0) { 931 release_temps(p); 932 return; 933 } 934 935 936 for (i = 0; i < MAX_LIGHTS; i++) { 937 if (p->state->unit[i].light_enabled) { 938 struct ureg half = undef; 939 struct ureg att = undef, VPpli = undef; 940 941 count++; 942 943 if (p->state->unit[i].light_eyepos3_is_zero) { 944 /* Can used precomputed constants in this case. 945 * Attenuation never applies to infinite lights. 946 */ 947 VPpli = register_param3(p, STATE_LIGHT, i, 948 STATE_POSITION_NORMALIZED); 949 half = register_param3(p, STATE_LIGHT, i, STATE_HALF); 950 } 951 else { 952 struct ureg Ppli = register_param3(p, STATE_LIGHT, i, 953 STATE_POSITION); 954 struct ureg V = get_eye_position(p); 955 struct ureg dist = get_temp(p); 956 957 VPpli = get_temp(p); 958 half = get_temp(p); 959 960 /* Calulate VPpli vector 961 */ 962 emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V); 963 964 /* Normalize VPpli. The dist value also used in 965 * attenuation below. 966 */ 967 emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli); 968 emit_op1(p, OPCODE_RSQ, dist, 0, dist); 969 emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist); 970 971 972 /* Calculate attenuation: 973 */ 974 if (!p->state->unit[i].light_spotcutoff_is_180 || 975 p->state->unit[i].light_attenuated) { 976 att = calculate_light_attenuation(p, i, VPpli, dist); 977 } 978 979 980 /* Calculate viewer direction, or use infinite viewer: 981 */ 982 if (p->state->light_local_viewer) { 983 struct ureg eye_hat = get_eye_position_normalized(p); 984 emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); 985 } 986 else { 987 struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z); 988 emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir); 989 } 990 991 emit_normalize_vec3(p, half, half); 992 993 release_temp(p, dist); 994 } 995 996 /* Calculate dot products: 997 */ 998 emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli); 999 emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half); 1000 1001 1002 /* Front face lighting: 1003 */ 1004 { 1005 struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT); 1006 struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE); 1007 struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR); 1008 struct ureg res0, res1; 1009 GLuint mask0, mask1; 1010 1011 emit_op1(p, OPCODE_LIT, lit, 0, dots); 1012 1013 if (!is_undef(att)) 1014 emit_op2(p, OPCODE_MUL, lit, 0, lit, att); 1015 1016 1017 if (count == nr_lights) { 1018 if (separate) { 1019 mask0 = WRITEMASK_XYZ; 1020 mask1 = WRITEMASK_XYZ; 1021 res0 = register_output( p, VERT_RESULT_COL0 ); 1022 res1 = register_output( p, VERT_RESULT_COL1 ); 1023 } 1024 else { 1025 mask0 = 0; 1026 mask1 = WRITEMASK_XYZ; 1027 res0 = _col0; 1028 res1 = register_output( p, VERT_RESULT_COL0 ); 1029 } 1030 } else { 1031 mask0 = 0; 1032 mask1 = 0; 1033 res0 = _col0; 1034 res1 = _col1; 1035 } 1036 1037 emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0); 1038 emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0); 1039 emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1); 1040 1041 release_temp(p, ambient); 1042 release_temp(p, diffuse); 1043 release_temp(p, specular); 1044 } 1045 1046 /* Back face lighting: 1047 */ 1048 if (twoside) { 1049 struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT); 1050 struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE); 1051 struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR); 1052 struct ureg res0, res1; 1053 GLuint mask0, mask1; 1054 1055 emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z))); 1056 1057 if (!is_undef(att)) 1058 emit_op2(p, OPCODE_MUL, lit, 0, lit, att); 1059 1060 if (count == nr_lights) { 1061 if (separate) { 1062 mask0 = WRITEMASK_XYZ; 1063 mask1 = WRITEMASK_XYZ; 1064 res0 = register_output( p, VERT_RESULT_BFC0 ); 1065 res1 = register_output( p, VERT_RESULT_BFC1 ); 1066 } 1067 else { 1068 mask0 = 0; 1069 mask1 = WRITEMASK_XYZ; 1070 res0 = _bfc0; 1071 res1 = register_output( p, VERT_RESULT_BFC0 ); 1072 } 1073 } else { 1074 res0 = _bfc0; 1075 res1 = _bfc1; 1076 mask0 = 0; 1077 mask1 = 0; 1078 } 1079 1080 emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0); 1081 emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0); 1082 emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1); 1083 1084 release_temp(p, ambient); 1085 release_temp(p, diffuse); 1086 release_temp(p, specular); 1087 } 1088 1089 release_temp(p, half); 1090 release_temp(p, VPpli); 1091 release_temp(p, att); 1092 } 1093 } 1094 1095 release_temps( p ); 1096} 1097 1098 1099static void build_fog( struct tnl_program *p ) 1100{ 1101 struct ureg fog = register_output(p, VERT_RESULT_FOGC); 1102 struct ureg input; 1103 1104 if (p->state->fog_source_is_depth) { 1105 input = swizzle1(get_eye_position(p), Z); 1106 } 1107 else { 1108 input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X); 1109 } 1110 1111 if (p->state->tnl_do_vertex_fog) { 1112 struct ureg params = register_param1(p, STATE_FOG_PARAMS); 1113 struct ureg tmp = get_temp(p); 1114 1115 switch (p->state->fog_mode) { 1116 case FOG_LINEAR: { 1117 struct ureg id = get_identity_param(p); 1118 emit_op2(p, OPCODE_SUB, tmp, 0, swizzle1(params,Z), input); 1119 emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,W)); 1120 emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */ 1121 emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W)); 1122 break; 1123 } 1124 case FOG_EXP: 1125 emit_op1(p, OPCODE_ABS, tmp, 0, input); 1126 emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,X)); 1127 emit_op2(p, OPCODE_POW, fog, WRITEMASK_X, 1128 register_const1f(p, M_E), negate(tmp)); 1129 break; 1130 case FOG_EXP2: 1131 emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,X)); 1132 emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp); 1133 emit_op2(p, OPCODE_POW, fog, WRITEMASK_X, 1134 register_const1f(p, M_E), negate(tmp)); 1135 break; 1136 } 1137 1138 release_temp(p, tmp); 1139 } 1140 else { 1141 /* results = incoming fog coords (compute fog per-fragment later) 1142 * 1143 * KW: Is it really necessary to do anything in this case? 1144 */ 1145 emit_op1(p, OPCODE_MOV, fog, WRITEMASK_X, input); 1146 } 1147} 1148 1149static void build_reflect_texgen( struct tnl_program *p, 1150 struct ureg dest, 1151 GLuint writemask ) 1152{ 1153 struct ureg normal = get_eye_normal(p); 1154 struct ureg eye_hat = get_eye_position_normalized(p); 1155 struct ureg tmp = get_temp(p); 1156 1157 /* n.u */ 1158 emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat); 1159 /* 2n.u */ 1160 emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp); 1161 /* (-2n.u)n + u */ 1162 emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat); 1163 1164 release_temp(p, tmp); 1165} 1166 1167static void build_sphere_texgen( struct tnl_program *p, 1168 struct ureg dest, 1169 GLuint writemask ) 1170{ 1171 struct ureg normal = get_eye_normal(p); 1172 struct ureg eye_hat = get_eye_position_normalized(p); 1173 struct ureg tmp = get_temp(p); 1174 struct ureg half = register_scalar_const(p, .5); 1175 struct ureg r = get_temp(p); 1176 struct ureg inv_m = get_temp(p); 1177 struct ureg id = get_identity_param(p); 1178 1179 /* Could share the above calculations, but it would be 1180 * a fairly odd state for someone to set (both sphere and 1181 * reflection active for different texture coordinate 1182 * components. Of course - if two texture units enable 1183 * reflect and/or sphere, things start to tilt in favour 1184 * of seperating this out: 1185 */ 1186 1187 /* n.u */ 1188 emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat); 1189 /* 2n.u */ 1190 emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp); 1191 /* (-2n.u)n + u */ 1192 emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat); 1193 /* r + 0,0,1 */ 1194 emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z)); 1195 /* rx^2 + ry^2 + (rz+1)^2 */ 1196 emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp); 1197 /* 2/m */ 1198 emit_op1(p, OPCODE_RSQ, tmp, 0, tmp); 1199 /* 1/m */ 1200 emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half); 1201 /* r/m + 1/2 */ 1202 emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half); 1203 1204 release_temp(p, tmp); 1205 release_temp(p, r); 1206 release_temp(p, inv_m); 1207} 1208 1209 1210static void build_texture_transform( struct tnl_program *p ) 1211{ 1212 GLuint i, j; 1213 1214 for (i = 0; i < MAX_TEXTURE_UNITS; i++) { 1215 1216 if (!(p->state->fragprog_inputs_read & FRAG_BIT_TEX(i))) 1217 continue; 1218 1219 if (p->state->unit[i].texgen_enabled || 1220 p->state->unit[i].texmat_enabled) { 1221 1222 GLuint texmat_enabled = p->state->unit[i].texmat_enabled; 1223 struct ureg out = register_output(p, VERT_RESULT_TEX0 + i); 1224 struct ureg out_texgen = undef; 1225 1226 if (p->state->unit[i].texgen_enabled) { 1227 GLuint copy_mask = 0; 1228 GLuint sphere_mask = 0; 1229 GLuint reflect_mask = 0; 1230 GLuint normal_mask = 0; 1231 GLuint modes[4]; 1232 1233 if (texmat_enabled) 1234 out_texgen = get_temp(p); 1235 else 1236 out_texgen = out; 1237 1238 modes[0] = p->state->unit[i].texgen_mode0; 1239 modes[1] = p->state->unit[i].texgen_mode1; 1240 modes[2] = p->state->unit[i].texgen_mode2; 1241 modes[3] = p->state->unit[i].texgen_mode3; 1242 1243 for (j = 0; j < 4; j++) { 1244 switch (modes[j]) { 1245 case TXG_OBJ_LINEAR: { 1246 struct ureg obj = register_input(p, VERT_ATTRIB_POS); 1247 struct ureg plane = 1248 register_param3(p, STATE_TEXGEN, i, 1249 STATE_TEXGEN_OBJECT_S + j); 1250 1251 emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j, 1252 obj, plane ); 1253 break; 1254 } 1255 case TXG_EYE_LINEAR: { 1256 struct ureg eye = get_eye_position(p); 1257 struct ureg plane = 1258 register_param3(p, STATE_TEXGEN, i, 1259 STATE_TEXGEN_EYE_S + j); 1260 1261 emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j, 1262 eye, plane ); 1263 break; 1264 } 1265 case TXG_SPHERE_MAP: 1266 sphere_mask |= WRITEMASK_X << j; 1267 break; 1268 case TXG_REFLECTION_MAP: 1269 reflect_mask |= WRITEMASK_X << j; 1270 break; 1271 case TXG_NORMAL_MAP: 1272 normal_mask |= WRITEMASK_X << j; 1273 break; 1274 case TXG_NONE: 1275 copy_mask |= WRITEMASK_X << j; 1276 } 1277 1278 } 1279 1280 1281 if (sphere_mask) { 1282 build_sphere_texgen(p, out_texgen, sphere_mask); 1283 } 1284 1285 if (reflect_mask) { 1286 build_reflect_texgen(p, out_texgen, reflect_mask); 1287 } 1288 1289 if (normal_mask) { 1290 struct ureg normal = get_eye_normal(p); 1291 emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal ); 1292 } 1293 1294 if (copy_mask) { 1295 struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i); 1296 emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in ); 1297 } 1298 } 1299 1300 if (texmat_enabled) { 1301 struct ureg texmat[4]; 1302 struct ureg in = (!is_undef(out_texgen) ? 1303 out_texgen : 1304 register_input(p, VERT_ATTRIB_TEX0+i)); 1305 if (PREFER_DP4) { 1306 register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i, 1307 0, 3, STATE_MATRIX, texmat ); 1308 emit_matrix_transform_vec4( p, out, texmat, in ); 1309 } 1310 else { 1311 register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i, 1312 0, 3, STATE_MATRIX_TRANSPOSE, texmat ); 1313 emit_transpose_matrix_transform_vec4( p, out, texmat, in ); 1314 } 1315 } 1316 1317 release_temps(p); 1318 } 1319 else { 1320 emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i); 1321 } 1322 } 1323} 1324 1325 1326/* Seems like it could be tighter: 1327 */ 1328static void build_pointsize( struct tnl_program *p ) 1329{ 1330 struct ureg eye = get_eye_position(p); 1331 struct ureg state_size = register_param1(p, STATE_POINT_SIZE); 1332 struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION); 1333 struct ureg out = register_output(p, VERT_RESULT_PSIZ); 1334 struct ureg ut = get_temp(p); 1335 1336 /* 1, -Z, Z * Z, 1 */ 1337 emit_op1(p, OPCODE_MOV, ut, 0, swizzle1(get_identity_param(p), W)); 1338 emit_op2(p, OPCODE_MUL, ut, WRITEMASK_YZ, ut, negate(swizzle1(eye, Z))); 1339 emit_op2(p, OPCODE_MUL, ut, WRITEMASK_Z, ut, negate(swizzle1(eye, Z))); 1340 1341 1342 /* p1 + p2 * dist + p3 * dist * dist, 0 */ 1343 emit_op2(p, OPCODE_DP3, ut, 0, ut, state_attenuation); 1344 1345 /* 1 / factor */ 1346 emit_op1(p, OPCODE_RCP, ut, 0, ut ); 1347 1348 /* out = pointSize / factor */ 1349 emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size); 1350 1351 release_temp(p, ut); 1352} 1353 1354static void build_tnl_program( struct tnl_program *p ) 1355{ /* Emit the program, starting with modelviewproject: 1356 */ 1357 build_hpos(p); 1358 1359 /* Lighting calculations: 1360 */ 1361 if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) { 1362 if (p->state->light_global_enabled) 1363 build_lighting(p); 1364 else { 1365 if (p->state->fragprog_inputs_read & FRAG_BIT_COL0) 1366 emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0); 1367 1368 if (p->state->fragprog_inputs_read & FRAG_BIT_COL1) 1369 emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1); 1370 } 1371 } 1372 1373 if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) || 1374 p->state->fog_mode != FOG_NONE) 1375 build_fog(p); 1376 1377 if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY) 1378 build_texture_transform(p); 1379 1380 if (p->state->point_attenuated) 1381 build_pointsize(p); 1382 1383 /* Finish up: 1384 */ 1385 emit_op1(p, OPCODE_END, undef, 0, undef); 1386 1387 /* Disassemble: 1388 */ 1389 if (DISASSEM) { 1390 _mesa_printf ("\n"); 1391 } 1392} 1393 1394 1395static void 1396create_new_program( const struct state_key *key, 1397 struct gl_vertex_program *program, 1398 GLuint max_temps) 1399{ 1400 struct tnl_program p; 1401 1402 _mesa_memset(&p, 0, sizeof(p)); 1403 p.state = key; 1404 p.program = program; 1405 p.eye_position = undef; 1406 p.eye_position_normalized = undef; 1407 p.eye_normal = undef; 1408 p.identity = undef; 1409 p.temp_in_use = 0; 1410 1411 if (max_temps >= sizeof(int) * 8) 1412 p.temp_reserved = 0; 1413 else 1414 p.temp_reserved = ~((1<<max_temps)-1); 1415 1416 p.program->Base.Instructions 1417 = (struct prog_instruction*) MALLOC(sizeof(struct prog_instruction) * MAX_INSN); 1418 p.program->Base.String = 0; 1419 p.program->Base.NumInstructions = 1420 p.program->Base.NumTemporaries = 1421 p.program->Base.NumParameters = 1422 p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0; 1423 p.program->Base.Parameters = _mesa_new_parameter_list(); 1424 p.program->Base.InputsRead = 0; 1425 p.program->Base.OutputsWritten = 0; 1426 1427 build_tnl_program( &p ); 1428} 1429 1430static void *search_cache( struct tnl_cache *cache, 1431 GLuint hash, 1432 const void *key, 1433 GLuint keysize) 1434{ 1435 struct tnl_cache_item *c; 1436 1437 for (c = cache->items[hash % cache->size]; c; c = c->next) { 1438 if (c->hash == hash && _mesa_memcmp(c->key, key, keysize) == 0) 1439 return c->data; 1440 } 1441 1442 return NULL; 1443} 1444 1445static void rehash( struct tnl_cache *cache ) 1446{ 1447 struct tnl_cache_item **items; 1448 struct tnl_cache_item *c, *next; 1449 GLuint size, i; 1450 1451 size = cache->size * 3; 1452 items = (struct tnl_cache_item**) _mesa_malloc(size * sizeof(*items)); 1453 _mesa_memset(items, 0, size * sizeof(*items)); 1454 1455 for (i = 0; i < cache->size; i++) 1456 for (c = cache->items[i]; c; c = next) { 1457 next = c->next; 1458 c->next = items[c->hash % size]; 1459 items[c->hash % size] = c; 1460 } 1461 1462 FREE(cache->items); 1463 cache->items = items; 1464 cache->size = size; 1465} 1466 1467static void cache_item( struct tnl_cache *cache, 1468 GLuint hash, 1469 void *key, 1470 void *data ) 1471{ 1472 struct tnl_cache_item *c = (struct tnl_cache_item*) _mesa_malloc(sizeof(*c)); 1473 c->hash = hash; 1474 c->key = key; 1475 c->data = data; 1476 1477 if (++cache->n_items > cache->size * 1.5) 1478 rehash(cache); 1479 1480 c->next = cache->items[hash % cache->size]; 1481 cache->items[hash % cache->size] = c; 1482} 1483 1484static GLuint hash_key( struct state_key *key ) 1485{ 1486 GLuint *ikey = (GLuint *)key; 1487 GLuint hash = 0, i; 1488 1489 /* I'm sure this can be improved on, but speed is important: 1490 */ 1491 for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++) 1492 hash ^= ikey[i]; 1493 1494 return hash; 1495} 1496 1497void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx ) 1498{ 1499 TNLcontext *tnl = TNL_CONTEXT(ctx); 1500 struct state_key *key; 1501 GLuint hash; 1502 const struct gl_vertex_program *prev = ctx->VertexProgram._Current; 1503 1504 if (!ctx->VertexProgram._Current) { 1505 /* Grab all the relevent state and put it in a single structure: 1506 */ 1507 key = make_state_key(ctx); 1508 hash = hash_key(key); 1509 1510 /* Look for an already-prepared program for this state: 1511 */ 1512 ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *) 1513 search_cache( tnl->vp_cache, hash, key, sizeof(*key) ); 1514 1515 /* OK, we'll have to build a new one: 1516 */ 1517 if (!ctx->VertexProgram._TnlProgram) { 1518 if (0) 1519 _mesa_printf("Build new TNL program\n"); 1520 1521 ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *) 1522 ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0); 1523 1524 create_new_program( key, ctx->VertexProgram._TnlProgram, 1525 ctx->Const.VertexProgram.MaxTemps ); 1526 1527 if (ctx->Driver.ProgramStringNotify) 1528 ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB, 1529 &ctx->VertexProgram._TnlProgram->Base ); 1530 1531 cache_item(tnl->vp_cache, hash, key, ctx->VertexProgram._TnlProgram ); 1532 } 1533 else { 1534 FREE(key); 1535 if (0) 1536 _mesa_printf("Found existing TNL program for key %x\n", hash); 1537 } 1538 ctx->VertexProgram._Current = ctx->VertexProgram._TnlProgram; 1539 } 1540 1541 /* Tell the driver about the change. Could define a new target for 1542 * this? 1543 */ 1544 if (ctx->VertexProgram._Current != prev && ctx->Driver.BindProgram) { 1545 ctx->Driver.BindProgram(ctx, GL_VERTEX_PROGRAM_ARB, 1546 (struct gl_program *) ctx->VertexProgram._Current); 1547 } 1548} 1549 1550void _tnl_ProgramCacheInit( GLcontext *ctx ) 1551{ 1552 TNLcontext *tnl = TNL_CONTEXT(ctx); 1553 1554 tnl->vp_cache = (struct tnl_cache *) MALLOC(sizeof(*tnl->vp_cache)); 1555 tnl->vp_cache->size = 17; 1556 tnl->vp_cache->n_items = 0; 1557 tnl->vp_cache->items = (struct tnl_cache_item**) 1558 _mesa_calloc(tnl->vp_cache->size * sizeof(*tnl->vp_cache->items)); 1559} 1560 1561void _tnl_ProgramCacheDestroy( GLcontext *ctx ) 1562{ 1563 TNLcontext *tnl = TNL_CONTEXT(ctx); 1564 struct tnl_cache_item *c, *next; 1565 GLuint i; 1566 1567 for (i = 0; i < tnl->vp_cache->size; i++) 1568 for (c = tnl->vp_cache->items[i]; c; c = next) { 1569 next = c->next; 1570 FREE(c->key); 1571 FREE(c->data); 1572 FREE(c); 1573 } 1574 1575 FREE(tnl->vp_cache->items); 1576 FREE(tnl->vp_cache); 1577} 1578