t_vp_build.c revision 5f430c9976c71d7167f9327623e1df5fce53a970
1/* 2 * Mesa 3-D graphics library 3 * Version: 6.3 4 * 5 * Copyright (C) 2005 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 <strings.h> 34 35#include "glheader.h" 36#include "macros.h" 37#include "enums.h" 38#include "t_context.h" 39#include "t_vp_build.h" 40 41#include "shader/program.h" 42#include "shader/nvvertprog.h" 43#include "shader/arbvertparse.h" 44 45 46/* Very useful debugging tool - produces annotated listing of 47 * generated program with line/function references for each 48 * instruction back into this file: 49 */ 50#define DISASSEM 0 51 52/* Should be tunable by the driver - do we want to do matrix 53 * multiplications with DP4's or with MUL/MAD's? SSE works better 54 * with the latter, drivers may differ. 55 */ 56#define PREFER_DP4 1 57 58#define MAX_INSN 200 59 60/* Use uregs to represent registers internally, translate to Mesa's 61 * expected formats on emit. 62 * 63 * NOTE: These are passed by value extensively in this file rather 64 * than as usual by pointer reference. If this disturbs you, try 65 * remembering they are just 32bits in size. 66 * 67 * GCC is smart enough to deal with these dword-sized structures in 68 * much the same way as if I had defined them as dwords and was using 69 * macros to access and set the fields. This is much nicer and easier 70 * to evolve. 71 */ 72struct ureg { 73 GLuint file:4; 74 GLuint idx:8; 75 GLuint negate:1; 76 GLuint swz:12; 77 GLuint pad:7; 78}; 79 80 81struct tnl_program { 82 GLcontext *ctx; 83 struct vertex_program *program; 84 85 GLuint temp_in_use; 86 GLuint temp_reserved; 87 88 struct ureg eye_position; 89 struct ureg eye_position_normalized; 90 struct ureg eye_normal; 91 struct ureg identity; 92 93 GLuint materials; 94 GLuint color_materials; 95}; 96 97 98const static struct ureg undef = { 99 ~0, 100 ~0, 101 0, 102 0, 103 0 104}; 105 106/* Local shorthand: 107 */ 108#define X SWIZZLE_X 109#define Y SWIZZLE_Y 110#define Z SWIZZLE_Z 111#define W SWIZZLE_W 112 113 114/* Construct a ureg: 115 */ 116static struct ureg make_ureg(GLuint file, GLuint idx) 117{ 118 struct ureg reg; 119 reg.file = file; 120 reg.idx = idx; 121 reg.negate = 0; 122 reg.swz = SWIZZLE_NOOP; 123 reg.pad = 0; 124 return reg; 125} 126 127 128 129static struct ureg negate( struct ureg reg ) 130{ 131 reg.negate ^= 1; 132 return reg; 133} 134 135 136static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w ) 137{ 138 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x), 139 GET_SWZ(reg.swz, y), 140 GET_SWZ(reg.swz, z), 141 GET_SWZ(reg.swz, w)); 142 143 return reg; 144} 145 146static struct ureg swizzle1( struct ureg reg, int x ) 147{ 148 return swizzle(reg, x, x, x, x); 149} 150 151static struct ureg get_temp( struct tnl_program *p ) 152{ 153 int bit = ffs( ~p->temp_in_use ); 154 if (!bit) { 155 fprintf(stderr, "%s: out of temporaries\n", __FILE__); 156 exit(1); 157 } 158 159 p->temp_in_use |= 1<<(bit-1); 160 return make_ureg(PROGRAM_TEMPORARY, bit-1); 161} 162 163static struct ureg reserve_temp( struct tnl_program *p ) 164{ 165 struct ureg temp = get_temp( p ); 166 p->temp_reserved |= 1<<temp.idx; 167 return temp; 168} 169 170static void release_temp( struct tnl_program *p, struct ureg reg ) 171{ 172 if (reg.file == PROGRAM_TEMPORARY) { 173 p->temp_in_use &= ~(1<<reg.idx); 174 p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */ 175 } 176} 177 178static void release_temps( struct tnl_program *p ) 179{ 180 p->temp_in_use = p->temp_reserved; 181} 182 183 184 185static struct ureg register_input( struct tnl_program *p, GLuint input ) 186{ 187 p->program->InputsRead |= (1<<input); 188 return make_ureg(PROGRAM_INPUT, input); 189} 190 191static struct ureg register_output( struct tnl_program *p, GLuint output ) 192{ 193 p->program->OutputsWritten |= (1<<output); 194 return make_ureg(PROGRAM_OUTPUT, output); 195} 196 197static struct ureg register_const4f( struct tnl_program *p, 198 GLfloat s0, 199 GLfloat s1, 200 GLfloat s2, 201 GLfloat s3) 202{ 203 GLfloat values[4]; 204 GLuint idx; 205 values[0] = s0; 206 values[1] = s1; 207 values[2] = s2; 208 values[3] = s3; 209 idx = _mesa_add_unnamed_constant( p->program->Parameters, values ); 210 return make_ureg(PROGRAM_STATE_VAR, idx); 211} 212 213#define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1) 214#define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1) 215#define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1) 216 217static GLboolean is_undef( struct ureg reg ) 218{ 219 return reg.file == 0xf; 220} 221 222static struct ureg get_identity_param( struct tnl_program *p ) 223{ 224 if (is_undef(p->identity)) 225 p->identity = register_const4f(p, 0,0,0,1); 226 227 return p->identity; 228} 229 230static struct ureg register_param6( struct tnl_program *p, 231 GLint s0, 232 GLint s1, 233 GLint s2, 234 GLint s3, 235 GLint s4, 236 GLint s5) 237{ 238 GLint tokens[6]; 239 GLuint idx; 240 tokens[0] = s0; 241 tokens[1] = s1; 242 tokens[2] = s2; 243 tokens[3] = s3; 244 tokens[4] = s4; 245 tokens[5] = s5; 246 idx = _mesa_add_state_reference( p->program->Parameters, tokens ); 247 return make_ureg(PROGRAM_STATE_VAR, idx); 248} 249 250 251#define register_param1(p,s0) register_param6(p,s0,0,0,0,0,0) 252#define register_param2(p,s0,s1) register_param6(p,s0,s1,0,0,0,0) 253#define register_param3(p,s0,s1,s2) register_param6(p,s0,s1,s2,0,0,0) 254#define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0) 255 256 257static void register_matrix_param6( struct tnl_program *p, 258 GLint s0, 259 GLint s1, 260 GLint s2, 261 GLint s3, 262 GLint s4, 263 GLint s5, 264 struct ureg *matrix ) 265{ 266 GLuint i; 267 268 /* This is a bit sad as the support is there to pull the whole 269 * matrix out in one go: 270 */ 271 for (i = 0; i <= s4 - s3; i++) 272 matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 ); 273} 274 275 276static void emit_arg( struct vp_src_register *src, 277 struct ureg reg ) 278{ 279 src->File = reg.file; 280 src->Index = reg.idx; 281 src->Swizzle = reg.swz; 282 src->Negate = reg.negate; 283 src->RelAddr = 0; 284 src->pad = 0; 285} 286 287static void emit_dst( struct vp_dst_register *dst, 288 struct ureg reg, GLuint mask ) 289{ 290 dst->File = reg.file; 291 dst->Index = reg.idx; 292 /* allow zero as a shorthand for xyzw */ 293 dst->WriteMask = mask ? mask : WRITEMASK_XYZW; 294 dst->pad = 0; 295} 296 297static void debug_insn( struct vp_instruction *inst, const char *fn, 298 GLuint line ) 299{ 300#if DISASSEM 301 static const char *last_fn; 302 303 if (fn != last_fn) { 304 last_fn = fn; 305 _mesa_printf("%s:\n", fn); 306 } 307 308 _mesa_printf("%d:\t", line); 309 _mesa_debug_vp_inst(1, inst); 310#endif 311} 312 313 314static void emit_op3fn(struct tnl_program *p, 315 GLuint op, 316 struct ureg dest, 317 GLuint mask, 318 struct ureg src0, 319 struct ureg src1, 320 struct ureg src2, 321 const char *fn, 322 GLuint line) 323{ 324 GLuint nr = p->program->Base.NumInstructions++; 325 struct vp_instruction *inst = &p->program->Instructions[nr]; 326 327 if (p->program->Base.NumInstructions > MAX_INSN) { 328 _mesa_problem(p->ctx, "Out of instructions in emit_op3fn\n"); 329 return; 330 } 331 332 inst->Opcode = op; 333 inst->StringPos = 0; 334 inst->Data = 0; 335 336 emit_arg( &inst->SrcReg[0], src0 ); 337 emit_arg( &inst->SrcReg[1], src1 ); 338 emit_arg( &inst->SrcReg[2], src2 ); 339 340 emit_dst( &inst->DstReg, dest, mask ); 341 342 debug_insn(inst, fn, line); 343} 344 345 346 347#define emit_op3(p, op, dst, mask, src0, src1, src2) \ 348 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__) 349 350#define emit_op2(p, op, dst, mask, src0, src1) \ 351 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__) 352 353#define emit_op1(p, op, dst, mask, src0) \ 354 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__) 355 356 357static struct ureg make_temp( struct tnl_program *p, struct ureg reg ) 358{ 359 if (reg.file == PROGRAM_TEMPORARY && 360 !(p->temp_reserved & (1<<reg.idx))) 361 return reg; 362 else { 363 struct ureg temp = get_temp(p); 364 emit_op1(p, VP_OPCODE_MOV, temp, 0, reg); 365 return temp; 366 } 367} 368 369 370/* Currently no tracking performed of input/output/register size or 371 * active elements. Could be used to reduce these operations, as 372 * could the matrix type. 373 */ 374static void emit_matrix_transform_vec4( struct tnl_program *p, 375 struct ureg dest, 376 const struct ureg *mat, 377 struct ureg src) 378{ 379 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]); 380 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]); 381 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]); 382 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]); 383} 384 385/* This version is much easier to implement if writemasks are not 386 * supported natively on the target or (like SSE), the target doesn't 387 * have a clean/obvious dotproduct implementation. 388 */ 389static void emit_transpose_matrix_transform_vec4( struct tnl_program *p, 390 struct ureg dest, 391 const struct ureg *mat, 392 struct ureg src) 393{ 394 struct ureg tmp; 395 396 if (dest.file != PROGRAM_TEMPORARY) 397 tmp = get_temp(p); 398 else 399 tmp = dest; 400 401 emit_op2(p, VP_OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]); 402 emit_op3(p, VP_OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp); 403 emit_op3(p, VP_OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp); 404 emit_op3(p, VP_OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp); 405 406 if (dest.file != PROGRAM_TEMPORARY) 407 release_temp(p, tmp); 408} 409 410static void emit_matrix_transform_vec3( struct tnl_program *p, 411 struct ureg dest, 412 const struct ureg *mat, 413 struct ureg src) 414{ 415 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]); 416 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]); 417 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]); 418} 419 420 421static void emit_normalize_vec3( struct tnl_program *p, 422 struct ureg dest, 423 struct ureg src ) 424{ 425 struct ureg tmp = get_temp(p); 426 emit_op2(p, VP_OPCODE_DP3, tmp, 0, src, src); 427 emit_op1(p, VP_OPCODE_RSQ, tmp, 0, tmp); 428 emit_op2(p, VP_OPCODE_MUL, dest, 0, src, tmp); 429 release_temp(p, tmp); 430} 431 432static void emit_passthrough( struct tnl_program *p, 433 GLuint input, 434 GLuint output ) 435{ 436 struct ureg out = register_output(p, output); 437 emit_op1(p, VP_OPCODE_MOV, out, 0, register_input(p, input)); 438} 439 440static struct ureg get_eye_position( struct tnl_program *p ) 441{ 442 if (is_undef(p->eye_position)) { 443 struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 444 struct ureg modelview[4]; 445 446 p->eye_position = reserve_temp(p); 447 448 if (PREFER_DP4) { 449 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3, 450 STATE_MATRIX, modelview ); 451 452 emit_matrix_transform_vec4(p, p->eye_position, modelview, pos); 453 } 454 else { 455 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3, 456 STATE_MATRIX_TRANSPOSE, modelview ); 457 458 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos); 459 } 460 } 461 462 return p->eye_position; 463} 464 465 466static struct ureg get_eye_position_normalized( struct tnl_program *p ) 467{ 468 if (is_undef(p->eye_position_normalized)) { 469 struct ureg eye = get_eye_position(p); 470 p->eye_position_normalized = reserve_temp(p); 471 emit_normalize_vec3(p, p->eye_position_normalized, eye); 472 } 473 474 return p->eye_position_normalized; 475} 476 477 478static struct ureg get_eye_normal( struct tnl_program *p ) 479{ 480 if (is_undef(p->eye_normal)) { 481 struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL ); 482 struct ureg mvinv[3]; 483 484 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 2, 485 STATE_MATRIX_INVTRANS, mvinv ); 486 487 p->eye_normal = reserve_temp(p); 488 489 /* Transform to eye space: 490 */ 491 emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal ); 492 493 /* Normalize/Rescale: 494 */ 495 if (p->ctx->Transform.Normalize) { 496 emit_normalize_vec3( p, p->eye_normal, p->eye_normal ); 497 } 498 else if (p->ctx->Transform.RescaleNormals) { 499 struct ureg rescale = register_param2(p, STATE_INTERNAL, 500 STATE_NORMAL_SCALE); 501 502 emit_op2( p, VP_OPCODE_MUL, p->eye_normal, 0, normal, 503 swizzle1(rescale, X)); 504 } 505 } 506 507 return p->eye_normal; 508} 509 510 511 512static void build_hpos( struct tnl_program *p ) 513{ 514 struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 515 struct ureg hpos = register_output( p, VERT_RESULT_HPOS ); 516 struct ureg mvp[4]; 517 518 if (PREFER_DP4) { 519 register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3, 520 STATE_MATRIX, mvp ); 521 emit_matrix_transform_vec4( p, hpos, mvp, pos ); 522 } 523 else { 524 register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3, 525 STATE_MATRIX_TRANSPOSE, mvp ); 526 emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos ); 527 } 528} 529 530 531static GLuint material_attrib( GLuint side, GLuint property ) 532{ 533 return (_TNL_ATTRIB_MAT_FRONT_AMBIENT + 534 (property - STATE_AMBIENT) * 2 + 535 side); 536} 537 538static void set_material_flags( struct tnl_program *p ) 539{ 540 GLcontext *ctx = p->ctx; 541 TNLcontext *tnl = TNL_CONTEXT(ctx); 542 GLuint i; 543 544 p->color_materials = 0; 545 p->materials = 0; 546 547 if (ctx->Light.ColorMaterialEnabled) { 548 p->materials = 549 p->color_materials = 550 ctx->Light.ColorMaterialBitmask << _TNL_ATTRIB_MAT_FRONT_AMBIENT; 551 } 552 553 for (i = _TNL_ATTRIB_MAT_FRONT_AMBIENT ; i < _TNL_ATTRIB_INDEX ; i++) 554 if (tnl->vb.AttribPtr[i]->stride) 555 p->materials |= 1<<i; 556} 557 558 559static struct ureg get_material( struct tnl_program *p, GLuint side, 560 GLuint property ) 561{ 562 GLuint attrib = material_attrib(side, property); 563 564 if (p->color_materials & (1<<attrib)) 565 return register_input(p, VERT_ATTRIB_COLOR0); 566 else if (p->materials & (1<<attrib)) 567 return register_input( p, attrib ); 568 else 569 return register_param3( p, STATE_MATERIAL, side, property ); 570} 571 572#define SCENE_COLOR_BITS(side) (( _TNL_BIT_MAT_FRONT_EMISSION | \ 573 _TNL_BIT_MAT_FRONT_AMBIENT | \ 574 _TNL_BIT_MAT_FRONT_DIFFUSE) << (side)) 575 576/* Either return a precalculated constant value or emit code to 577 * calculate these values dynamically in the case where material calls 578 * are present between begin/end pairs. 579 * 580 * Probably want to shift this to the program compilation phase - if 581 * we always emitted the calculation here, a smart compiler could 582 * detect that it was constant (given a certain set of inputs), and 583 * lift it out of the main loop. That way the programs created here 584 * would be independent of the vertex_buffer details. 585 */ 586static struct ureg get_scenecolor( struct tnl_program *p, GLuint side ) 587{ 588 if (p->materials & SCENE_COLOR_BITS(side)) { 589 struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT); 590 struct ureg material_emission = get_material(p, side, STATE_EMISSION); 591 struct ureg material_ambient = get_material(p, side, STATE_AMBIENT); 592 struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE); 593 struct ureg tmp = make_temp(p, material_diffuse); 594 emit_op3(p, VP_OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient, 595 material_ambient, material_emission); 596 return tmp; 597 } 598 else 599 return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side ); 600} 601 602 603static struct ureg get_lightprod( struct tnl_program *p, GLuint light, 604 GLuint side, GLuint property ) 605{ 606 GLuint attrib = material_attrib(side, property); 607 if (p->materials & (1<<attrib)) { 608 struct ureg light_value = 609 register_param3(p, STATE_LIGHT, light, property); 610 struct ureg material_value = get_material(p, side, property); 611 struct ureg tmp = get_temp(p); 612 emit_op2(p, VP_OPCODE_MUL, tmp, 0, light_value, material_value); 613 return tmp; 614 } 615 else 616 return register_param4(p, STATE_LIGHTPROD, light, side, property); 617} 618 619static struct ureg calculate_light_attenuation( struct tnl_program *p, 620 GLuint i, 621 struct gl_light *light, 622 struct ureg VPpli, 623 struct ureg dist ) 624{ 625 struct ureg attenuation = register_param3(p, STATE_LIGHT, i, 626 STATE_ATTENUATION); 627 struct ureg att = get_temp(p); 628 629 /* Calculate spot attenuation: 630 */ 631 if (light->SpotCutoff != 180.0F) { 632 struct ureg spot_dir = register_param3(p, STATE_LIGHT, i, 633 STATE_SPOT_DIRECTION); 634 struct ureg spot = get_temp(p); 635 struct ureg slt = get_temp(p); 636 637 emit_normalize_vec3( p, spot, spot_dir ); /* XXX: precompute! */ 638 emit_op2(p, VP_OPCODE_DP3, spot, 0, negate(VPpli), spot_dir); 639 emit_op2(p, VP_OPCODE_SLT, slt, 0, swizzle1(spot_dir,W), spot); 640 emit_op2(p, VP_OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W)); 641 emit_op2(p, VP_OPCODE_MUL, att, 0, slt, spot); 642 643 release_temp(p, spot); 644 release_temp(p, slt); 645 } 646 647 /* Calculate distance attenuation: 648 */ 649 if (light->ConstantAttenuation != 1.0 || 650 light->LinearAttenuation != 1.0 || 651 light->QuadraticAttenuation != 1.0) { 652 653 /* 1/d,d,d,1/d */ 654 emit_op1(p, VP_OPCODE_RCP, dist, WRITEMASK_YZ, dist); 655 /* 1,d,d*d,1/d */ 656 emit_op2(p, VP_OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y)); 657 /* 1/dist-atten */ 658 emit_op2(p, VP_OPCODE_DP3, dist, 0, attenuation, dist); 659 660 if (light->SpotCutoff != 180.0F) { 661 /* dist-atten */ 662 emit_op1(p, VP_OPCODE_RCP, dist, 0, dist); 663 /* spot-atten * dist-atten */ 664 emit_op2(p, VP_OPCODE_MUL, att, 0, dist, att); 665 } else { 666 /* dist-atten */ 667 emit_op1(p, VP_OPCODE_RCP, att, 0, dist); 668 } 669 } 670 671 return att; 672} 673 674 675 676 677 678/* Need to add some addtional parameters to allow lighting in object 679 * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye 680 * space lighting. 681 */ 682static void build_lighting( struct tnl_program *p ) 683{ 684 GLcontext *ctx = p->ctx; 685 const GLboolean twoside = ctx->Light.Model.TwoSide; 686 const GLboolean separate = (ctx->Light.Model.ColorControl == 687 GL_SEPARATE_SPECULAR_COLOR); 688 GLuint nr_lights = 0, count = 0; 689 struct ureg normal = get_eye_normal(p); 690 struct ureg lit = get_temp(p); 691 struct ureg dots = get_temp(p); 692 struct ureg _col0 = undef, _col1 = undef; 693 struct ureg _bfc0 = undef, _bfc1 = undef; 694 GLuint i; 695 696 for (i = 0; i < MAX_LIGHTS; i++) 697 if (ctx->Light.Light[i].Enabled) 698 nr_lights++; 699 700 set_material_flags(p); 701 702 { 703 struct ureg shininess = get_material(p, 0, STATE_SHININESS); 704 emit_op1(p, VP_OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X)); 705 release_temp(p, shininess); 706 707 _col0 = make_temp(p, get_scenecolor(p, 0)); 708 if (separate) 709 _col1 = make_temp(p, get_identity_param(p)); 710 else 711 _col1 = _col0; 712 713 } 714 715 if (twoside) { 716 struct ureg shininess = get_material(p, 1, STATE_SHININESS); 717 emit_op1(p, VP_OPCODE_MOV, dots, WRITEMASK_Z, 718 negate(swizzle1(shininess,X))); 719 release_temp(p, shininess); 720 721 _bfc0 = make_temp(p, get_scenecolor(p, 1)); 722 if (separate) 723 _bfc1 = make_temp(p, get_identity_param(p)); 724 else 725 _bfc1 = _bfc0; 726 } 727 728 729 /* If no lights, still need to emit the scenecolor. 730 */ 731 if (nr_lights == 0) { 732 { 733 struct ureg res0 = register_output( p, VERT_RESULT_COL0 ); 734 emit_op1(p, VP_OPCODE_MOV, res0, 0, _col0); 735 } 736 737 if (separate) { 738 struct ureg res1 = register_output( p, VERT_RESULT_COL1 ); 739 emit_op1(p, VP_OPCODE_MOV, res1, 0, _col1); 740 } 741 742 if (twoside) { 743 struct ureg res0 = register_output( p, VERT_RESULT_BFC0 ); 744 emit_op1(p, VP_OPCODE_MOV, res0, 0, _bfc0); 745 } 746 747 if (twoside && separate) { 748 struct ureg res1 = register_output( p, VERT_RESULT_BFC1 ); 749 emit_op1(p, VP_OPCODE_MOV, res1, 0, _bfc1); 750 } 751 752 release_temps(p); 753 return; 754 } 755 756 757 for (i = 0; i < MAX_LIGHTS; i++) { 758 struct gl_light *light = &ctx->Light.Light[i]; 759 760 if (light->Enabled) { 761 struct ureg half = undef; 762 struct ureg att = undef, VPpli = undef; 763 764 count++; 765 766 if (light->EyePosition[3] == 0) { 767 /* Can used precomputed constants in this case. 768 * Attenuation never applies to infinite lights. 769 */ 770 VPpli = register_param3(p, STATE_LIGHT, i, 771 STATE_POSITION_NORMALIZED); 772 half = register_param3(p, STATE_LIGHT, i, STATE_HALF); 773 } 774 else { 775 struct ureg Ppli = register_param3(p, STATE_LIGHT, i, 776 STATE_POSITION); 777 struct ureg V = get_eye_position(p); 778 struct ureg dist = get_temp(p); 779 780 VPpli = get_temp(p); 781 half = get_temp(p); 782 783 /* Calulate VPpli vector 784 */ 785 emit_op2(p, VP_OPCODE_SUB, VPpli, 0, Ppli, V); 786 787 /* Normalize VPpli. The dist value also used in 788 * attenuation below. 789 */ 790 emit_op2(p, VP_OPCODE_DP3, dist, 0, VPpli, VPpli); 791 emit_op1(p, VP_OPCODE_RSQ, dist, 0, dist); 792 emit_op2(p, VP_OPCODE_MUL, VPpli, 0, VPpli, dist); 793 794 795 /* Calculate attenuation: 796 */ 797 if (light->SpotCutoff != 180.0 || 798 light->ConstantAttenuation != 1.0 || 799 light->LinearAttenuation != 1.0 || 800 light->QuadraticAttenuation != 1.0) { 801 att = calculate_light_attenuation(p, i, light, VPpli, dist); 802 } 803 804 805 /* Calculate viewer direction, or use infinite viewer: 806 */ 807 if (ctx->Light.Model.LocalViewer) { 808 struct ureg eye_hat = get_eye_position_normalized(p); 809 emit_op2(p, VP_OPCODE_SUB, half, 0, VPpli, eye_hat); 810 } 811 else { 812 struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z); 813 emit_op2(p, VP_OPCODE_ADD, half, 0, VPpli, z_dir); 814 } 815 816 emit_normalize_vec3(p, half, half); 817 818 release_temp(p, dist); 819 } 820 821 /* Calculate dot products: 822 */ 823 emit_op2(p, VP_OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli); 824 emit_op2(p, VP_OPCODE_DP3, dots, WRITEMASK_Y, normal, half); 825 826 827 /* Front face lighting: 828 */ 829 { 830 struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT); 831 struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE); 832 struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR); 833 struct ureg res0, res1; 834 835 emit_op1(p, VP_OPCODE_LIT, lit, 0, dots); 836 837 if (!is_undef(att)) 838 emit_op2(p, VP_OPCODE_MUL, lit, 0, lit, att); 839 840 841 if (count == nr_lights) { 842 if (separate) { 843 res0 = register_output( p, VERT_RESULT_COL0 ); 844 res1 = register_output( p, VERT_RESULT_COL1 ); 845 } 846 else { 847 res0 = _col0; 848 res1 = register_output( p, VERT_RESULT_COL0 ); 849 } 850 } else { 851 res0 = _col0; 852 res1 = _col1; 853 } 854 855 emit_op3(p, VP_OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0); 856 emit_op3(p, VP_OPCODE_MAD, res0, 0, swizzle1(lit,Y), diffuse, _col0); 857 emit_op3(p, VP_OPCODE_MAD, res1, 0, swizzle1(lit,Z), specular, _col1); 858 859 release_temp(p, ambient); 860 release_temp(p, diffuse); 861 release_temp(p, specular); 862 } 863 864 /* Back face lighting: 865 */ 866 if (twoside) { 867 struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT); 868 struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE); 869 struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR); 870 struct ureg res0, res1; 871 872 emit_op1(p, VP_OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z))); 873 874 if (!is_undef(att)) 875 emit_op2(p, VP_OPCODE_MUL, lit, 0, lit, att); 876 877 if (count == nr_lights) { 878 if (separate) { 879 res0 = register_output( p, VERT_RESULT_BFC0 ); 880 res1 = register_output( p, VERT_RESULT_BFC1 ); 881 } 882 else { 883 res0 = _bfc0; 884 res1 = register_output( p, VERT_RESULT_BFC0 ); 885 } 886 } else { 887 res0 = _bfc0; 888 res1 = _bfc1; 889 } 890 891 892 emit_op3(p, VP_OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0); 893 emit_op3(p, VP_OPCODE_MAD, res0, 0, swizzle1(lit,Y), diffuse, _bfc0); 894 emit_op3(p, VP_OPCODE_MAD, res1, 0, swizzle1(lit,Z), specular, _bfc1); 895 896 release_temp(p, ambient); 897 release_temp(p, diffuse); 898 release_temp(p, specular); 899 } 900 901 release_temp(p, half); 902 release_temp(p, VPpli); 903 release_temp(p, att); 904 } 905 } 906 907 release_temps( p ); 908} 909 910 911static void build_fog( struct tnl_program *p ) 912{ 913 GLcontext *ctx = p->ctx; 914 TNLcontext *tnl = TNL_CONTEXT(ctx); 915 struct ureg fog = register_output(p, VERT_RESULT_FOGC); 916 struct ureg input; 917 918 if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT) { 919 input = swizzle1(get_eye_position(p), Z); 920 } 921 else { 922 input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X); 923 } 924 925 if (tnl->_DoVertexFog) { 926 struct ureg params = register_param1(p, STATE_FOG_PARAMS); 927 struct ureg tmp = get_temp(p); 928 929 switch (ctx->Fog.Mode) { 930 case GL_LINEAR: { 931 struct ureg id = get_identity_param(p); 932 emit_op2(p, VP_OPCODE_SUB, tmp, 0, swizzle1(params,Z), input); 933 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, swizzle1(params,W)); 934 emit_op2(p, VP_OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */ 935 emit_op2(p, VP_OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W)); 936 break; 937 } 938 case GL_EXP: 939 emit_op1(p, VP_OPCODE_ABS, tmp, 0, input); 940 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, swizzle1(params,X)); 941 emit_op2(p, VP_OPCODE_POW, fog, WRITEMASK_X, 942 register_const1f(p, M_E), negate(tmp)); 943 break; 944 case GL_EXP2: 945 emit_op2(p, VP_OPCODE_MUL, tmp, 0, input, swizzle1(params,X)); 946 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, tmp); 947 emit_op2(p, VP_OPCODE_POW, fog, WRITEMASK_X, 948 register_const1f(p, M_E), negate(tmp)); 949 break; 950 } 951 952 release_temp(p, tmp); 953 } 954 else { 955 /* results = incoming fog coords (compute fog per-fragment later) 956 * 957 * KW: Is it really necessary to do anything in this case? 958 */ 959 emit_op1(p, VP_OPCODE_MOV, fog, WRITEMASK_X, input); 960 } 961} 962 963static void build_reflect_texgen( struct tnl_program *p, 964 struct ureg dest, 965 GLuint writemask ) 966{ 967 struct ureg normal = get_eye_normal(p); 968 struct ureg eye_hat = get_eye_position_normalized(p); 969 struct ureg tmp = get_temp(p); 970 971 /* n.u */ 972 emit_op2(p, VP_OPCODE_DP3, tmp, 0, normal, eye_hat); 973 /* 2n.u */ 974 emit_op2(p, VP_OPCODE_ADD, tmp, 0, tmp, tmp); 975 /* (-2n.u)n + u */ 976 emit_op3(p, VP_OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat); 977} 978 979static void build_sphere_texgen( struct tnl_program *p, 980 struct ureg dest, 981 GLuint writemask ) 982{ 983 struct ureg normal = get_eye_normal(p); 984 struct ureg eye_hat = get_eye_position_normalized(p); 985 struct ureg tmp = get_temp(p); 986 struct ureg half = register_const1f(p, .5); 987 struct ureg r = get_temp(p); 988 struct ureg inv_m = get_temp(p); 989 struct ureg id = get_identity_param(p); 990 991 /* Could share the above calculations, but it would be 992 * a fairly odd state for someone to set (both sphere and 993 * reflection active for different texture coordinate 994 * components. Of course - if two texture units enable 995 * reflect and/or sphere, things start to tilt in favour 996 * of seperating this out: 997 */ 998 999 /* n.u */ 1000 emit_op2(p, VP_OPCODE_DP3, tmp, 0, normal, eye_hat); 1001 /* 2n.u */ 1002 emit_op2(p, VP_OPCODE_ADD, tmp, 0, tmp, tmp); 1003 /* (-2n.u)n + u */ 1004 emit_op3(p, VP_OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat); 1005 /* r + 0,0,1 */ 1006 emit_op2(p, VP_OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z)); 1007 /* rx^2 + ry^2 + (rz+1)^2 */ 1008 emit_op2(p, VP_OPCODE_DP3, tmp, 0, tmp, tmp); 1009 /* 2/m */ 1010 emit_op1(p, VP_OPCODE_RSQ, tmp, 0, tmp); 1011 /* 1/m */ 1012 emit_op2(p, VP_OPCODE_MUL, inv_m, 0, tmp, swizzle1(half,X)); 1013 /* r/m + 1/2 */ 1014 emit_op3(p, VP_OPCODE_MAD, dest, writemask, r, inv_m, swizzle1(half,X)); 1015 1016 release_temp(p, tmp); 1017 release_temp(p, r); 1018 release_temp(p, inv_m); 1019} 1020 1021 1022static void build_texture_transform( struct tnl_program *p ) 1023{ 1024 GLcontext *ctx = p->ctx; 1025 GLuint i, j; 1026 1027 for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) { 1028 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i]; 1029 GLuint texmat_enabled = ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i); 1030 1031 if (texUnit->TexGenEnabled || texmat_enabled) { 1032 struct ureg out = register_output(p, VERT_RESULT_TEX0 + i); 1033 struct ureg out_texgen = undef; 1034 1035 if (texUnit->TexGenEnabled) { 1036 GLuint copy_mask = 0; 1037 GLuint sphere_mask = 0; 1038 GLuint reflect_mask = 0; 1039 GLuint normal_mask = 0; 1040 GLuint modes[4]; 1041 1042 if (texmat_enabled) 1043 out_texgen = get_temp(p); 1044 else 1045 out_texgen = out; 1046 1047 modes[0] = texUnit->GenModeS; 1048 modes[1] = texUnit->GenModeT; 1049 modes[2] = texUnit->GenModeR; 1050 modes[3] = texUnit->GenModeQ; 1051 1052 for (j = 0; j < 4; j++) { 1053 if (texUnit->TexGenEnabled & (1<<j)) { 1054 switch (modes[j]) { 1055 case GL_OBJECT_LINEAR: { 1056 struct ureg obj = register_input(p, VERT_ATTRIB_POS); 1057 struct ureg plane = 1058 register_param3(p, STATE_TEXGEN, i, 1059 STATE_TEXGEN_OBJECT_S + j); 1060 1061 emit_op2(p, VP_OPCODE_DP4, out_texgen, WRITEMASK_X << j, 1062 obj, plane ); 1063 break; 1064 } 1065 case GL_EYE_LINEAR: { 1066 struct ureg eye = get_eye_position(p); 1067 struct ureg plane = 1068 register_param3(p, STATE_TEXGEN, i, 1069 STATE_TEXGEN_EYE_S + j); 1070 1071 emit_op2(p, VP_OPCODE_DP4, out_texgen, WRITEMASK_X << j, 1072 eye, plane ); 1073 break; 1074 } 1075 case GL_SPHERE_MAP: 1076 sphere_mask |= WRITEMASK_X << j; 1077 break; 1078 case GL_REFLECTION_MAP_NV: 1079 reflect_mask |= WRITEMASK_X << j; 1080 break; 1081 case GL_NORMAL_MAP_NV: 1082 normal_mask |= WRITEMASK_X << j; 1083 break; 1084 } 1085 } 1086 else 1087 copy_mask |= WRITEMASK_X << j; 1088 } 1089 1090 1091 if (sphere_mask) { 1092 build_sphere_texgen(p, out_texgen, sphere_mask); 1093 } 1094 1095 if (reflect_mask) { 1096 build_reflect_texgen(p, out_texgen, reflect_mask); 1097 } 1098 1099 if (normal_mask) { 1100 struct ureg normal = get_eye_normal(p); 1101 emit_op1(p, VP_OPCODE_MOV, out_texgen, normal_mask, normal ); 1102 } 1103 1104 if (copy_mask) { 1105 struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i); 1106 emit_op1(p, VP_OPCODE_MOV, out_texgen, copy_mask, in ); 1107 } 1108 } 1109 1110 if (texmat_enabled) { 1111 struct ureg texmat[4]; 1112 struct ureg in = (!is_undef(out_texgen) ? 1113 out_texgen : 1114 register_input(p, VERT_ATTRIB_TEX0+i)); 1115 if (PREFER_DP4) { 1116 register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i, 1117 0, 3, STATE_MATRIX, texmat ); 1118 emit_matrix_transform_vec4( p, out, texmat, in ); 1119 } 1120 else { 1121 register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i, 1122 0, 3, STATE_MATRIX_TRANSPOSE, texmat ); 1123 emit_matrix_transform_vec4( p, out, texmat, in ); 1124 } 1125 } 1126 1127 release_temps(p); 1128 } 1129 else if (texUnit->_ReallyEnabled) { 1130 /* KW: _ReallyEnabled isn't sufficient? Need to know whether 1131 * this texture unit is referenced by the fragment shader. 1132 */ 1133 emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i); 1134 } 1135 } 1136} 1137 1138 1139/* Seems like it could be tighter: 1140 */ 1141static void build_pointsize( struct tnl_program *p ) 1142{ 1143 struct ureg eye = get_eye_position(p); 1144 struct ureg state_size = register_param1(p, STATE_POINT_SIZE); 1145 struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION); 1146 struct ureg out = register_output(p, VERT_RESULT_PSIZ); 1147 struct ureg ut = get_temp(p); 1148 1149 /* 1, -Z, Z * Z, 1 */ 1150 emit_op1(p, VP_OPCODE_MOV, ut, 0, swizzle1(get_identity_param(p), W)); 1151 emit_op2(p, VP_OPCODE_MUL, ut, WRITEMASK_YZ, ut, negate(swizzle1(eye, Z))); 1152 emit_op2(p, VP_OPCODE_MUL, ut, WRITEMASK_Z, ut, negate(swizzle1(eye, Z))); 1153 1154 1155 /* p1 + p2 * dist + p3 * dist * dist, 0 */ 1156 emit_op2(p, VP_OPCODE_DP3, ut, 0, ut, state_attenuation); 1157 1158 /* 1 / factor */ 1159 emit_op1(p, VP_OPCODE_RCP, ut, 0, ut ); 1160 1161 /* out = pointSize / factor */ 1162 emit_op2(p, VP_OPCODE_MUL, out, WRITEMASK_X, ut, state_size); 1163 1164 release_temp(p, ut); 1165} 1166 1167 1168static GLboolean programs_eq( struct vertex_program *a, 1169 struct vertex_program *b ) 1170{ 1171 if (!a || !b) 1172 return GL_FALSE; 1173 1174 if (a->Base.NumInstructions != b->Base.NumInstructions || 1175 a->Parameters->NumParameters != b->Parameters->NumParameters) 1176 return GL_FALSE; 1177 1178 if (memcmp(a->Instructions, b->Instructions, 1179 a->Base.NumInstructions * sizeof(struct vp_instruction)) != 0) 1180 return GL_FALSE; 1181 1182 if (memcmp(a->Parameters->Parameters, b->Parameters->Parameters, 1183 a->Parameters->NumParameters * 1184 sizeof(struct program_parameter)) != 0) 1185 return GL_FALSE; 1186 1187 return GL_TRUE; 1188} 1189 1190 1191void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx ) 1192{ 1193 struct tnl_program p; 1194 1195 if (ctx->VertexProgram._Enabled) 1196 return; 1197 1198 1199 _mesa_memset(&p, 0, sizeof(p)); 1200 p.ctx = ctx; 1201 p.program = (struct vertex_program *)ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0); 1202 p.eye_position = undef; 1203 p.eye_position_normalized = undef; 1204 p.eye_normal = undef; 1205 p.identity = undef; 1206 1207 p.temp_in_use = 0; 1208 1209 if (ctx->Const.MaxVertexProgramTemps >= sizeof(int) * 8) 1210 p.temp_reserved = 0; 1211 else 1212 p.temp_reserved = ~((1<<ctx->Const.MaxVertexProgramTemps)-1); 1213 1214 p.program->Instructions = MALLOC(sizeof(struct vp_instruction) * MAX_INSN); 1215 1216 /* Initialize the arb_program struct */ 1217 p.program->Base.String = 0; 1218 p.program->Base.NumInstructions = 1219 p.program->Base.NumTemporaries = 1220 p.program->Base.NumParameters = 1221 p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0; 1222 1223 if (p.program->Parameters) 1224 _mesa_free_parameters(p.program->Parameters); 1225 else 1226 p.program->Parameters = _mesa_new_parameter_list(); 1227 1228 p.program->InputsRead = 0; 1229 p.program->OutputsWritten = 0; 1230 1231 /* Emit the program, starting with modelviewproject: 1232 */ 1233 build_hpos(&p); 1234 1235 /* Lighting calculations: 1236 */ 1237 if (ctx->Light.Enabled) 1238 build_lighting(&p); 1239 else 1240 emit_passthrough(&p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0); 1241 1242 if (ctx->Fog.Enabled) 1243 build_fog(&p); 1244 1245 if (ctx->Texture._TexGenEnabled || ctx->Texture._TexMatEnabled) 1246 build_texture_transform(&p); 1247 1248 if (ctx->Point._Attenuated) 1249 build_pointsize(&p); 1250 1251 /* Finish up: 1252 */ 1253 emit_op1(&p, VP_OPCODE_END, undef, 0, undef); 1254 1255 /* Disassemble: 1256 */ 1257 if (DISASSEM) { 1258 _mesa_printf ("\n"); 1259 } 1260 1261 1262 /* Notify driver the fragment program has (actually) changed. 1263 */ 1264 if (!programs_eq(ctx->_TnlProgram, p.program) != 0) { 1265 if (ctx->_TnlProgram) 1266 ctx->Driver.DeleteProgram( ctx, &ctx->_TnlProgram->Base ); 1267 ctx->_TnlProgram = p.program; 1268 } 1269 else if (p.program) { 1270 /* Nothing changed... 1271 */ 1272 ctx->Driver.DeleteProgram( ctx, &p.program->Base ); 1273 } 1274} 1275