t_vp_build.c revision 3509fd8c1b38d955a066a2bad429dbfba162fa30
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#define MAX_INSN 200 53 54/* Use uregs to represent registers internally, translate to Mesa's 55 * expected formats on emit. 56 * 57 * NOTE: These are passed by value extensively in this file rather 58 * than as usual by pointer reference. If this disturbs you, try 59 * remembering they are just 32bits in size. 60 * 61 * GCC is smart enough to deal with these dword-sized structures in 62 * much the same way as if I had defined them as dwords and was using 63 * macros to access and set the fields. This is much nicer and easier 64 * to evolve. 65 */ 66struct ureg { 67 GLuint file:4; 68 GLuint idx:8; 69 GLuint negate:1; 70 GLuint swz:12; 71 GLuint pad:7; 72}; 73 74 75struct tnl_program { 76 GLcontext *ctx; 77 struct vertex_program *program; 78 79 GLuint temp_in_use; 80 GLuint temp_reserved; 81 82 struct ureg eye_position; 83 struct ureg eye_position_normalized; 84 struct ureg eye_normal; 85 struct ureg identity; 86 87 GLuint materials; 88 GLuint color_materials; 89}; 90 91 92const static struct ureg undef = { 93 ~0, 94 ~0, 95 0, 96 0, 97 0 98}; 99 100/* Local shorthand: 101 */ 102#define X SWIZZLE_X 103#define Y SWIZZLE_Y 104#define Z SWIZZLE_Z 105#define W SWIZZLE_W 106 107 108/* Construct a ureg: 109 */ 110static struct ureg make_ureg(GLuint file, GLuint idx) 111{ 112 struct ureg reg; 113 reg.file = file; 114 reg.idx = idx; 115 reg.negate = 0; 116 reg.swz = SWIZZLE_NOOP; 117 reg.pad = 0; 118 return reg; 119} 120 121 122 123static struct ureg negate( struct ureg reg ) 124{ 125 reg.negate ^= 1; 126 return reg; 127} 128 129 130static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w ) 131{ 132 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x), 133 GET_SWZ(reg.swz, y), 134 GET_SWZ(reg.swz, z), 135 GET_SWZ(reg.swz, w)); 136 137 return reg; 138} 139 140static struct ureg swizzle1( struct ureg reg, int x ) 141{ 142 return swizzle(reg, x, x, x, x); 143} 144 145static struct ureg get_temp( struct tnl_program *p ) 146{ 147 int bit = ffs( ~p->temp_in_use ); 148 if (!bit) { 149 fprintf(stderr, "%s: out of temporaries\n", __FILE__); 150 exit(1); 151 } 152 153 p->temp_in_use |= 1<<(bit-1); 154 return make_ureg(PROGRAM_TEMPORARY, bit-1); 155} 156 157static struct ureg reserve_temp( struct tnl_program *p ) 158{ 159 struct ureg temp = get_temp( p ); 160 p->temp_reserved |= 1<<temp.idx; 161 return temp; 162} 163 164static void release_temp( struct tnl_program *p, struct ureg reg ) 165{ 166 if (reg.file == PROGRAM_TEMPORARY) { 167 p->temp_in_use &= ~(1<<reg.idx); 168 p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */ 169 } 170} 171 172static void release_temps( struct tnl_program *p ) 173{ 174 p->temp_in_use = p->temp_reserved; 175} 176 177 178 179static struct ureg register_input( struct tnl_program *p, GLuint input ) 180{ 181 p->program->InputsRead |= (1<<input); 182 return make_ureg(PROGRAM_INPUT, input); 183} 184 185static struct ureg register_output( struct tnl_program *p, GLuint output ) 186{ 187 p->program->OutputsWritten |= (1<<output); 188 return make_ureg(PROGRAM_OUTPUT, output); 189} 190 191static struct ureg register_const4f( struct tnl_program *p, 192 GLfloat s0, 193 GLfloat s1, 194 GLfloat s2, 195 GLfloat s3) 196{ 197 GLfloat values[4]; 198 GLuint idx; 199 values[0] = s0; 200 values[1] = s1; 201 values[2] = s2; 202 values[3] = s3; 203 idx = _mesa_add_unnamed_constant( p->program->Parameters, values ); 204 return make_ureg(PROGRAM_STATE_VAR, idx); 205} 206 207#define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1) 208#define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1) 209#define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1) 210 211static GLboolean is_undef( struct ureg reg ) 212{ 213 return reg.file == 0xf; 214} 215 216static struct ureg get_identity_param( struct tnl_program *p ) 217{ 218 if (is_undef(p->identity)) 219 p->identity = register_const4f(p, 0,0,0,1); 220 221 return p->identity; 222} 223 224static struct ureg register_param6( struct tnl_program *p, 225 GLint s0, 226 GLint s1, 227 GLint s2, 228 GLint s3, 229 GLint s4, 230 GLint s5) 231{ 232 GLint tokens[6]; 233 GLuint idx; 234 tokens[0] = s0; 235 tokens[1] = s1; 236 tokens[2] = s2; 237 tokens[3] = s3; 238 tokens[4] = s4; 239 tokens[5] = s5; 240 idx = _mesa_add_state_reference( p->program->Parameters, tokens ); 241 return make_ureg(PROGRAM_STATE_VAR, idx); 242} 243 244 245#define register_param1(p,s0) register_param6(p,s0,0,0,0,0,0) 246#define register_param2(p,s0,s1) register_param6(p,s0,s1,0,0,0,0) 247#define register_param3(p,s0,s1,s2) register_param6(p,s0,s1,s2,0,0,0) 248#define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0) 249 250 251static void register_matrix_param6( struct tnl_program *p, 252 GLint s0, 253 GLint s1, 254 GLint s2, 255 GLint s3, 256 GLint s4, 257 GLint s5, 258 struct ureg *matrix ) 259{ 260 GLuint i; 261 262 /* This is a bit sad as the support is there to pull the whole 263 * matrix out in one go: 264 */ 265 for (i = 0; i <= s4 - s3; i++) 266 matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 ); 267} 268 269 270static void emit_arg( struct vp_src_register *src, 271 struct ureg reg ) 272{ 273 src->File = reg.file; 274 src->Index = reg.idx; 275 src->Swizzle = reg.swz; 276 src->Negate = reg.negate; 277 src->RelAddr = 0; 278 src->pad = 0; 279} 280 281static void emit_dst( struct vp_dst_register *dst, 282 struct ureg reg, GLuint mask ) 283{ 284 dst->File = reg.file; 285 dst->Index = reg.idx; 286 /* allow zero as a shorthand for xyzw */ 287 dst->WriteMask = mask ? mask : WRITEMASK_XYZW; 288 dst->pad = 0; 289} 290 291static void debug_insn( struct vp_instruction *inst, const char *fn, 292 GLuint line ) 293{ 294#if DISASSEM 295 static const char *last_fn; 296 297 if (fn != last_fn) { 298 last_fn = fn; 299 _mesa_printf("%s:\n", fn); 300 } 301 302 _mesa_printf("%d:\t", line); 303 _mesa_debug_vp_inst(1, inst); 304#endif 305} 306 307 308static void emit_op3fn(struct tnl_program *p, 309 GLuint op, 310 struct ureg dest, 311 GLuint mask, 312 struct ureg src0, 313 struct ureg src1, 314 struct ureg src2, 315 const char *fn, 316 GLuint line) 317{ 318 GLuint nr = p->program->Base.NumInstructions++; 319 struct vp_instruction *inst = &p->program->Instructions[nr]; 320 321 if (p->program->Base.NumInstructions > MAX_INSN) { 322 _mesa_problem(p->ctx, "Out of instructions in emit_op3fn\n"); 323 return; 324 } 325 326 inst->Opcode = op; 327 inst->StringPos = 0; 328 inst->Data = 0; 329 330 emit_arg( &inst->SrcReg[0], src0 ); 331 emit_arg( &inst->SrcReg[1], src1 ); 332 emit_arg( &inst->SrcReg[2], src2 ); 333 334 emit_dst( &inst->DstReg, dest, mask ); 335 336 debug_insn(inst, fn, line); 337} 338 339 340 341#define emit_op3(p, op, dst, mask, src0, src1, src2) \ 342 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__) 343 344#define emit_op2(p, op, dst, mask, src0, src1) \ 345 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__) 346 347#define emit_op1(p, op, dst, mask, src0) \ 348 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__) 349 350 351static struct ureg make_temp( struct tnl_program *p, struct ureg reg ) 352{ 353 if (reg.file == PROGRAM_TEMPORARY && 354 !(p->temp_reserved & (1<<reg.idx))) 355 return reg; 356 else { 357 struct ureg temp = get_temp(p); 358 emit_op1(p, VP_OPCODE_MOV, temp, 0, reg); 359 return temp; 360 } 361} 362 363 364/* Currently no tracking performed of input/output/register size or 365 * active elements. Could be used to reduce these operations, as 366 * could the matrix type. 367 */ 368static void emit_matrix_transform_vec4( struct tnl_program *p, 369 struct ureg dest, 370 const struct ureg *mat, 371 struct ureg src) 372{ 373 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]); 374 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]); 375 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]); 376 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]); 377} 378 379/* This version is much easier to implement if writemasks are not 380 * supported natively on the target or (like SSE), the target doesn't 381 * have a clean/obvious dotproduct implementation. 382 */ 383static void emit_transpose_matrix_transform_vec4( struct tnl_program *p, 384 struct ureg dest, 385 const struct ureg *mat, 386 struct ureg src) 387{ 388 struct ureg tmp; 389 390 if (dest.file != PROGRAM_TEMPORARY) 391 tmp = get_temp(p); 392 else 393 tmp = dest; 394 395 emit_op2(p, VP_OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]); 396 emit_op3(p, VP_OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp); 397 emit_op3(p, VP_OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp); 398 emit_op3(p, VP_OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp); 399 400 if (dest.file != PROGRAM_TEMPORARY) 401 release_temp(p, tmp); 402} 403 404static void emit_matrix_transform_vec3( struct tnl_program *p, 405 struct ureg dest, 406 const struct ureg *mat, 407 struct ureg src) 408{ 409 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]); 410 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]); 411 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]); 412} 413 414 415static void emit_normalize_vec3( struct tnl_program *p, 416 struct ureg dest, 417 struct ureg src ) 418{ 419 struct ureg tmp = get_temp(p); 420 emit_op2(p, VP_OPCODE_DP3, tmp, 0, src, src); 421 emit_op1(p, VP_OPCODE_RSQ, tmp, 0, tmp); 422 emit_op2(p, VP_OPCODE_MUL, dest, 0, src, tmp); 423 release_temp(p, tmp); 424} 425 426static struct ureg get_eye_position( struct tnl_program *p ) 427{ 428 if (is_undef(p->eye_position)) { 429 struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 430 struct ureg modelview[4]; 431 432 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3, 433 STATE_MATRIX_TRANSPOSE, modelview ); 434 p->eye_position = reserve_temp(p); 435 436 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos); 437 } 438 439 return p->eye_position; 440} 441 442 443static struct ureg get_eye_position_normalized( struct tnl_program *p ) 444{ 445 if (is_undef(p->eye_position_normalized)) { 446 struct ureg eye = get_eye_position(p); 447 p->eye_position_normalized = reserve_temp(p); 448 emit_normalize_vec3(p, p->eye_position_normalized, eye); 449 } 450 451 return p->eye_position_normalized; 452} 453 454 455static struct ureg get_eye_normal( struct tnl_program *p ) 456{ 457 if (is_undef(p->eye_normal)) { 458 struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL ); 459 struct ureg mvinv[3]; 460 461 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 2, 462 STATE_MATRIX_INVTRANS, mvinv ); 463 464 p->eye_normal = reserve_temp(p); 465 466 /* Transform to eye space: 467 */ 468 emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal ); 469 470 /* Normalize/Rescale: 471 */ 472 if (p->ctx->Transform.Normalize) { 473 emit_normalize_vec3( p, p->eye_normal, p->eye_normal ); 474 } 475 else if (p->ctx->Transform.RescaleNormals) { 476 struct ureg rescale = register_param2(p, STATE_INTERNAL, 477 STATE_NORMAL_SCALE); 478 479 emit_op2( p, VP_OPCODE_MUL, p->eye_normal, 0, normal, 480 swizzle1(rescale, X)); 481 } 482 } 483 484 return p->eye_normal; 485} 486 487 488 489static void build_hpos( struct tnl_program *p ) 490{ 491 struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 492 struct ureg hpos = register_output( p, VERT_RESULT_HPOS ); 493 struct ureg mvp[4]; 494 495 register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3, 496 STATE_MATRIX_TRANSPOSE, mvp ); 497 emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos ); 498} 499 500 501static GLuint material_attrib( GLuint side, GLuint property ) 502{ 503 return (_TNL_ATTRIB_MAT_FRONT_AMBIENT + 504 (property - STATE_AMBIENT) * 2 + 505 side); 506} 507 508static void set_material_flags( struct tnl_program *p ) 509{ 510 GLcontext *ctx = p->ctx; 511 TNLcontext *tnl = TNL_CONTEXT(ctx); 512 GLuint i; 513 514 p->color_materials = 0; 515 p->materials = 0; 516 517 if (ctx->Light.ColorMaterialEnabled) { 518 p->materials = 519 p->color_materials = 520 ctx->Light.ColorMaterialBitmask << _TNL_ATTRIB_MAT_FRONT_AMBIENT; 521 } 522 523 for (i = _TNL_ATTRIB_MAT_FRONT_AMBIENT ; i < _TNL_ATTRIB_INDEX ; i++) 524 if (tnl->vb.AttribPtr[i]->stride) 525 p->materials |= 1<<i; 526} 527 528 529static struct ureg get_material( struct tnl_program *p, GLuint side, 530 GLuint property ) 531{ 532 GLuint attrib = material_attrib(side, property); 533 534 if (p->color_materials & (1<<attrib)) 535 return register_input(p, VERT_ATTRIB_COLOR0); 536 else if (p->materials & (1<<attrib)) 537 return register_input( p, attrib ); 538 else 539 return register_param3( p, STATE_MATERIAL, side, property ); 540} 541 542#define SCENE_COLOR_BITS(side) (( _TNL_BIT_MAT_FRONT_EMISSION | \ 543 _TNL_BIT_MAT_FRONT_AMBIENT | \ 544 _TNL_BIT_MAT_FRONT_DIFFUSE) << (side)) 545 546/* Either return a precalculated constant value or emit code to 547 * calculate these values dynamically in the case where material calls 548 * are present between begin/end pairs. 549 * 550 * Probably want to shift this to the program compilation phase - if 551 * we always emitted the calculation here, a smart compiler could 552 * detect that it was constant (given a certain set of inputs), and 553 * lift it out of the main loop. That way the programs created here 554 * would be independent of the vertex_buffer details. 555 */ 556static struct ureg get_scenecolor( struct tnl_program *p, GLuint side ) 557{ 558 if (p->materials & SCENE_COLOR_BITS(side)) { 559 struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT); 560 struct ureg material_emission = get_material(p, side, STATE_EMISSION); 561 struct ureg material_ambient = get_material(p, side, STATE_AMBIENT); 562 struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE); 563 struct ureg tmp = make_temp(p, material_diffuse); 564 emit_op3(p, VP_OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient, 565 material_ambient, material_emission); 566 return tmp; 567 } 568 else 569 return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side ); 570} 571 572 573static struct ureg get_lightprod( struct tnl_program *p, GLuint light, 574 GLuint side, GLuint property ) 575{ 576 GLuint attrib = material_attrib(side, property); 577 if (p->materials & (1<<attrib)) { 578 struct ureg light_value = 579 register_param3(p, STATE_LIGHT, light, property); 580 struct ureg material_value = get_material(p, side, property); 581 struct ureg tmp = get_temp(p); 582 emit_op2(p, VP_OPCODE_MUL, tmp, 0, light_value, material_value); 583 return tmp; 584 } 585 else 586 return register_param4(p, STATE_LIGHTPROD, light, side, property); 587} 588 589static struct ureg calculate_light_attenuation( struct tnl_program *p, 590 GLuint i, 591 struct gl_light *light, 592 struct ureg VPpli, 593 struct ureg dist ) 594{ 595 struct ureg attenuation = register_param3(p, STATE_LIGHT, i, 596 STATE_ATTENUATION); 597 struct ureg att = get_temp(p); 598 599 /* Calculate spot attenuation: 600 */ 601 if (light->SpotCutoff != 180.0F) { 602 struct ureg spot_dir = register_param3(p, STATE_LIGHT, i, 603 STATE_SPOT_DIRECTION); 604 struct ureg spot = get_temp(p); 605 struct ureg slt = get_temp(p); 606 607 emit_normalize_vec3( p, spot, spot_dir ); /* XXX: precompute! */ 608 emit_op2(p, VP_OPCODE_DP3, spot, 0, negate(VPpli), spot_dir); 609 emit_op2(p, VP_OPCODE_SLT, slt, 0, swizzle1(spot_dir,W), spot); 610 emit_op2(p, VP_OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W)); 611 emit_op2(p, VP_OPCODE_MUL, att, 0, slt, spot); 612 613 release_temp(p, spot); 614 release_temp(p, slt); 615 } 616 617 /* Calculate distance attenuation: 618 */ 619 if (light->ConstantAttenuation != 1.0 || 620 light->LinearAttenuation != 1.0 || 621 light->QuadraticAttenuation != 1.0) { 622 623 /* 1/d,d,d,1/d */ 624 emit_op1(p, VP_OPCODE_RCP, dist, WRITEMASK_YZ, dist); 625 /* 1,d,d*d,1/d */ 626 emit_op2(p, VP_OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y)); 627 /* 1/dist-atten */ 628 emit_op2(p, VP_OPCODE_DP3, dist, 0, attenuation, dist); 629 630 if (light->SpotCutoff != 180.0F) { 631 /* dist-atten */ 632 emit_op1(p, VP_OPCODE_RCP, dist, 0, dist); 633 /* spot-atten * dist-atten */ 634 emit_op2(p, VP_OPCODE_MUL, att, 0, dist, att); 635 } else { 636 /* dist-atten */ 637 emit_op1(p, VP_OPCODE_RCP, att, 0, dist); 638 } 639 } 640 641 return att; 642} 643 644 645 646 647 648/* Need to add some addtional parameters to allow lighting in object 649 * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye 650 * space lighting. 651 */ 652static void build_lighting( struct tnl_program *p ) 653{ 654 GLcontext *ctx = p->ctx; 655 const GLboolean twoside = ctx->Light.Model.TwoSide; 656 const GLboolean separate = (ctx->Light.Model.ColorControl == 657 GL_SEPARATE_SPECULAR_COLOR); 658 GLuint nr_lights = 0, count = 0; 659 struct ureg normal = get_eye_normal(p); 660 struct ureg lit = get_temp(p); 661 struct ureg dots = get_temp(p); 662 struct ureg _col0 = undef, _col1 = undef; 663 struct ureg _bfc0 = undef, _bfc1 = undef; 664 GLuint i; 665 666 for (i = 0; i < MAX_LIGHTS; i++) 667 if (ctx->Light.Light[i].Enabled) 668 nr_lights++; 669 670 set_material_flags(p); 671 672 { 673 struct ureg shininess = get_material(p, 0, STATE_SHININESS); 674 emit_op1(p, VP_OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X)); 675 release_temp(p, shininess); 676 677 _col0 = make_temp(p, get_scenecolor(p, 0)); 678 if (separate) 679 _col1 = make_temp(p, get_identity_param(p)); 680 else 681 _col1 = _col0; 682 683 } 684 685 if (twoside) { 686 struct ureg shininess = get_material(p, 1, STATE_SHININESS); 687 emit_op1(p, VP_OPCODE_MOV, dots, WRITEMASK_Z, 688 negate(swizzle1(shininess,X))); 689 release_temp(p, shininess); 690 691 _bfc0 = make_temp(p, get_scenecolor(p, 1)); 692 if (separate) 693 _bfc1 = make_temp(p, get_identity_param(p)); 694 else 695 _bfc1 = _bfc0; 696 } 697 698 for (i = 0; i < MAX_LIGHTS; i++) { 699 struct gl_light *light = &ctx->Light.Light[i]; 700 701 if (light->Enabled) { 702 struct ureg half = undef; 703 struct ureg att = undef, VPpli = undef; 704 705 count++; 706 707 if (light->EyePosition[3] == 0) { 708 /* Can used precomputed constants in this case. 709 * Attenuation never applies to infinite lights. 710 */ 711 VPpli = register_param3(p, STATE_LIGHT, i, 712 STATE_POSITION_NORMALIZED); 713 half = register_param3(p, STATE_LIGHT, i, STATE_HALF); 714 } 715 else { 716 struct ureg Ppli = register_param3(p, STATE_LIGHT, i, 717 STATE_POSITION); 718 struct ureg V = get_eye_position(p); 719 struct ureg dist = get_temp(p); 720 721 VPpli = get_temp(p); 722 half = get_temp(p); 723 724 /* Calulate VPpli vector 725 */ 726 emit_op2(p, VP_OPCODE_SUB, VPpli, 0, Ppli, V); 727 728 /* Normalize VPpli. The dist value also used in 729 * attenuation below. 730 */ 731 emit_op2(p, VP_OPCODE_DP3, dist, 0, VPpli, VPpli); 732 emit_op1(p, VP_OPCODE_RSQ, dist, 0, dist); 733 emit_op2(p, VP_OPCODE_MUL, VPpli, 0, VPpli, dist); 734 735 736 /* Calculate attenuation: 737 */ 738 if (light->SpotCutoff != 180.0 || 739 light->ConstantAttenuation != 1.0 || 740 light->LinearAttenuation != 1.0 || 741 light->QuadraticAttenuation != 1.0) { 742 att = calculate_light_attenuation(p, i, light, VPpli, dist); 743 } 744 745 746 /* Calculate viewer direction, or use infinite viewer: 747 */ 748 if (ctx->Light.Model.LocalViewer) { 749 struct ureg eye_hat = get_eye_position_normalized(p); 750 emit_op2(p, VP_OPCODE_SUB, half, 0, VPpli, eye_hat); 751 } 752 else { 753 struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z); 754 emit_op2(p, VP_OPCODE_ADD, half, 0, VPpli, z_dir); 755 } 756 757 emit_normalize_vec3(p, half, half); 758 759 release_temp(p, dist); 760 } 761 762 /* Calculate dot products: 763 */ 764 emit_op2(p, VP_OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli); 765 emit_op2(p, VP_OPCODE_DP3, dots, WRITEMASK_Y, normal, half); 766 767 768 /* Front face lighting: 769 */ 770 { 771 struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT); 772 struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE); 773 struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR); 774 struct ureg res0, res1; 775 776 emit_op1(p, VP_OPCODE_LIT, lit, 0, dots); 777 778 if (!is_undef(att)) 779 emit_op2(p, VP_OPCODE_MUL, lit, 0, lit, att); 780 781 782 if (count == nr_lights) { 783 if (separate) { 784 res0 = register_output( p, VERT_RESULT_COL0 ); 785 res1 = register_output( p, VERT_RESULT_COL1 ); 786 } 787 else { 788 res0 = _col0; 789 res1 = register_output( p, VERT_RESULT_COL0 ); 790 } 791 } else { 792 res0 = _col0; 793 res1 = _col1; 794 } 795 796 emit_op3(p, VP_OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0); 797 emit_op3(p, VP_OPCODE_MAD, res0, 0, swizzle1(lit,Y), diffuse, _col0); 798 emit_op3(p, VP_OPCODE_MAD, res1, 0, swizzle1(lit,Z), specular, _col1); 799 800 release_temp(p, ambient); 801 release_temp(p, diffuse); 802 release_temp(p, specular); 803 } 804 805 /* Back face lighting: 806 */ 807 if (twoside) { 808 struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT); 809 struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE); 810 struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR); 811 struct ureg res0, res1; 812 813 emit_op1(p, VP_OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z))); 814 815 if (!is_undef(att)) 816 emit_op2(p, VP_OPCODE_MUL, lit, 0, lit, att); 817 818 if (count == nr_lights) { 819 if (separate) { 820 res0 = register_output( p, VERT_RESULT_BFC0 ); 821 res1 = register_output( p, VERT_RESULT_BFC1 ); 822 } 823 else { 824 res0 = _bfc0; 825 res1 = register_output( p, VERT_RESULT_BFC0 ); 826 } 827 } else { 828 res0 = _bfc0; 829 res1 = _bfc1; 830 } 831 832 833 emit_op3(p, VP_OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0); 834 emit_op3(p, VP_OPCODE_MAD, res0, 0, swizzle1(lit,Y), diffuse, _bfc0); 835 emit_op3(p, VP_OPCODE_MAD, res1, 0, swizzle1(lit,Z), specular, _bfc1); 836 837 release_temp(p, ambient); 838 release_temp(p, diffuse); 839 release_temp(p, specular); 840 } 841 842 release_temp(p, half); 843 release_temp(p, VPpli); 844 release_temp(p, att); 845 } 846 } 847 848 release_temps( p ); 849} 850 851 852static void build_fog( struct tnl_program *p ) 853{ 854 GLcontext *ctx = p->ctx; 855 TNLcontext *tnl = TNL_CONTEXT(ctx); 856 struct ureg fog = register_output(p, VERT_RESULT_FOGC); 857 struct ureg input; 858 859 if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT) { 860 input = swizzle1(get_eye_position(p), Z); 861 } 862 else { 863 input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X); 864 } 865 866 if (tnl->_DoVertexFog) { 867 struct ureg params = register_param1(p, STATE_FOG_PARAMS); 868 struct ureg tmp = get_temp(p); 869 870 switch (ctx->Fog.Mode) { 871 case GL_LINEAR: { 872 struct ureg id = get_identity_param(p); 873 emit_op2(p, VP_OPCODE_SUB, tmp, 0, swizzle1(params,Z), input); 874 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, swizzle1(params,W)); 875 emit_op2(p, VP_OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */ 876 emit_op2(p, VP_OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W)); 877 break; 878 } 879 case GL_EXP: 880 emit_op1(p, VP_OPCODE_ABS, tmp, 0, input); 881 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, swizzle1(params,X)); 882 emit_op2(p, VP_OPCODE_POW, fog, WRITEMASK_X, 883 register_const1f(p, M_E), negate(tmp)); 884 break; 885 case GL_EXP2: 886 emit_op2(p, VP_OPCODE_MUL, tmp, 0, input, swizzle1(params,X)); 887 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, tmp); 888 emit_op2(p, VP_OPCODE_POW, fog, WRITEMASK_X, 889 register_const1f(p, M_E), negate(tmp)); 890 break; 891 } 892 893 release_temp(p, tmp); 894 } 895 else { 896 /* results = incoming fog coords (compute fog per-fragment later) 897 * 898 * KW: Is it really necessary to do anything in this case? 899 */ 900 emit_op1(p, VP_OPCODE_MOV, fog, WRITEMASK_X, input); 901 } 902} 903 904static void build_reflect_texgen( struct tnl_program *p, 905 struct ureg dest, 906 GLuint writemask ) 907{ 908 struct ureg normal = get_eye_normal(p); 909 struct ureg eye_hat = get_eye_position_normalized(p); 910 struct ureg tmp = get_temp(p); 911 912 /* n.u */ 913 emit_op2(p, VP_OPCODE_DP3, tmp, 0, normal, eye_hat); 914 /* 2n.u */ 915 emit_op2(p, VP_OPCODE_ADD, tmp, 0, tmp, tmp); 916 /* (-2n.u)n + u */ 917 emit_op3(p, VP_OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat); 918} 919 920static void build_sphere_texgen( struct tnl_program *p, 921 struct ureg dest, 922 GLuint writemask ) 923{ 924 struct ureg normal = get_eye_normal(p); 925 struct ureg eye_hat = get_eye_position_normalized(p); 926 struct ureg tmp = get_temp(p); 927 struct ureg half = register_const1f(p, .5); 928 struct ureg r = get_temp(p); 929 struct ureg inv_m = get_temp(p); 930 struct ureg id = get_identity_param(p); 931 932 /* Could share the above calculations, but it would be 933 * a fairly odd state for someone to set (both sphere and 934 * reflection active for different texture coordinate 935 * components. Of course - if two texture units enable 936 * reflect and/or sphere, things start to tilt in favour 937 * of seperating this out: 938 */ 939 940 /* n.u */ 941 emit_op2(p, VP_OPCODE_DP3, tmp, 0, normal, eye_hat); 942 /* 2n.u */ 943 emit_op2(p, VP_OPCODE_ADD, tmp, 0, tmp, tmp); 944 /* (-2n.u)n + u */ 945 emit_op3(p, VP_OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat); 946 /* r + 0,0,1 */ 947 emit_op2(p, VP_OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z)); 948 /* rx^2 + ry^2 + (rz+1)^2 */ 949 emit_op2(p, VP_OPCODE_DP3, tmp, 0, tmp, tmp); 950 /* 2/m */ 951 emit_op1(p, VP_OPCODE_RSQ, tmp, 0, tmp); 952 /* 1/m */ 953 emit_op2(p, VP_OPCODE_MUL, inv_m, 0, tmp, swizzle1(half,X)); 954 /* r/m + 1/2 */ 955 emit_op3(p, VP_OPCODE_MAD, dest, writemask, r, inv_m, swizzle1(half,X)); 956 957 release_temp(p, tmp); 958 release_temp(p, r); 959 release_temp(p, inv_m); 960} 961 962 963static void build_texture_transform( struct tnl_program *p ) 964{ 965 GLcontext *ctx = p->ctx; 966 GLuint i, j; 967 968 for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) { 969 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i]; 970 GLuint texmat_enabled = ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i); 971 struct ureg out = register_output(p, VERT_RESULT_TEX0 + i); 972 973 if (texUnit->TexGenEnabled || texmat_enabled) { 974 struct ureg out_texgen = undef; 975 976 if (texUnit->TexGenEnabled) { 977 GLuint copy_mask = 0; 978 GLuint sphere_mask = 0; 979 GLuint reflect_mask = 0; 980 GLuint normal_mask = 0; 981 GLuint modes[4]; 982 983 if (texmat_enabled) 984 out_texgen = get_temp(p); 985 else 986 out_texgen = out; 987 988 modes[0] = texUnit->GenModeS; 989 modes[1] = texUnit->GenModeT; 990 modes[2] = texUnit->GenModeR; 991 modes[3] = texUnit->GenModeQ; 992 993 for (j = 0; j < 4; j++) { 994 if (texUnit->TexGenEnabled & (1<<j)) { 995 switch (modes[j]) { 996 case GL_OBJECT_LINEAR: { 997 struct ureg obj = register_input(p, VERT_ATTRIB_POS); 998 struct ureg plane = 999 register_param3(p, STATE_TEXGEN, i, 1000 STATE_TEXGEN_OBJECT_S + j); 1001 1002 emit_op2(p, VP_OPCODE_DP4, out_texgen, WRITEMASK_X << j, 1003 obj, plane ); 1004 break; 1005 } 1006 case GL_EYE_LINEAR: { 1007 struct ureg eye = get_eye_position(p); 1008 struct ureg plane = 1009 register_param3(p, STATE_TEXGEN, i, 1010 STATE_TEXGEN_EYE_S + j); 1011 1012 emit_op2(p, VP_OPCODE_DP4, out_texgen, WRITEMASK_X << j, 1013 eye, plane ); 1014 break; 1015 } 1016 case GL_SPHERE_MAP: 1017 sphere_mask |= WRITEMASK_X << j; 1018 break; 1019 case GL_REFLECTION_MAP_NV: 1020 reflect_mask |= WRITEMASK_X << j; 1021 break; 1022 case GL_NORMAL_MAP_NV: 1023 normal_mask |= WRITEMASK_X << j; 1024 break; 1025 } 1026 } 1027 else 1028 copy_mask |= WRITEMASK_X << j; 1029 } 1030 1031 1032 if (sphere_mask) { 1033 build_sphere_texgen(p, out_texgen, sphere_mask); 1034 } 1035 1036 if (reflect_mask) { 1037 build_reflect_texgen(p, out_texgen, reflect_mask); 1038 } 1039 1040 if (normal_mask) { 1041 struct ureg normal = get_eye_normal(p); 1042 emit_op1(p, VP_OPCODE_MOV, out_texgen, normal_mask, normal ); 1043 } 1044 1045 if (copy_mask) { 1046 struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i); 1047 emit_op1(p, VP_OPCODE_MOV, out_texgen, copy_mask, in ); 1048 } 1049 } 1050 1051 if (texmat_enabled) { 1052 struct ureg texmat[4]; 1053 struct ureg in = (!is_undef(out_texgen) ? 1054 out_texgen : 1055 register_input(p, VERT_ATTRIB_TEX0+i)); 1056 register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i, 1057 0, 3, 0, texmat ); 1058 emit_matrix_transform_vec4( p, out, texmat, in ); 1059 } 1060 1061 release_temps(p); 1062 } 1063 } 1064} 1065 1066 1067/* Seems like it could be tighter: 1068 */ 1069static void build_pointsize( struct tnl_program *p ) 1070{ 1071 struct ureg eye = get_eye_position(p); 1072 struct ureg state_size = register_param1(p, STATE_POINT_SIZE); 1073 struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION); 1074 struct ureg out = register_output(p, VERT_RESULT_PSIZ); 1075 struct ureg ut = get_temp(p); 1076 1077 /* 1, -Z, Z * Z, 1 */ 1078 emit_op1(p, VP_OPCODE_MOV, ut, 0, swizzle1(get_identity_param(p), W)); 1079 emit_op2(p, VP_OPCODE_MUL, ut, WRITEMASK_YZ, ut, negate(swizzle1(eye, Z))); 1080 emit_op2(p, VP_OPCODE_MUL, ut, WRITEMASK_Z, ut, negate(swizzle1(eye, Z))); 1081 1082 1083 /* p1 + p2 * dist + p3 * dist * dist, 0 */ 1084 emit_op2(p, VP_OPCODE_DP3, ut, 0, ut, state_attenuation); 1085 1086 /* 1 / factor */ 1087 emit_op1(p, VP_OPCODE_RCP, ut, 0, ut ); 1088 1089 /* out = pointSize / factor */ 1090 emit_op2(p, VP_OPCODE_MUL, out, WRITEMASK_X, ut, state_size); 1091 1092 release_temp(p, ut); 1093} 1094 1095 1096static void build_passthrough( struct tnl_program *p, GLuint inputs ) 1097{ 1098} 1099 1100 1101static GLboolean programs_eq( struct vertex_program *a, 1102 struct vertex_program *b ) 1103{ 1104 if (!a || !b) 1105 return GL_FALSE; 1106 1107 if (a->Base.NumInstructions != b->Base.NumInstructions || 1108 a->Parameters->NumParameters != b->Parameters->NumParameters) 1109 return GL_FALSE; 1110 1111 if (memcmp(a->Instructions, b->Instructions, 1112 a->Base.NumInstructions * sizeof(struct vp_instruction)) != 0) 1113 return GL_FALSE; 1114 1115 if (memcmp(a->Parameters->Parameters, b->Parameters->Parameters, 1116 a->Parameters->NumParameters * 1117 sizeof(struct program_parameter)) != 0) 1118 return GL_FALSE; 1119 1120 return GL_TRUE; 1121} 1122 1123 1124void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx ) 1125{ 1126 TNLcontext *tnl = TNL_CONTEXT(ctx); 1127 struct tnl_program p; 1128 1129 if (ctx->VertexProgram._Enabled) 1130 return; 1131 1132 1133 memset(&p, 0, sizeof(p)); 1134 p.ctx = ctx; 1135 p.program = (struct vertex_program *)ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0); 1136 p.eye_position = undef; 1137 p.eye_position_normalized = undef; 1138 p.eye_normal = undef; 1139 p.identity = undef; 1140 1141 p.temp_in_use = 0; 1142 1143 if (ctx->Const.MaxVertexProgramTemps >= sizeof(int) * 8) 1144 p.temp_reserved = 0; 1145 else 1146 p.temp_reserved = ~((1<<ctx->Const.MaxVertexProgramTemps)-1); 1147 1148 p.program->Instructions = MALLOC(sizeof(struct vp_instruction) * MAX_INSN); 1149 1150 /* Initialize the arb_program struct */ 1151 p.program->Base.String = 0; 1152 p.program->Base.NumInstructions = 1153 p.program->Base.NumTemporaries = 1154 p.program->Base.NumParameters = 1155 p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0; 1156 1157 if (p.program->Parameters) 1158 _mesa_free_parameters(p.program->Parameters); 1159 else 1160 p.program->Parameters = _mesa_new_parameter_list(); 1161 1162 p.program->InputsRead = 0; 1163 p.program->OutputsWritten = 0; 1164 1165 /* Emit the program, starting with modelviewproject: 1166 */ 1167 build_hpos(&p); 1168 1169 /* Lighting calculations: 1170 */ 1171 if (ctx->Light.Enabled) 1172 build_lighting(&p); 1173 1174 if (ctx->Fog.Enabled) 1175 build_fog(&p); 1176 1177 if (ctx->Texture._TexGenEnabled || ctx->Texture._TexMatEnabled) 1178 build_texture_transform(&p); 1179 1180 if (ctx->Point._Attenuated) 1181 build_pointsize(&p); 1182 1183 /* Is there a need to copy inputs to outputs? The software 1184 * implementation might do this more efficiently by just assigning 1185 * the missing results to point at input arrays. 1186 */ 1187 if (/* tnl->vp_copy_inputs && */ 1188 (tnl->render_inputs & ~p.program->OutputsWritten)) { 1189 build_passthrough(&p, tnl->render_inputs); 1190 } 1191 1192 1193 /* Finish up: 1194 */ 1195 emit_op1(&p, VP_OPCODE_END, undef, 0, undef); 1196 1197 /* Disassemble: 1198 */ 1199 if (DISASSEM) { 1200 _mesa_printf ("\n"); 1201 } 1202 1203 1204 /* Notify driver the fragment program has (actually) changed. 1205 */ 1206 if (!programs_eq(ctx->_TnlProgram, p.program) != 0) { 1207 if (ctx->_TnlProgram) 1208 ctx->Driver.DeleteProgram( ctx, &ctx->_TnlProgram->Base ); 1209 ctx->_TnlProgram = p.program; 1210 } 1211 else if (p.program) { 1212 /* Nothing changed... 1213 */ 1214 ctx->Driver.DeleteProgram( ctx, &p.program->Base ); 1215 } 1216} 1217