1 /************************************************************************** 2 * 3 * Copyright 2007 VMware, Inc. 4 * All Rights Reserved. 5 * Copyright 2009 VMware, Inc. All Rights Reserved. 6 * Copyright 2010-2011 Intel Corporation 7 * 8 * Permission is hereby granted, free of charge, to any person obtaining a 9 * copy of this software and associated documentation files (the 10 * "Software"), to deal in the Software without restriction, including 11 * without limitation the rights to use, copy, modify, merge, publish, 12 * distribute, sub license, and/or sell copies of the Software, and to 13 * permit persons to whom the Software is furnished to do so, subject to 14 * the following conditions: 15 * 16 * The above copyright notice and this permission notice (including the 17 * next paragraph) shall be included in all copies or substantial portions 18 * of the Software. 19 * 20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 21 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 23 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR 24 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 25 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 26 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 27 * 28 **************************************************************************/ 29 30 #include "main/glheader.h" 31 #include "main/context.h" 32 #include "main/imports.h" 33 #include "main/macros.h" 34 #include "main/samplerobj.h" 35 #include "main/shaderobj.h" 36 #include "main/texenvprogram.h" 37 #include "main/texobj.h" 38 #include "main/uniforms.h" 39 #include "compiler/glsl/ir_builder.h" 40 #include "compiler/glsl/ir_optimization.h" 41 #include "compiler/glsl/glsl_parser_extras.h" 42 #include "compiler/glsl/glsl_symbol_table.h" 43 #include "compiler/glsl_types.h" 44 #include "program/ir_to_mesa.h" 45 #include "program/program.h" 46 #include "program/programopt.h" 47 #include "program/prog_cache.h" 48 #include "program/prog_instruction.h" 49 #include "program/prog_parameter.h" 50 #include "program/prog_print.h" 51 #include "program/prog_statevars.h" 52 #include "util/bitscan.h" 53 54 using namespace ir_builder; 55 56 /* 57 * Note on texture units: 58 * 59 * The number of texture units supported by fixed-function fragment 60 * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS. 61 * That's because there's a one-to-one correspondence between texture 62 * coordinates and samplers in fixed-function processing. 63 * 64 * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS 65 * sets of texcoords, so is fixed-function fragment processing. 66 * 67 * We can safely use ctx->Const.MaxTextureUnits for loop bounds. 68 */ 69 70 71 struct texenvprog_cache_item 72 { 73 GLuint hash; 74 void *key; 75 struct gl_shader_program *data; 76 struct texenvprog_cache_item *next; 77 }; 78 79 static GLboolean 80 texenv_doing_secondary_color(struct gl_context *ctx) 81 { 82 if (ctx->Light.Enabled && 83 (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR)) 84 return GL_TRUE; 85 86 if (ctx->Fog.ColorSumEnabled) 87 return GL_TRUE; 88 89 return GL_FALSE; 90 } 91 92 struct mode_opt { 93 #ifdef __GNUC__ 94 __extension__ GLubyte Source:4; /**< SRC_x */ 95 __extension__ GLubyte Operand:3; /**< OPR_x */ 96 #else 97 GLubyte Source; /**< SRC_x */ 98 GLubyte Operand; /**< OPR_x */ 99 #endif 100 }; 101 102 struct state_key { 103 GLuint nr_enabled_units:8; 104 GLuint enabled_units:8; 105 GLuint separate_specular:1; 106 GLuint fog_mode:2; /**< FOG_x */ 107 GLuint inputs_available:12; 108 GLuint num_draw_buffers:4; 109 110 /* NOTE: This array of structs must be last! (see "keySize" below) */ 111 struct { 112 GLuint enabled:1; 113 GLuint source_index:4; /**< TEXTURE_x_INDEX */ 114 GLuint shadow:1; 115 GLuint ScaleShiftRGB:2; 116 GLuint ScaleShiftA:2; 117 118 GLuint NumArgsRGB:3; /**< up to MAX_COMBINER_TERMS */ 119 GLuint ModeRGB:5; /**< MODE_x */ 120 121 GLuint NumArgsA:3; /**< up to MAX_COMBINER_TERMS */ 122 GLuint ModeA:5; /**< MODE_x */ 123 124 struct mode_opt OptRGB[MAX_COMBINER_TERMS]; 125 struct mode_opt OptA[MAX_COMBINER_TERMS]; 126 } unit[MAX_TEXTURE_UNITS]; 127 }; 128 129 #define FOG_NONE 0 130 #define FOG_LINEAR 1 131 #define FOG_EXP 2 132 #define FOG_EXP2 3 133 134 static GLuint translate_fog_mode( GLenum mode ) 135 { 136 switch (mode) { 137 case GL_LINEAR: return FOG_LINEAR; 138 case GL_EXP: return FOG_EXP; 139 case GL_EXP2: return FOG_EXP2; 140 default: return FOG_NONE; 141 } 142 } 143 144 #define OPR_SRC_COLOR 0 145 #define OPR_ONE_MINUS_SRC_COLOR 1 146 #define OPR_SRC_ALPHA 2 147 #define OPR_ONE_MINUS_SRC_ALPHA 3 148 #define OPR_ZERO 4 149 #define OPR_ONE 5 150 #define OPR_UNKNOWN 7 151 152 static GLuint translate_operand( GLenum operand ) 153 { 154 switch (operand) { 155 case GL_SRC_COLOR: return OPR_SRC_COLOR; 156 case GL_ONE_MINUS_SRC_COLOR: return OPR_ONE_MINUS_SRC_COLOR; 157 case GL_SRC_ALPHA: return OPR_SRC_ALPHA; 158 case GL_ONE_MINUS_SRC_ALPHA: return OPR_ONE_MINUS_SRC_ALPHA; 159 case GL_ZERO: return OPR_ZERO; 160 case GL_ONE: return OPR_ONE; 161 default: 162 assert(0); 163 return OPR_UNKNOWN; 164 } 165 } 166 167 #define SRC_TEXTURE 0 168 #define SRC_TEXTURE0 1 169 #define SRC_TEXTURE1 2 170 #define SRC_TEXTURE2 3 171 #define SRC_TEXTURE3 4 172 #define SRC_TEXTURE4 5 173 #define SRC_TEXTURE5 6 174 #define SRC_TEXTURE6 7 175 #define SRC_TEXTURE7 8 176 #define SRC_CONSTANT 9 177 #define SRC_PRIMARY_COLOR 10 178 #define SRC_PREVIOUS 11 179 #define SRC_ZERO 12 180 #define SRC_UNKNOWN 15 181 182 static GLuint translate_source( GLenum src ) 183 { 184 switch (src) { 185 case GL_TEXTURE: return SRC_TEXTURE; 186 case GL_TEXTURE0: 187 case GL_TEXTURE1: 188 case GL_TEXTURE2: 189 case GL_TEXTURE3: 190 case GL_TEXTURE4: 191 case GL_TEXTURE5: 192 case GL_TEXTURE6: 193 case GL_TEXTURE7: return SRC_TEXTURE0 + (src - GL_TEXTURE0); 194 case GL_CONSTANT: return SRC_CONSTANT; 195 case GL_PRIMARY_COLOR: return SRC_PRIMARY_COLOR; 196 case GL_PREVIOUS: return SRC_PREVIOUS; 197 case GL_ZERO: 198 return SRC_ZERO; 199 default: 200 assert(0); 201 return SRC_UNKNOWN; 202 } 203 } 204 205 #define MODE_REPLACE 0 /* r = a0 */ 206 #define MODE_MODULATE 1 /* r = a0 * a1 */ 207 #define MODE_ADD 2 /* r = a0 + a1 */ 208 #define MODE_ADD_SIGNED 3 /* r = a0 + a1 - 0.5 */ 209 #define MODE_INTERPOLATE 4 /* r = a0 * a2 + a1 * (1 - a2) */ 210 #define MODE_SUBTRACT 5 /* r = a0 - a1 */ 211 #define MODE_DOT3_RGB 6 /* r = a0 . a1 */ 212 #define MODE_DOT3_RGB_EXT 7 /* r = a0 . a1 */ 213 #define MODE_DOT3_RGBA 8 /* r = a0 . a1 */ 214 #define MODE_DOT3_RGBA_EXT 9 /* r = a0 . a1 */ 215 #define MODE_MODULATE_ADD_ATI 10 /* r = a0 * a2 + a1 */ 216 #define MODE_MODULATE_SIGNED_ADD_ATI 11 /* r = a0 * a2 + a1 - 0.5 */ 217 #define MODE_MODULATE_SUBTRACT_ATI 12 /* r = a0 * a2 - a1 */ 218 #define MODE_ADD_PRODUCTS 13 /* r = a0 * a1 + a2 * a3 */ 219 #define MODE_ADD_PRODUCTS_SIGNED 14 /* r = a0 * a1 + a2 * a3 - 0.5 */ 220 #define MODE_UNKNOWN 16 221 222 /** 223 * Translate GL combiner state into a MODE_x value 224 */ 225 static GLuint translate_mode( GLenum envMode, GLenum mode ) 226 { 227 switch (mode) { 228 case GL_REPLACE: return MODE_REPLACE; 229 case GL_MODULATE: return MODE_MODULATE; 230 case GL_ADD: 231 if (envMode == GL_COMBINE4_NV) 232 return MODE_ADD_PRODUCTS; 233 else 234 return MODE_ADD; 235 case GL_ADD_SIGNED: 236 if (envMode == GL_COMBINE4_NV) 237 return MODE_ADD_PRODUCTS_SIGNED; 238 else 239 return MODE_ADD_SIGNED; 240 case GL_INTERPOLATE: return MODE_INTERPOLATE; 241 case GL_SUBTRACT: return MODE_SUBTRACT; 242 case GL_DOT3_RGB: return MODE_DOT3_RGB; 243 case GL_DOT3_RGB_EXT: return MODE_DOT3_RGB_EXT; 244 case GL_DOT3_RGBA: return MODE_DOT3_RGBA; 245 case GL_DOT3_RGBA_EXT: return MODE_DOT3_RGBA_EXT; 246 case GL_MODULATE_ADD_ATI: return MODE_MODULATE_ADD_ATI; 247 case GL_MODULATE_SIGNED_ADD_ATI: return MODE_MODULATE_SIGNED_ADD_ATI; 248 case GL_MODULATE_SUBTRACT_ATI: return MODE_MODULATE_SUBTRACT_ATI; 249 default: 250 assert(0); 251 return MODE_UNKNOWN; 252 } 253 } 254 255 256 /** 257 * Do we need to clamp the results of the given texture env/combine mode? 258 * If the inputs to the mode are in [0,1] we don't always have to clamp 259 * the results. 260 */ 261 static GLboolean 262 need_saturate( GLuint mode ) 263 { 264 switch (mode) { 265 case MODE_REPLACE: 266 case MODE_MODULATE: 267 case MODE_INTERPOLATE: 268 return GL_FALSE; 269 case MODE_ADD: 270 case MODE_ADD_SIGNED: 271 case MODE_SUBTRACT: 272 case MODE_DOT3_RGB: 273 case MODE_DOT3_RGB_EXT: 274 case MODE_DOT3_RGBA: 275 case MODE_DOT3_RGBA_EXT: 276 case MODE_MODULATE_ADD_ATI: 277 case MODE_MODULATE_SIGNED_ADD_ATI: 278 case MODE_MODULATE_SUBTRACT_ATI: 279 case MODE_ADD_PRODUCTS: 280 case MODE_ADD_PRODUCTS_SIGNED: 281 return GL_TRUE; 282 default: 283 assert(0); 284 return GL_FALSE; 285 } 286 } 287 288 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0) 289 290 /** 291 * Identify all possible varying inputs. The fragment program will 292 * never reference non-varying inputs, but will track them via state 293 * constants instead. 294 * 295 * This function figures out all the inputs that the fragment program 296 * has access to. The bitmask is later reduced to just those which 297 * are actually referenced. 298 */ 299 static GLbitfield get_fp_input_mask( struct gl_context *ctx ) 300 { 301 /* _NEW_PROGRAM */ 302 const GLboolean vertexShader = 303 (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX] && 304 ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]->data->LinkStatus && 305 ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]->_LinkedShaders[MESA_SHADER_VERTEX]); 306 const GLboolean vertexProgram = ctx->VertexProgram._Enabled; 307 GLbitfield fp_inputs = 0x0; 308 309 if (ctx->VertexProgram._Overriden) { 310 /* Somebody's messing with the vertex program and we don't have 311 * a clue what's happening. Assume that it could be producing 312 * all possible outputs. 313 */ 314 fp_inputs = ~0; 315 } 316 else if (ctx->RenderMode == GL_FEEDBACK) { 317 /* _NEW_RENDERMODE */ 318 fp_inputs = (VARYING_BIT_COL0 | VARYING_BIT_TEX0); 319 } 320 else if (!(vertexProgram || vertexShader)) { 321 /* Fixed function vertex logic */ 322 /* _NEW_VARYING_VP_INPUTS */ 323 GLbitfield64 varying_inputs = ctx->varying_vp_inputs; 324 325 /* These get generated in the setup routine regardless of the 326 * vertex program: 327 */ 328 /* _NEW_POINT */ 329 if (ctx->Point.PointSprite) 330 varying_inputs |= VARYING_BITS_TEX_ANY; 331 332 /* First look at what values may be computed by the generated 333 * vertex program: 334 */ 335 /* _NEW_LIGHT */ 336 if (ctx->Light.Enabled) { 337 fp_inputs |= VARYING_BIT_COL0; 338 339 if (texenv_doing_secondary_color(ctx)) 340 fp_inputs |= VARYING_BIT_COL1; 341 } 342 343 /* _NEW_TEXTURE */ 344 fp_inputs |= (ctx->Texture._TexGenEnabled | 345 ctx->Texture._TexMatEnabled) << VARYING_SLOT_TEX0; 346 347 /* Then look at what might be varying as a result of enabled 348 * arrays, etc: 349 */ 350 if (varying_inputs & VERT_BIT_COLOR0) 351 fp_inputs |= VARYING_BIT_COL0; 352 if (varying_inputs & VERT_BIT_COLOR1) 353 fp_inputs |= VARYING_BIT_COL1; 354 355 fp_inputs |= (((varying_inputs & VERT_BIT_TEX_ANY) >> VERT_ATTRIB_TEX0) 356 << VARYING_SLOT_TEX0); 357 358 } 359 else { 360 /* calculate from vp->outputs */ 361 struct gl_program *vprog; 362 GLbitfield64 vp_outputs; 363 364 /* Choose GLSL vertex shader over ARB vertex program. Need this 365 * since vertex shader state validation comes after fragment state 366 * validation (see additional comments in state.c). 367 */ 368 if (vertexShader) 369 vprog = ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]->_LinkedShaders[MESA_SHADER_VERTEX]->Program; 370 else 371 vprog = ctx->VertexProgram.Current; 372 373 vp_outputs = vprog->info.outputs_written; 374 375 /* These get generated in the setup routine regardless of the 376 * vertex program: 377 */ 378 /* _NEW_POINT */ 379 if (ctx->Point.PointSprite) 380 vp_outputs |= VARYING_BITS_TEX_ANY; 381 382 if (vp_outputs & (1 << VARYING_SLOT_COL0)) 383 fp_inputs |= VARYING_BIT_COL0; 384 if (vp_outputs & (1 << VARYING_SLOT_COL1)) 385 fp_inputs |= VARYING_BIT_COL1; 386 387 fp_inputs |= (((vp_outputs & VARYING_BITS_TEX_ANY) >> VARYING_SLOT_TEX0) 388 << VARYING_SLOT_TEX0); 389 } 390 391 return fp_inputs; 392 } 393 394 395 /** 396 * Examine current texture environment state and generate a unique 397 * key to identify it. 398 */ 399 static GLuint make_state_key( struct gl_context *ctx, struct state_key *key ) 400 { 401 GLuint j; 402 GLbitfield inputs_referenced = VARYING_BIT_COL0; 403 const GLbitfield inputs_available = get_fp_input_mask( ctx ); 404 GLbitfield mask; 405 GLuint keySize; 406 407 memset(key, 0, sizeof(*key)); 408 409 /* _NEW_TEXTURE */ 410 mask = ctx->Texture._EnabledCoordUnits; 411 while (mask) { 412 const int i = u_bit_scan(&mask); 413 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i]; 414 const struct gl_texture_object *texObj = texUnit->_Current; 415 const struct gl_tex_env_combine_state *comb = texUnit->_CurrentCombine; 416 const struct gl_sampler_object *samp; 417 GLenum format; 418 419 if (!texObj) 420 continue; 421 422 samp = _mesa_get_samplerobj(ctx, i); 423 format = _mesa_texture_base_format(texObj); 424 425 key->unit[i].enabled = 1; 426 key->enabled_units |= (1<<i); 427 key->nr_enabled_units = i + 1; 428 inputs_referenced |= VARYING_BIT_TEX(i); 429 430 key->unit[i].source_index = _mesa_tex_target_to_index(ctx, 431 texObj->Target); 432 433 key->unit[i].shadow = 434 ((samp->CompareMode == GL_COMPARE_R_TO_TEXTURE) && 435 ((format == GL_DEPTH_COMPONENT) || 436 (format == GL_DEPTH_STENCIL_EXT))); 437 438 key->unit[i].NumArgsRGB = comb->_NumArgsRGB; 439 key->unit[i].NumArgsA = comb->_NumArgsA; 440 441 key->unit[i].ModeRGB = 442 translate_mode(texUnit->EnvMode, comb->ModeRGB); 443 key->unit[i].ModeA = 444 translate_mode(texUnit->EnvMode, comb->ModeA); 445 446 key->unit[i].ScaleShiftRGB = comb->ScaleShiftRGB; 447 key->unit[i].ScaleShiftA = comb->ScaleShiftA; 448 449 for (j = 0; j < MAX_COMBINER_TERMS; j++) { 450 key->unit[i].OptRGB[j].Operand = translate_operand(comb->OperandRGB[j]); 451 key->unit[i].OptA[j].Operand = translate_operand(comb->OperandA[j]); 452 key->unit[i].OptRGB[j].Source = translate_source(comb->SourceRGB[j]); 453 key->unit[i].OptA[j].Source = translate_source(comb->SourceA[j]); 454 } 455 } 456 457 /* _NEW_LIGHT | _NEW_FOG */ 458 if (texenv_doing_secondary_color(ctx)) { 459 key->separate_specular = 1; 460 inputs_referenced |= VARYING_BIT_COL1; 461 } 462 463 /* _NEW_FOG */ 464 if (ctx->Fog.Enabled) { 465 key->fog_mode = translate_fog_mode(ctx->Fog.Mode); 466 inputs_referenced |= VARYING_BIT_FOGC; /* maybe */ 467 } 468 469 /* _NEW_BUFFERS */ 470 key->num_draw_buffers = ctx->DrawBuffer->_NumColorDrawBuffers; 471 472 /* _NEW_COLOR */ 473 if (ctx->Color.AlphaEnabled && key->num_draw_buffers == 0) { 474 /* if alpha test is enabled we need to emit at least one color */ 475 key->num_draw_buffers = 1; 476 } 477 478 key->inputs_available = (inputs_available & inputs_referenced); 479 480 /* compute size of state key, ignoring unused texture units */ 481 keySize = sizeof(*key) - sizeof(key->unit) 482 + key->nr_enabled_units * sizeof(key->unit[0]); 483 484 return keySize; 485 } 486 487 488 /** State used to build the fragment program: 489 */ 490 class texenv_fragment_program : public ir_factory { 491 public: 492 struct gl_shader_program *shader_program; 493 struct gl_shader *shader; 494 exec_list *top_instructions; 495 struct state_key *state; 496 497 ir_variable *src_texture[MAX_TEXTURE_COORD_UNITS]; 498 /* Reg containing each texture unit's sampled texture color, 499 * else undef. 500 */ 501 502 /* Texcoord override from bumpmapping. */ 503 ir_variable *texcoord_tex[MAX_TEXTURE_COORD_UNITS]; 504 505 /* Reg containing texcoord for a texture unit, 506 * needed for bump mapping, else undef. 507 */ 508 509 ir_rvalue *src_previous; /**< Reg containing color from previous 510 * stage. May need to be decl'd. 511 */ 512 }; 513 514 static ir_rvalue * 515 get_current_attrib(texenv_fragment_program *p, GLuint attrib) 516 { 517 ir_variable *current; 518 ir_rvalue *val; 519 520 current = p->shader->symbols->get_variable("gl_CurrentAttribFragMESA"); 521 assert(current); 522 current->data.max_array_access = MAX2(current->data.max_array_access, (int)attrib); 523 val = new(p->mem_ctx) ir_dereference_variable(current); 524 ir_rvalue *index = new(p->mem_ctx) ir_constant(attrib); 525 return new(p->mem_ctx) ir_dereference_array(val, index); 526 } 527 528 static ir_rvalue * 529 get_gl_Color(texenv_fragment_program *p) 530 { 531 if (p->state->inputs_available & VARYING_BIT_COL0) { 532 ir_variable *var = p->shader->symbols->get_variable("gl_Color"); 533 assert(var); 534 return new(p->mem_ctx) ir_dereference_variable(var); 535 } else { 536 return get_current_attrib(p, VERT_ATTRIB_COLOR0); 537 } 538 } 539 540 static ir_rvalue * 541 get_source(texenv_fragment_program *p, 542 GLuint src, GLuint unit) 543 { 544 ir_variable *var; 545 ir_dereference *deref; 546 547 switch (src) { 548 case SRC_TEXTURE: 549 return new(p->mem_ctx) ir_dereference_variable(p->src_texture[unit]); 550 551 case SRC_TEXTURE0: 552 case SRC_TEXTURE1: 553 case SRC_TEXTURE2: 554 case SRC_TEXTURE3: 555 case SRC_TEXTURE4: 556 case SRC_TEXTURE5: 557 case SRC_TEXTURE6: 558 case SRC_TEXTURE7: 559 return new(p->mem_ctx) 560 ir_dereference_variable(p->src_texture[src - SRC_TEXTURE0]); 561 562 case SRC_CONSTANT: 563 var = p->shader->symbols->get_variable("gl_TextureEnvColor"); 564 assert(var); 565 deref = new(p->mem_ctx) ir_dereference_variable(var); 566 var->data.max_array_access = MAX2(var->data.max_array_access, (int)unit); 567 return new(p->mem_ctx) ir_dereference_array(deref, 568 new(p->mem_ctx) ir_constant(unit)); 569 570 case SRC_PRIMARY_COLOR: 571 var = p->shader->symbols->get_variable("gl_Color"); 572 assert(var); 573 return new(p->mem_ctx) ir_dereference_variable(var); 574 575 case SRC_ZERO: 576 return new(p->mem_ctx) ir_constant(0.0f); 577 578 case SRC_PREVIOUS: 579 if (!p->src_previous) { 580 return get_gl_Color(p); 581 } else { 582 return p->src_previous->clone(p->mem_ctx, NULL); 583 } 584 585 default: 586 assert(0); 587 return NULL; 588 } 589 } 590 591 static ir_rvalue * 592 emit_combine_source(texenv_fragment_program *p, 593 GLuint unit, 594 GLuint source, 595 GLuint operand) 596 { 597 ir_rvalue *src; 598 599 src = get_source(p, source, unit); 600 601 switch (operand) { 602 case OPR_ONE_MINUS_SRC_COLOR: 603 return sub(new(p->mem_ctx) ir_constant(1.0f), src); 604 605 case OPR_SRC_ALPHA: 606 return src->type->is_scalar() ? src : swizzle_w(src); 607 608 case OPR_ONE_MINUS_SRC_ALPHA: { 609 ir_rvalue *const scalar = src->type->is_scalar() ? src : swizzle_w(src); 610 611 return sub(new(p->mem_ctx) ir_constant(1.0f), scalar); 612 } 613 614 case OPR_ZERO: 615 return new(p->mem_ctx) ir_constant(0.0f); 616 case OPR_ONE: 617 return new(p->mem_ctx) ir_constant(1.0f); 618 case OPR_SRC_COLOR: 619 return src; 620 default: 621 assert(0); 622 return src; 623 } 624 } 625 626 /** 627 * Check if the RGB and Alpha sources and operands match for the given 628 * texture unit's combinder state. When the RGB and A sources and 629 * operands match, we can emit fewer instructions. 630 */ 631 static GLboolean args_match( const struct state_key *key, GLuint unit ) 632 { 633 GLuint i, numArgs = key->unit[unit].NumArgsRGB; 634 635 for (i = 0; i < numArgs; i++) { 636 if (key->unit[unit].OptA[i].Source != key->unit[unit].OptRGB[i].Source) 637 return GL_FALSE; 638 639 switch (key->unit[unit].OptA[i].Operand) { 640 case OPR_SRC_ALPHA: 641 switch (key->unit[unit].OptRGB[i].Operand) { 642 case OPR_SRC_COLOR: 643 case OPR_SRC_ALPHA: 644 break; 645 default: 646 return GL_FALSE; 647 } 648 break; 649 case OPR_ONE_MINUS_SRC_ALPHA: 650 switch (key->unit[unit].OptRGB[i].Operand) { 651 case OPR_ONE_MINUS_SRC_COLOR: 652 case OPR_ONE_MINUS_SRC_ALPHA: 653 break; 654 default: 655 return GL_FALSE; 656 } 657 break; 658 default: 659 return GL_FALSE; /* impossible */ 660 } 661 } 662 663 return GL_TRUE; 664 } 665 666 static ir_rvalue * 667 smear(ir_rvalue *val) 668 { 669 if (!val->type->is_scalar()) 670 return val; 671 672 return swizzle_xxxx(val); 673 } 674 675 static ir_rvalue * 676 emit_combine(texenv_fragment_program *p, 677 GLuint unit, 678 GLuint nr, 679 GLuint mode, 680 const struct mode_opt *opt) 681 { 682 ir_rvalue *src[MAX_COMBINER_TERMS]; 683 ir_rvalue *tmp0, *tmp1; 684 GLuint i; 685 686 assert(nr <= MAX_COMBINER_TERMS); 687 688 for (i = 0; i < nr; i++) 689 src[i] = emit_combine_source( p, unit, opt[i].Source, opt[i].Operand ); 690 691 switch (mode) { 692 case MODE_REPLACE: 693 return src[0]; 694 695 case MODE_MODULATE: 696 return mul(src[0], src[1]); 697 698 case MODE_ADD: 699 return add(src[0], src[1]); 700 701 case MODE_ADD_SIGNED: 702 return add(add(src[0], src[1]), new(p->mem_ctx) ir_constant(-0.5f)); 703 704 case MODE_INTERPOLATE: 705 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) */ 706 tmp0 = mul(src[0], src[2]); 707 tmp1 = mul(src[1], sub(new(p->mem_ctx) ir_constant(1.0f), 708 src[2]->clone(p->mem_ctx, NULL))); 709 return add(tmp0, tmp1); 710 711 case MODE_SUBTRACT: 712 return sub(src[0], src[1]); 713 714 case MODE_DOT3_RGBA: 715 case MODE_DOT3_RGBA_EXT: 716 case MODE_DOT3_RGB_EXT: 717 case MODE_DOT3_RGB: { 718 tmp0 = mul(src[0], new(p->mem_ctx) ir_constant(2.0f)); 719 tmp0 = add(tmp0, new(p->mem_ctx) ir_constant(-1.0f)); 720 721 tmp1 = mul(src[1], new(p->mem_ctx) ir_constant(2.0f)); 722 tmp1 = add(tmp1, new(p->mem_ctx) ir_constant(-1.0f)); 723 724 return dot(swizzle_xyz(smear(tmp0)), swizzle_xyz(smear(tmp1))); 725 } 726 case MODE_MODULATE_ADD_ATI: 727 return add(mul(src[0], src[2]), src[1]); 728 729 case MODE_MODULATE_SIGNED_ADD_ATI: 730 return add(add(mul(src[0], src[2]), src[1]), 731 new(p->mem_ctx) ir_constant(-0.5f)); 732 733 case MODE_MODULATE_SUBTRACT_ATI: 734 return sub(mul(src[0], src[2]), src[1]); 735 736 case MODE_ADD_PRODUCTS: 737 return add(mul(src[0], src[1]), mul(src[2], src[3])); 738 739 case MODE_ADD_PRODUCTS_SIGNED: 740 return add(add(mul(src[0], src[1]), mul(src[2], src[3])), 741 new(p->mem_ctx) ir_constant(-0.5f)); 742 default: 743 assert(0); 744 return src[0]; 745 } 746 } 747 748 /** 749 * Generate instructions for one texture unit's env/combiner mode. 750 */ 751 static ir_rvalue * 752 emit_texenv(texenv_fragment_program *p, GLuint unit) 753 { 754 const struct state_key *key = p->state; 755 GLboolean rgb_saturate, alpha_saturate; 756 GLuint rgb_shift, alpha_shift; 757 758 if (!key->unit[unit].enabled) { 759 return get_source(p, SRC_PREVIOUS, 0); 760 } 761 762 switch (key->unit[unit].ModeRGB) { 763 case MODE_DOT3_RGB_EXT: 764 alpha_shift = key->unit[unit].ScaleShiftA; 765 rgb_shift = 0; 766 break; 767 case MODE_DOT3_RGBA_EXT: 768 alpha_shift = 0; 769 rgb_shift = 0; 770 break; 771 default: 772 rgb_shift = key->unit[unit].ScaleShiftRGB; 773 alpha_shift = key->unit[unit].ScaleShiftA; 774 break; 775 } 776 777 /* If we'll do rgb/alpha shifting don't saturate in emit_combine(). 778 * We don't want to clamp twice. 779 */ 780 if (rgb_shift) 781 rgb_saturate = GL_FALSE; /* saturate after rgb shift */ 782 else if (need_saturate(key->unit[unit].ModeRGB)) 783 rgb_saturate = GL_TRUE; 784 else 785 rgb_saturate = GL_FALSE; 786 787 if (alpha_shift) 788 alpha_saturate = GL_FALSE; /* saturate after alpha shift */ 789 else if (need_saturate(key->unit[unit].ModeA)) 790 alpha_saturate = GL_TRUE; 791 else 792 alpha_saturate = GL_FALSE; 793 794 ir_variable *temp_var = p->make_temp(glsl_type::vec4_type, "texenv_combine"); 795 ir_dereference *deref; 796 ir_rvalue *val; 797 798 /* Emit the RGB and A combine ops 799 */ 800 if (key->unit[unit].ModeRGB == key->unit[unit].ModeA && 801 args_match(key, unit)) { 802 val = emit_combine(p, unit, 803 key->unit[unit].NumArgsRGB, 804 key->unit[unit].ModeRGB, 805 key->unit[unit].OptRGB); 806 val = smear(val); 807 if (rgb_saturate) 808 val = saturate(val); 809 810 p->emit(assign(temp_var, val)); 811 } 812 else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT || 813 key->unit[unit].ModeRGB == MODE_DOT3_RGBA) { 814 ir_rvalue *val = emit_combine(p, unit, 815 key->unit[unit].NumArgsRGB, 816 key->unit[unit].ModeRGB, 817 key->unit[unit].OptRGB); 818 val = smear(val); 819 if (rgb_saturate) 820 val = saturate(val); 821 p->emit(assign(temp_var, val)); 822 } 823 else { 824 /* Need to do something to stop from re-emitting identical 825 * argument calculations here: 826 */ 827 val = emit_combine(p, unit, 828 key->unit[unit].NumArgsRGB, 829 key->unit[unit].ModeRGB, 830 key->unit[unit].OptRGB); 831 val = swizzle_xyz(smear(val)); 832 if (rgb_saturate) 833 val = saturate(val); 834 p->emit(assign(temp_var, val, WRITEMASK_XYZ)); 835 836 val = emit_combine(p, unit, 837 key->unit[unit].NumArgsA, 838 key->unit[unit].ModeA, 839 key->unit[unit].OptA); 840 val = swizzle_w(smear(val)); 841 if (alpha_saturate) 842 val = saturate(val); 843 p->emit(assign(temp_var, val, WRITEMASK_W)); 844 } 845 846 deref = new(p->mem_ctx) ir_dereference_variable(temp_var); 847 848 /* Deal with the final shift: 849 */ 850 if (alpha_shift || rgb_shift) { 851 ir_constant *shift; 852 853 if (rgb_shift == alpha_shift) { 854 shift = new(p->mem_ctx) ir_constant((float)(1 << rgb_shift)); 855 } 856 else { 857 ir_constant_data const_data; 858 859 const_data.f[0] = float(1 << rgb_shift); 860 const_data.f[1] = float(1 << rgb_shift); 861 const_data.f[2] = float(1 << rgb_shift); 862 const_data.f[3] = float(1 << alpha_shift); 863 864 shift = new(p->mem_ctx) ir_constant(glsl_type::vec4_type, 865 &const_data); 866 } 867 868 return saturate(mul(deref, shift)); 869 } 870 else 871 return deref; 872 } 873 874 875 /** 876 * Generate instruction for getting a texture source term. 877 */ 878 static void load_texture( texenv_fragment_program *p, GLuint unit ) 879 { 880 ir_dereference *deref; 881 882 if (p->src_texture[unit]) 883 return; 884 885 const GLuint texTarget = p->state->unit[unit].source_index; 886 ir_rvalue *texcoord; 887 888 if (!(p->state->inputs_available & (VARYING_BIT_TEX0 << unit))) { 889 texcoord = get_current_attrib(p, VERT_ATTRIB_TEX0 + unit); 890 } else if (p->texcoord_tex[unit]) { 891 texcoord = new(p->mem_ctx) ir_dereference_variable(p->texcoord_tex[unit]); 892 } else { 893 ir_variable *tc_array = p->shader->symbols->get_variable("gl_TexCoord"); 894 assert(tc_array); 895 texcoord = new(p->mem_ctx) ir_dereference_variable(tc_array); 896 ir_rvalue *index = new(p->mem_ctx) ir_constant(unit); 897 texcoord = new(p->mem_ctx) ir_dereference_array(texcoord, index); 898 tc_array->data.max_array_access = MAX2(tc_array->data.max_array_access, (int)unit); 899 } 900 901 if (!p->state->unit[unit].enabled) { 902 p->src_texture[unit] = p->make_temp(glsl_type::vec4_type, 903 "dummy_tex"); 904 p->emit(p->src_texture[unit]); 905 906 p->emit(assign(p->src_texture[unit], new(p->mem_ctx) ir_constant(0.0f))); 907 return ; 908 } 909 910 const glsl_type *sampler_type = NULL; 911 int coords = 0; 912 913 switch (texTarget) { 914 case TEXTURE_1D_INDEX: 915 if (p->state->unit[unit].shadow) 916 sampler_type = glsl_type::sampler1DShadow_type; 917 else 918 sampler_type = glsl_type::sampler1D_type; 919 coords = 1; 920 break; 921 case TEXTURE_1D_ARRAY_INDEX: 922 if (p->state->unit[unit].shadow) 923 sampler_type = glsl_type::sampler1DArrayShadow_type; 924 else 925 sampler_type = glsl_type::sampler1DArray_type; 926 coords = 2; 927 break; 928 case TEXTURE_2D_INDEX: 929 if (p->state->unit[unit].shadow) 930 sampler_type = glsl_type::sampler2DShadow_type; 931 else 932 sampler_type = glsl_type::sampler2D_type; 933 coords = 2; 934 break; 935 case TEXTURE_2D_ARRAY_INDEX: 936 if (p->state->unit[unit].shadow) 937 sampler_type = glsl_type::sampler2DArrayShadow_type; 938 else 939 sampler_type = glsl_type::sampler2DArray_type; 940 coords = 3; 941 break; 942 case TEXTURE_RECT_INDEX: 943 if (p->state->unit[unit].shadow) 944 sampler_type = glsl_type::sampler2DRectShadow_type; 945 else 946 sampler_type = glsl_type::sampler2DRect_type; 947 coords = 2; 948 break; 949 case TEXTURE_3D_INDEX: 950 assert(!p->state->unit[unit].shadow); 951 sampler_type = glsl_type::sampler3D_type; 952 coords = 3; 953 break; 954 case TEXTURE_CUBE_INDEX: 955 if (p->state->unit[unit].shadow) 956 sampler_type = glsl_type::samplerCubeShadow_type; 957 else 958 sampler_type = glsl_type::samplerCube_type; 959 coords = 3; 960 break; 961 case TEXTURE_EXTERNAL_INDEX: 962 assert(!p->state->unit[unit].shadow); 963 sampler_type = glsl_type::samplerExternalOES_type; 964 coords = 2; 965 break; 966 } 967 968 p->src_texture[unit] = p->make_temp(glsl_type::vec4_type, 969 "tex"); 970 971 ir_texture *tex = new(p->mem_ctx) ir_texture(ir_tex); 972 973 974 char *sampler_name = ralloc_asprintf(p->mem_ctx, "sampler_%d", unit); 975 ir_variable *sampler = new(p->mem_ctx) ir_variable(sampler_type, 976 sampler_name, 977 ir_var_uniform); 978 p->top_instructions->push_head(sampler); 979 980 /* Set the texture unit for this sampler in the same way that 981 * layout(binding=X) would. 982 */ 983 sampler->data.explicit_binding = true; 984 sampler->data.binding = unit; 985 986 deref = new(p->mem_ctx) ir_dereference_variable(sampler); 987 tex->set_sampler(deref, glsl_type::vec4_type); 988 989 tex->coordinate = new(p->mem_ctx) ir_swizzle(texcoord, 0, 1, 2, 3, coords); 990 991 if (p->state->unit[unit].shadow) { 992 texcoord = texcoord->clone(p->mem_ctx, NULL); 993 tex->shadow_comparator = new(p->mem_ctx) ir_swizzle(texcoord, 994 coords, 0, 0, 0, 995 1); 996 coords++; 997 } 998 999 texcoord = texcoord->clone(p->mem_ctx, NULL); 1000 tex->projector = swizzle_w(texcoord); 1001 1002 p->emit(assign(p->src_texture[unit], tex)); 1003 } 1004 1005 static void 1006 load_texenv_source(texenv_fragment_program *p, 1007 GLuint src, GLuint unit) 1008 { 1009 switch (src) { 1010 case SRC_TEXTURE: 1011 load_texture(p, unit); 1012 break; 1013 1014 case SRC_TEXTURE0: 1015 case SRC_TEXTURE1: 1016 case SRC_TEXTURE2: 1017 case SRC_TEXTURE3: 1018 case SRC_TEXTURE4: 1019 case SRC_TEXTURE5: 1020 case SRC_TEXTURE6: 1021 case SRC_TEXTURE7: 1022 load_texture(p, src - SRC_TEXTURE0); 1023 break; 1024 1025 default: 1026 /* not a texture src - do nothing */ 1027 break; 1028 } 1029 } 1030 1031 1032 /** 1033 * Generate instructions for loading all texture source terms. 1034 */ 1035 static GLboolean 1036 load_texunit_sources( texenv_fragment_program *p, GLuint unit ) 1037 { 1038 const struct state_key *key = p->state; 1039 GLuint i; 1040 1041 for (i = 0; i < key->unit[unit].NumArgsRGB; i++) { 1042 load_texenv_source( p, key->unit[unit].OptRGB[i].Source, unit ); 1043 } 1044 1045 for (i = 0; i < key->unit[unit].NumArgsA; i++) { 1046 load_texenv_source( p, key->unit[unit].OptA[i].Source, unit ); 1047 } 1048 1049 return GL_TRUE; 1050 } 1051 1052 /** 1053 * Applies the fog calculations. 1054 * 1055 * This is basically like the ARB_fragment_prorgam fog options. Note 1056 * that ffvertex_prog.c produces fogcoord for us when 1057 * GL_FOG_COORDINATE_EXT is set to GL_FRAGMENT_DEPTH_EXT. 1058 */ 1059 static ir_rvalue * 1060 emit_fog_instructions(texenv_fragment_program *p, 1061 ir_rvalue *fragcolor) 1062 { 1063 struct state_key *key = p->state; 1064 ir_rvalue *f, *temp; 1065 ir_variable *params, *oparams; 1066 ir_variable *fogcoord; 1067 1068 /* Temporary storage for the whole fog result. Fog calculations 1069 * only affect rgb so we're hanging on to the .a value of fragcolor 1070 * this way. 1071 */ 1072 ir_variable *fog_result = p->make_temp(glsl_type::vec4_type, "fog_result"); 1073 p->emit(assign(fog_result, fragcolor)); 1074 1075 fragcolor = swizzle_xyz(fog_result); 1076 1077 oparams = p->shader->symbols->get_variable("gl_FogParamsOptimizedMESA"); 1078 assert(oparams); 1079 fogcoord = p->shader->symbols->get_variable("gl_FogFragCoord"); 1080 assert(fogcoord); 1081 params = p->shader->symbols->get_variable("gl_Fog"); 1082 assert(params); 1083 f = new(p->mem_ctx) ir_dereference_variable(fogcoord); 1084 1085 ir_variable *f_var = p->make_temp(glsl_type::float_type, "fog_factor"); 1086 1087 switch (key->fog_mode) { 1088 case FOG_LINEAR: 1089 /* f = (end - z) / (end - start) 1090 * 1091 * gl_MesaFogParamsOptimized gives us (-1 / (end - start)) and 1092 * (end / (end - start)) so we can generate a single MAD. 1093 */ 1094 f = add(mul(f, swizzle_x(oparams)), swizzle_y(oparams)); 1095 break; 1096 case FOG_EXP: 1097 /* f = e^(-(density * fogcoord)) 1098 * 1099 * gl_MesaFogParamsOptimized gives us density/ln(2) so we can 1100 * use EXP2 which is generally the native instruction without 1101 * having to do any further math on the fog density uniform. 1102 */ 1103 f = mul(f, swizzle_z(oparams)); 1104 f = new(p->mem_ctx) ir_expression(ir_unop_neg, f); 1105 f = new(p->mem_ctx) ir_expression(ir_unop_exp2, f); 1106 break; 1107 case FOG_EXP2: 1108 /* f = e^(-(density * fogcoord)^2) 1109 * 1110 * gl_MesaFogParamsOptimized gives us density/sqrt(ln(2)) so we 1111 * can do this like FOG_EXP but with a squaring after the 1112 * multiply by density. 1113 */ 1114 ir_variable *temp_var = p->make_temp(glsl_type::float_type, "fog_temp"); 1115 p->emit(assign(temp_var, mul(f, swizzle_w(oparams)))); 1116 1117 f = mul(temp_var, temp_var); 1118 f = new(p->mem_ctx) ir_expression(ir_unop_neg, f); 1119 f = new(p->mem_ctx) ir_expression(ir_unop_exp2, f); 1120 break; 1121 } 1122 1123 p->emit(assign(f_var, saturate(f))); 1124 1125 f = sub(new(p->mem_ctx) ir_constant(1.0f), f_var); 1126 temp = new(p->mem_ctx) ir_dereference_variable(params); 1127 temp = new(p->mem_ctx) ir_dereference_record(temp, "color"); 1128 temp = mul(swizzle_xyz(temp), f); 1129 1130 p->emit(assign(fog_result, add(temp, mul(fragcolor, f_var)), WRITEMASK_XYZ)); 1131 1132 return new(p->mem_ctx) ir_dereference_variable(fog_result); 1133 } 1134 1135 static void 1136 emit_instructions(texenv_fragment_program *p) 1137 { 1138 struct state_key *key = p->state; 1139 GLuint unit; 1140 1141 if (key->enabled_units) { 1142 /* First pass - to support texture_env_crossbar, first identify 1143 * all referenced texture sources and emit texld instructions 1144 * for each: 1145 */ 1146 for (unit = 0; unit < key->nr_enabled_units; unit++) 1147 if (key->unit[unit].enabled) { 1148 load_texunit_sources(p, unit); 1149 } 1150 1151 /* Second pass - emit combine instructions to build final color: 1152 */ 1153 for (unit = 0; unit < key->nr_enabled_units; unit++) { 1154 if (key->unit[unit].enabled) { 1155 p->src_previous = emit_texenv(p, unit); 1156 } 1157 } 1158 } 1159 1160 ir_rvalue *cf = get_source(p, SRC_PREVIOUS, 0); 1161 1162 if (key->separate_specular) { 1163 ir_variable *spec_result = p->make_temp(glsl_type::vec4_type, 1164 "specular_add"); 1165 p->emit(assign(spec_result, cf)); 1166 1167 ir_rvalue *secondary; 1168 if (p->state->inputs_available & VARYING_BIT_COL1) { 1169 ir_variable *var = 1170 p->shader->symbols->get_variable("gl_SecondaryColor"); 1171 assert(var); 1172 secondary = swizzle_xyz(var); 1173 } else { 1174 secondary = swizzle_xyz(get_current_attrib(p, VERT_ATTRIB_COLOR1)); 1175 } 1176 1177 p->emit(assign(spec_result, add(swizzle_xyz(spec_result), secondary), 1178 WRITEMASK_XYZ)); 1179 1180 cf = new(p->mem_ctx) ir_dereference_variable(spec_result); 1181 } 1182 1183 if (key->fog_mode) { 1184 cf = emit_fog_instructions(p, cf); 1185 } 1186 1187 ir_variable *frag_color = p->shader->symbols->get_variable("gl_FragColor"); 1188 assert(frag_color); 1189 p->emit(assign(frag_color, cf)); 1190 } 1191 1192 /** 1193 * Generate a new fragment program which implements the context's 1194 * current texture env/combine mode. 1195 */ 1196 static struct gl_shader_program * 1197 create_new_program(struct gl_context *ctx, struct state_key *key) 1198 { 1199 texenv_fragment_program p; 1200 unsigned int unit; 1201 _mesa_glsl_parse_state *state; 1202 1203 p.mem_ctx = ralloc_context(NULL); 1204 p.shader = _mesa_new_shader(0, MESA_SHADER_FRAGMENT); 1205 #ifdef DEBUG 1206 p.shader->SourceChecksum = 0xf18ed; /* fixed */ 1207 #endif 1208 p.shader->ir = new(p.shader) exec_list; 1209 state = new(p.shader) _mesa_glsl_parse_state(ctx, MESA_SHADER_FRAGMENT, 1210 p.shader); 1211 p.shader->symbols = state->symbols; 1212 p.top_instructions = p.shader->ir; 1213 p.instructions = p.shader->ir; 1214 p.state = key; 1215 p.shader_program = _mesa_new_shader_program(0); 1216 1217 /* Tell the linker to ignore the fact that we're building a 1218 * separate shader, in case we're in a GLES2 context that would 1219 * normally reject that. The real problem is that we're building a 1220 * fixed function program in a GLES2 context at all, but that's a 1221 * big mess to clean up. 1222 */ 1223 p.shader_program->SeparateShader = GL_TRUE; 1224 1225 /* The legacy GLSL shadow functions follow the depth texture 1226 * mode and return vec4. The GLSL 1.30 shadow functions return float and 1227 * ignore the depth texture mode. That's a shader and state dependency 1228 * that's difficult to deal with. st/mesa uses a simple but not 1229 * completely correct solution: if the shader declares GLSL >= 1.30 and 1230 * the depth texture mode is GL_ALPHA (000X), it sets the XXXX swizzle 1231 * instead. Thus, the GLSL 1.30 shadow function will get the result in .x 1232 * and legacy shadow functions will get it in .w as expected. 1233 * For the fixed-function fragment shader, use 120 to get correct behavior 1234 * for GL_ALPHA. 1235 */ 1236 state->language_version = 120; 1237 1238 state->es_shader = false; 1239 if (_mesa_is_gles(ctx) && ctx->Extensions.OES_EGL_image_external) 1240 state->OES_EGL_image_external_enable = true; 1241 _mesa_glsl_initialize_types(state); 1242 _mesa_glsl_initialize_variables(p.instructions, state); 1243 1244 for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) { 1245 p.src_texture[unit] = NULL; 1246 p.texcoord_tex[unit] = NULL; 1247 } 1248 1249 p.src_previous = NULL; 1250 1251 ir_function *main_f = new(p.mem_ctx) ir_function("main"); 1252 p.emit(main_f); 1253 state->symbols->add_function(main_f); 1254 1255 ir_function_signature *main_sig = 1256 new(p.mem_ctx) ir_function_signature(glsl_type::void_type); 1257 main_sig->is_defined = true; 1258 main_f->add_signature(main_sig); 1259 1260 p.instructions = &main_sig->body; 1261 if (key->num_draw_buffers) 1262 emit_instructions(&p); 1263 1264 validate_ir_tree(p.shader->ir); 1265 1266 const struct gl_shader_compiler_options *options = 1267 &ctx->Const.ShaderCompilerOptions[MESA_SHADER_FRAGMENT]; 1268 1269 /* Conservative approach: Don't optimize here, the linker does it too. */ 1270 if (!ctx->Const.GLSLOptimizeConservatively) { 1271 while (do_common_optimization(p.shader->ir, false, false, options, 1272 ctx->Const.NativeIntegers)) 1273 ; 1274 } 1275 1276 reparent_ir(p.shader->ir, p.shader->ir); 1277 1278 p.shader->CompileStatus = true; 1279 p.shader->Version = state->language_version; 1280 p.shader_program->Shaders = 1281 (gl_shader **)malloc(sizeof(*p.shader_program->Shaders)); 1282 p.shader_program->Shaders[0] = p.shader; 1283 p.shader_program->NumShaders = 1; 1284 1285 _mesa_glsl_link_shader(ctx, p.shader_program); 1286 1287 if (!p.shader_program->data->LinkStatus) 1288 _mesa_problem(ctx, "Failed to link fixed function fragment shader: %s\n", 1289 p.shader_program->data->InfoLog); 1290 1291 ralloc_free(p.mem_ctx); 1292 return p.shader_program; 1293 } 1294 1295 extern "C" { 1296 1297 /** 1298 * Return a fragment program which implements the current 1299 * fixed-function texture, fog and color-sum operations. 1300 */ 1301 struct gl_shader_program * 1302 _mesa_get_fixed_func_fragment_program(struct gl_context *ctx) 1303 { 1304 struct gl_shader_program *shader_program; 1305 struct state_key key; 1306 GLuint keySize; 1307 1308 keySize = make_state_key(ctx, &key); 1309 1310 shader_program = (struct gl_shader_program *) 1311 _mesa_search_program_cache(ctx->FragmentProgram.Cache, 1312 &key, keySize); 1313 1314 if (!shader_program) { 1315 shader_program = create_new_program(ctx, &key); 1316 1317 _mesa_shader_cache_insert(ctx, ctx->FragmentProgram.Cache, 1318 &key, keySize, shader_program); 1319 } 1320 1321 return shader_program; 1322 } 1323 1324 } 1325
