src/mesa/tnl/t_vb_program.c

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  6.5.3
   4  *
   5  * Copyright (C) 1999-2007  Brian Paul   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  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  21  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  22  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25
  26 /**
  27  * \file tnl/t_vb_program.c
  28  * \brief Pipeline stage for executing vertex programs.
  29  * \author Brian Paul,  Keith Whitwell
  30  */
  31
  32
  33 #include "glheader.h"
  34 #include "colormac.h"
  35 #include "context.h"
  36 #include "macros.h"
  37 #include "imports.h"
  38 #include "shader/prog_instruction.h"
  39 #include "shader/prog_statevars.h"
  40 #include "shader/prog_execute.h"
  41 #include "swrast/s_context.h"
  42 #include "swrast/s_texfilter.h"
  43
  44 #include "tnl.h"
  45 #include "t_context.h"
  46 #include "t_pipeline.h"
  47
  48
  49
  50 /*!
  51  * Private storage for the vertex program pipeline stage.
  52  */
  53 struct vp_stage_data {
  54    /** The results of running the vertex program go into these arrays. */
  55    GLvector4f results[VERT_RESULT_MAX];
  56
  57    GLvector4f ndcCoords;              /**< normalized device coords */
  58    GLubyte *clipmask;                 /**< clip flags */
  59    GLubyte ormask, andmask;           /**< for clipping */
  60 };
  61
  62
  63 #define VP_STAGE_DATA(stage) ((struct vp_stage_data *)(stage->privatePtr))
  64
  65
  66 /**
  67  * XXX the texture sampling code in this module is a bit of a hack.
  68  * The texture sampling code is in swrast, though it doesn't have any
  69  * real dependencies on the rest of swrast.  It should probably be
  70  * moved into main/ someday.
  71  */
  72
  73 static void userclip( GLcontext *ctx,
  74                       GLvector4f *clip,
  75                       GLubyte *clipmask,
  76                       GLubyte *clipormask,
  77                       GLubyte *clipandmask )
  78 {
  79    GLuint p;
  80
  81    for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
  82       if (ctx->Transform.ClipPlanesEnabled & (1 << p)) {
  83          GLuint nr, i;
  84          const GLfloat a = ctx->Transform._ClipUserPlane[p][0];
  85          const GLfloat b = ctx->Transform._ClipUserPlane[p][1];
  86          const GLfloat c = ctx->Transform._ClipUserPlane[p][2];
  87          const GLfloat d = ctx->Transform._ClipUserPlane[p][3];
  88          GLfloat *coord = (GLfloat *)clip->data;
  89          GLuint stride = clip->stride;
  90          GLuint count = clip->count;
  91
  92          for (nr = 0, i = 0 ; i < count ; i++) {
  93             GLfloat dp = (coord[0] * a +
  94                           coord[1] * b +
  95                           coord[2] * c +
  96                           coord[3] * d);
  97
  98             if (dp < 0) {
  99                nr++;
 100                clipmask[i] |= CLIP_USER_BIT;
 101             }
 102
 103             STRIDE_F(coord, stride);
 104          }
 105
 106          if (nr > 0) {
 107             *clipormask |= CLIP_USER_BIT;
 108             if (nr == count) {
 109                *clipandmask |= CLIP_USER_BIT;
 110                return;
 111             }
 112          }
 113       }
 114    }
 115 }
 116
 117
 118 static GLboolean
 119 do_ndc_cliptest(GLcontext *ctx, struct vp_stage_data *store)
 120 {
 121    TNLcontext *tnl = TNL_CONTEXT(ctx);
 122    struct vertex_buffer *VB = &tnl->vb;
 123    /* Cliptest and perspective divide.  Clip functions must clear
 124     * the clipmask.
 125     */
 126    store->ormask = 0;
 127    store->andmask = CLIP_FRUSTUM_BITS;
 128
 129    if (tnl->NeedNdcCoords) {
 130       VB->NdcPtr =
 131          _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
 132                                             &store->ndcCoords,
 133                                             store->clipmask,
 134                                             &store->ormask,
 135                                             &store->andmask );
 136    }
 137    else {
 138       VB->NdcPtr = NULL;
 139       _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
 140                                             NULL,
 141                                             store->clipmask,
 142                                             &store->ormask,
 143                                             &store->andmask );
 144    }
 145
 146    if (store->andmask) {
 147       /* All vertices are outside the frustum */
 148       return GL_FALSE;
 149    }
 150
 151    /* Test userclip planes.  This contributes to VB->ClipMask.
 152     */
 153    /** XXX NEW_SLANG _Enabled ??? */
 154    if (ctx->Transform.ClipPlanesEnabled && (!ctx->VertexProgram._Enabled ||
 155       ctx->VertexProgram.Current->IsPositionInvariant)) {
 156       userclip( ctx,
 157                 VB->ClipPtr,
 158                 store->clipmask,
 159                 &store->ormask,
 160                 &store->andmask );
 161
 162       if (store->andmask) {
 163          return GL_FALSE;
 164       }
 165    }
 166
 167    VB->ClipAndMask = store->andmask;
 168    VB->ClipOrMask = store->ormask;
 169    VB->ClipMask = store->clipmask;
 170
 171    return GL_TRUE;
 172 }
 173
 174
 175 static void
 176 vp_fetch_texel(GLcontext *ctx, const GLfloat texcoord[4], GLfloat lambda,
 177                GLuint unit, GLfloat color[4])
 178 {
 179    GLchan rgba[4];
 180    SWcontext *swrast = SWRAST_CONTEXT(ctx);
 181
 182    /* XXX use a float-valued TextureSample routine here!!! */
 183    swrast->TextureSample[unit](ctx, ctx->Texture.Unit[unit]._Current,
 184                                1, (const GLfloat (*)[4]) texcoord,
 185                                &lambda, &rgba);
 186    color[0] = CHAN_TO_FLOAT(rgba[0]);
 187    color[1] = CHAN_TO_FLOAT(rgba[1]);
 188    color[2] = CHAN_TO_FLOAT(rgba[2]);
 189    color[3] = CHAN_TO_FLOAT(rgba[3]);
 190 }
 191
 192
 193 /**
 194  * Called via ctx->Driver.ProgramStringNotify() after a new vertex program
 195  * string has been parsed.
 196  */
 197 void
 198 _tnl_program_string(GLcontext *ctx, GLenum target, struct gl_program *program)
 199 {
 200    /* No-op.
 201     * If we had derived anything from the program that was private to this
 202     * stage we'd recompute/validate it here.
 203     */
 204 }
 205
 206
 207 /**
 208  * Initialize virtual machine state prior to executing vertex program.
 209  */
 210 static void
 211 init_machine(GLcontext *ctx, struct gl_program_machine *machine)
 212 {
 213    /* Input registers get initialized from the current vertex attribs */
 214    MEMCPY(machine->VertAttribs, ctx->Current.Attrib,
 215           MAX_VERTEX_PROGRAM_ATTRIBS * 4 * sizeof(GLfloat));
 216
 217    if (ctx->VertexProgram._Current->IsNVProgram) {
 218       GLuint i;
 219       /* Output/result regs are initialized to [0,0,0,1] */
 220       for (i = 0; i < MAX_NV_VERTEX_PROGRAM_OUTPUTS; i++) {
 221          ASSIGN_4V(machine->Outputs[i], 0.0F, 0.0F, 0.0F, 1.0F);
 222       }
 223       /* Temp regs are initialized to [0,0,0,0] */
 224       for (i = 0; i < MAX_NV_VERTEX_PROGRAM_TEMPS; i++) {
 225          ASSIGN_4V(machine->Temporaries[i], 0.0F, 0.0F, 0.0F, 0.0F);
 226       }
 227       for (i = 0; i < MAX_VERTEX_PROGRAM_ADDRESS_REGS; i++) {
 228          ASSIGN_4V(machine->AddressReg[i], 0, 0, 0, 0);
 229       }
 230    }
 231
 232    machine->NumDeriv = 0;
 233
 234    /* init condition codes */
 235    machine->CondCodes[0] = COND_EQ;
 236    machine->CondCodes[1] = COND_EQ;
 237    machine->CondCodes[2] = COND_EQ;
 238    machine->CondCodes[3] = COND_EQ;
 239
 240    /* init call stack */
 241    machine->StackDepth = 0;
 242
 243    machine->FetchTexelLod = vp_fetch_texel;
 244    machine->FetchTexelDeriv = NULL; /* not used by vertex programs */
 245 }
 246
 247
 248 /**
 249  * Copy the 16 elements of a matrix into four consecutive program
 250  * registers starting at 'pos'.
 251  */
 252 static void
 253 load_matrix(GLfloat registers[][4], GLuint pos, const GLfloat mat[16])
 254 {
 255    GLuint i;
 256    for (i = 0; i < 4; i++) {
 257       registers[pos + i][0] = mat[0 + i];
 258       registers[pos + i][1] = mat[4 + i];
 259       registers[pos + i][2] = mat[8 + i];
 260       registers[pos + i][3] = mat[12 + i];
 261    }
 262 }
 263
 264
 265 /**
 266  * As above, but transpose the matrix.
 267  */
 268 static void
 269 load_transpose_matrix(GLfloat registers[][4], GLuint pos,
 270                       const GLfloat mat[16])
 271 {
 272    MEMCPY(registers[pos], mat, 16 * sizeof(GLfloat));
 273 }
 274
 275
 276 /**
 277  * Load current vertex program's parameter registers with tracked
 278  * matrices (if NV program).  This only needs to be done per
 279  * glBegin/glEnd, not per-vertex.
 280  */
 281 void
 282 _mesa_load_tracked_matrices(GLcontext *ctx)
 283 {
 284    GLuint i;
 285
 286    for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS / 4; i++) {
 287       /* point 'mat' at source matrix */
 288       GLmatrix *mat;
 289       if (ctx->VertexProgram.TrackMatrix[i] == GL_MODELVIEW) {
 290          mat = ctx->ModelviewMatrixStack.Top;
 291       }
 292       else if (ctx->VertexProgram.TrackMatrix[i] == GL_PROJECTION) {
 293          mat = ctx->ProjectionMatrixStack.Top;
 294       }
 295       else if (ctx->VertexProgram.TrackMatrix[i] == GL_TEXTURE) {
 296          mat = ctx->TextureMatrixStack[ctx->Texture.CurrentUnit].Top;
 297       }
 298       else if (ctx->VertexProgram.TrackMatrix[i] == GL_COLOR) {
 299          mat = ctx->ColorMatrixStack.Top;
 300       }
 301       else if (ctx->VertexProgram.TrackMatrix[i]==GL_MODELVIEW_PROJECTION_NV) {
 302          /* XXX verify the combined matrix is up to date */
 303          mat = &ctx->_ModelProjectMatrix;
 304       }
 305       else if (ctx->VertexProgram.TrackMatrix[i] >= GL_MATRIX0_NV &&
 306                ctx->VertexProgram.TrackMatrix[i] <= GL_MATRIX7_NV) {
 307          GLuint n = ctx->VertexProgram.TrackMatrix[i] - GL_MATRIX0_NV;
 308          ASSERT(n < MAX_PROGRAM_MATRICES);
 309          mat = ctx->ProgramMatrixStack[n].Top;
 310       }
 311       else {
 312          /* no matrix is tracked, but we leave the register values as-is */
 313          assert(ctx->VertexProgram.TrackMatrix[i] == GL_NONE);
 314          continue;
 315       }
 316
 317       /* load the matrix values into sequential registers */
 318       if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_IDENTITY_NV) {
 319          load_matrix(ctx->VertexProgram.Parameters, i*4, mat->m);
 320       }
 321       else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_INVERSE_NV) {
 322          _math_matrix_analyse(mat); /* update the inverse */
 323          ASSERT(!_math_matrix_is_dirty(mat));
 324          load_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv);
 325       }
 326       else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_TRANSPOSE_NV) {
 327          load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->m);
 328       }
 329       else {
 330          assert(ctx->VertexProgram.TrackMatrixTransform[i]
 331                 == GL_INVERSE_TRANSPOSE_NV);
 332          _math_matrix_analyse(mat); /* update the inverse */
 333          ASSERT(!_math_matrix_is_dirty(mat));
 334          load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv);
 335       }
 336    }
 337 }
 338
 339
 340 /**
 341  * This function executes vertex programs
 342  */
 343 static GLboolean
 344 run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
 345 {
 346    TNLcontext *tnl = TNL_CONTEXT(ctx);
 347    struct vp_stage_data *store = VP_STAGE_DATA(stage);
 348    struct vertex_buffer *VB = &tnl->vb;
 349    struct gl_vertex_program *program = ctx->VertexProgram._Current;
 350    struct gl_program_machine machine;
 351    GLuint outputs[VERT_RESULT_MAX], numOutputs;
 352    GLuint i, j;
 353
 354    if (!program)
 355       return GL_TRUE;
 356
 357    if (program->IsNVProgram) {
 358       _mesa_load_tracked_matrices(ctx);
 359    }
 360    else {
 361       /* ARB program or vertex shader */
 362       _mesa_load_state_parameters(ctx, program->Base.Parameters);
 363    }
 364
 365    numOutputs = 0;
 366    for (i = 0; i < VERT_RESULT_MAX; i++) {
 367       if (program->Base.OutputsWritten & (1 << i)) {
 368          outputs[numOutputs++] = i;
 369       }
 370    }
 371
 372    for (i = 0; i < VB->Count; i++) {
 373       GLuint attr;
 374
 375       init_machine(ctx, &machine);
 376
 377 #if 0
 378       printf("Input  %d: %f, %f, %f, %f\n", i,
 379              VB->AttribPtr[0]->data[i][0],
 380              VB->AttribPtr[0]->data[i][1],
 381              VB->AttribPtr[0]->data[i][2],
 382              VB->AttribPtr[0]->data[i][3]);
 383       printf("   color: %f, %f, %f, %f\n",
 384              VB->AttribPtr[3]->data[i][0],
 385              VB->AttribPtr[3]->data[i][1],
 386              VB->AttribPtr[3]->data[i][2],
 387              VB->AttribPtr[3]->data[i][3]);
 388       printf("  normal: %f, %f, %f, %f\n",
 389              VB->AttribPtr[2]->data[i][0],
 390              VB->AttribPtr[2]->data[i][1],
 391              VB->AttribPtr[2]->data[i][2],
 392              VB->AttribPtr[2]->data[i][3]);
 393 #endif
 394
 395       /* the vertex array case */
 396       for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
 397          if (program->Base.InputsRead & (1 << attr)) {
 398             const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
 399             const GLuint size = VB->AttribPtr[attr]->size;
 400             const GLuint stride = VB->AttribPtr[attr]->stride;
 401             const GLfloat *data = (GLfloat *) (ptr + stride * i);
 402             COPY_CLEAN_4V(machine.VertAttribs[attr], size, data);
 403          }
 404       }
 405
 406       /* execute the program */
 407       _mesa_execute_program(ctx, &program->Base, &machine);
 408
 409       /* copy the output registers into the VB->attribs arrays */
 410       for (j = 0; j < numOutputs; j++) {
 411          const GLuint attr = outputs[j];
 412          COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
 413       }
 414 #if 0
 415       printf("HPOS: %f %f %f %f\n",
 416              machine.Outputs[0][0],
 417              machine.Outputs[0][1],
 418              machine.Outputs[0][2],
 419              machine.Outputs[0][3]);
 420 #endif
 421    }
 422
 423    /* Fixup fog and point size results if needed */
 424    if (program->IsNVProgram) {
 425       if (ctx->Fog.Enabled &&
 426           (program->Base.OutputsWritten & (1 << VERT_RESULT_FOGC)) == 0) {
 427          for (i = 0; i < VB->Count; i++) {
 428             store->results[VERT_RESULT_FOGC].data[i][0] = 1.0;
 429          }
 430       }
 431
 432       if (ctx->VertexProgram.PointSizeEnabled &&
 433           (program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) == 0) {
 434          for (i = 0; i < VB->Count; i++) {
 435             store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size;
 436          }
 437       }
 438    }
 439
 440    if (program->IsPositionInvariant) {
 441       /* We need the exact same transform as in the fixed function path here
 442          to guarantee invariance, depending on compiler optimization flags results
 443          could be different otherwise */
 444       VB->ClipPtr = TransformRaw( &store->results[0],
 445                                   &ctx->_ModelProjectMatrix,
 446                                   VB->AttribPtr[0] );
 447
 448       /* Drivers expect this to be clean to element 4...
 449        */
 450       switch (VB->ClipPtr->size) {
 451       case 1:
 452          /* impossible */
 453       case 2:
 454          _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
 455          /* fall-through */
 456       case 3:
 457          _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
 458          /* fall-through */
 459       case 4:
 460          break;
 461       }
 462    }
 463
 464
 465    /* Setup the VB pointers so that the next pipeline stages get
 466     * their data from the right place (the program output arrays).
 467     */
 468    else {
 469       VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
 470       VB->ClipPtr->size = 4;
 471       VB->ClipPtr->count = VB->Count;
 472    }
 473    VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0];
 474    VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0];
 475    VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1];
 476    VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1];
 477    VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC];
 478
 479    VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
 480    VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
 481    VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
 482    VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
 483
 484    for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
 485       VB->TexCoordPtr[i] =
 486       VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
 487          = &store->results[VERT_RESULT_TEX0 + i];
 488    }
 489
 490    for (i = 0; i < ctx->Const.MaxVarying; i++) {
 491       if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
 492          /* Note: varying results get put into the generic attributes */
 493          VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
 494             = &store->results[VERT_RESULT_VAR0 + i];
 495       }
 496    }
 497
 498
 499    /* Perform NDC and cliptest operations:
 500     */
 501    return do_ndc_cliptest(ctx, store);
 502 }
 503
 504
 505 /**
 506  * Called the first time stage->run is called.  In effect, don't
 507  * allocate data until the first time the stage is run.
 508  */
 509 static GLboolean
 510 init_vp(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 511 {
 512    TNLcontext *tnl = TNL_CONTEXT(ctx);
 513    struct vertex_buffer *VB = &(tnl->vb);
 514    struct vp_stage_data *store;
 515    const GLuint size = VB->Size;
 516    GLuint i;
 517
 518    stage->privatePtr = MALLOC(sizeof(*store));
 519    store = VP_STAGE_DATA(stage);
 520    if (!store)
 521       return GL_FALSE;
 522
 523    /* Allocate arrays of vertex output values */
 524    for (i = 0; i < VERT_RESULT_MAX; i++) {
 525       _mesa_vector4f_alloc( &store->results[i], 0, size, 32 );
 526       store->results[i].size = 4;
 527    }
 528
 529    /* a few other misc allocations */
 530    _mesa_vector4f_alloc( &store->ndcCoords, 0, size, 32 );
 531    store->clipmask = (GLubyte *) ALIGN_MALLOC(sizeof(GLubyte)*size, 32 );
 532
 533    return GL_TRUE;
 534 }
 535
 536
 537 /**
 538  * Destructor for this pipeline stage.
 539  */
 540 static void
 541 dtr(struct tnl_pipeline_stage *stage)
 542 {
 543    struct vp_stage_data *store = VP_STAGE_DATA(stage);
 544
 545    if (store) {
 546       GLuint i;
 547
 548       /* free the vertex program result arrays */
 549       for (i = 0; i < VERT_RESULT_MAX; i++)
 550          _mesa_vector4f_free( &store->results[i] );
 551
 552       /* free misc arrays */
 553       _mesa_vector4f_free( &store->ndcCoords );
 554       ALIGN_FREE( store->clipmask );
 555
 556       FREE( store );
 557       stage->privatePtr = NULL;
 558    }
 559 }
 560
 561
 562 static void
 563 validate_vp_stage(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 564 {
 565    if (ctx->VertexProgram._Current) {
 566       _swrast_update_texture_samplers(ctx);
 567    }
 568 }
 569
 570
 571
 572 /**
 573  * Public description of this pipeline stage.
 574  */
 575 const struct tnl_pipeline_stage _tnl_vertex_program_stage =
 576 {
 577    "vertex-program",
 578    NULL,                        /* private_data */
 579    init_vp,                     /* create */
 580    dtr,                         /* destroy */
 581    validate_vp_stage,           /* validate */
 582    run_vp                       /* run -- initially set to ctr */
 583 };