src/mesa/tnl/t_vb_program.c

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  7.6
   4  *
   5  * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
   6  * Copyright (C) 2009  VMware, Inc.  All Rights Reserved.
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a
   9  * copy of this software and associated documentation files (the "Software"),
  10  * to deal in the Software without restriction, including without limitation
  11  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  12  * and/or sell copies of the Software, and to permit persons to whom the
  13  * Software is furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice shall be included
  16  * in all copies or substantial portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  21  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, 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 /**
  28  * \file tnl/t_vb_program.c
  29  * \brief Pipeline stage for executing vertex programs.
  30  * \author Brian Paul,  Keith Whitwell
  31  */
  32
  33
  34 #include "main/glheader.h"
  35 #include "main/colormac.h"
  36 #include "main/context.h"
  37 #include "main/macros.h"
  38 #include "main/imports.h"
  39 #include "shader/prog_instruction.h"
  40 #include "shader/prog_statevars.h"
  41 #include "shader/prog_execute.h"
  42 #include "swrast/s_context.h"
  43 #include "swrast/s_texfilter.h"
  44
  45 #include "tnl/tnl.h"
  46 #include "tnl/t_context.h"
  47 #include "tnl/t_pipeline.h"
  48
  49
  50 #ifdef NAN_CHECK
  51 /** Check for NaNs and very large values */
  52 static INLINE void
  53 check_float(float x)
  54 {
  55    assert(!IS_INF_OR_NAN(x));
  56    assert(1.0e-15 <= x && x <= 1.0e15);
  57 }
  58 #endif
  59
  60
  61 /*!
  62  * Private storage for the vertex program pipeline stage.
  63  */
  64 struct vp_stage_data {
  65    /** The results of running the vertex program go into these arrays. */
  66    GLvector4f results[VERT_RESULT_MAX];
  67
  68    GLvector4f ndcCoords;              /**< normalized device coords */
  69    GLubyte *clipmask;                 /**< clip flags */
  70    GLubyte ormask, andmask;           /**< for clipping */
  71 };
  72
  73
  74 #define VP_STAGE_DATA(stage) ((struct vp_stage_data *)(stage->privatePtr))
  75
  76
  77 static void
  78 userclip( GLcontext *ctx,
  79           GLvector4f *clip,
  80           GLubyte *clipmask,
  81           GLubyte *clipormask,
  82           GLubyte *clipandmask )
  83 {
  84    GLuint p;
  85
  86    for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
  87       if (ctx->Transform.ClipPlanesEnabled & (1 << p)) {
  88          GLuint nr, i;
  89          const GLfloat a = ctx->Transform._ClipUserPlane[p][0];
  90          const GLfloat b = ctx->Transform._ClipUserPlane[p][1];
  91          const GLfloat c = ctx->Transform._ClipUserPlane[p][2];
  92          const GLfloat d = ctx->Transform._ClipUserPlane[p][3];
  93          GLfloat *coord = (GLfloat *)clip->data;
  94          GLuint stride = clip->stride;
  95          GLuint count = clip->count;
  96
  97          for (nr = 0, i = 0 ; i < count ; i++) {
  98             GLfloat dp = (coord[0] * a +
  99                           coord[1] * b +
 100                           coord[2] * c +
 101                           coord[3] * d);
 102
 103             if (dp < 0) {
 104                nr++;
 105                clipmask[i] |= CLIP_USER_BIT;
 106             }
 107
 108             STRIDE_F(coord, stride);
 109          }
 110
 111          if (nr > 0) {
 112             *clipormask |= CLIP_USER_BIT;
 113             if (nr == count) {
 114                *clipandmask |= CLIP_USER_BIT;
 115                return;
 116             }
 117          }
 118       }
 119    }
 120 }
 121
 122
 123 static GLboolean
 124 do_ndc_cliptest(GLcontext *ctx, struct vp_stage_data *store)
 125 {
 126    TNLcontext *tnl = TNL_CONTEXT(ctx);
 127    struct vertex_buffer *VB = &tnl->vb;
 128    /* Cliptest and perspective divide.  Clip functions must clear
 129     * the clipmask.
 130     */
 131    store->ormask = 0;
 132    store->andmask = CLIP_FRUSTUM_BITS;
 133
 134    if (tnl->NeedNdcCoords) {
 135       VB->NdcPtr =
 136          _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
 137                                             &store->ndcCoords,
 138                                             store->clipmask,
 139                                             &store->ormask,
 140                                             &store->andmask );
 141    }
 142    else {
 143       VB->NdcPtr = NULL;
 144       _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
 145                                             NULL,
 146                                             store->clipmask,
 147                                             &store->ormask,
 148                                             &store->andmask );
 149    }
 150
 151    if (store->andmask) {
 152       /* All vertices are outside the frustum */
 153       return GL_FALSE;
 154    }
 155
 156    /* Test userclip planes.  This contributes to VB->ClipMask.
 157     */
 158    /** XXX NEW_SLANG _Enabled ??? */
 159    if (ctx->Transform.ClipPlanesEnabled && (!ctx->VertexProgram._Enabled ||
 160       ctx->VertexProgram.Current->IsPositionInvariant)) {
 161       userclip( ctx,
 162                 VB->ClipPtr,
 163                 store->clipmask,
 164                 &store->ormask,
 165                 &store->andmask );
 166
 167       if (store->andmask) {
 168          return GL_FALSE;
 169       }
 170    }
 171
 172    VB->ClipAndMask = store->andmask;
 173    VB->ClipOrMask = store->ormask;
 174    VB->ClipMask = store->clipmask;
 175
 176    return GL_TRUE;
 177 }
 178
 179
 180 /**
 181  * XXX the texture sampling code in this module is a bit of a hack.
 182  * The texture sampling code is in swrast, though it doesn't have any
 183  * real dependencies on the rest of swrast.  It should probably be
 184  * moved into main/ someday.
 185  */
 186 static void
 187 vp_fetch_texel(GLcontext *ctx, const GLfloat texcoord[4], GLfloat lambda,
 188                GLuint unit, GLfloat color[4])
 189 {
 190    SWcontext *swrast = SWRAST_CONTEXT(ctx);
 191
 192    /* XXX use a float-valued TextureSample routine here!!! */
 193    swrast->TextureSample[unit](ctx, ctx->Texture.Unit[unit]._Current,
 194                                1, (const GLfloat (*)[4]) texcoord,
 195                                &lambda,  (GLfloat (*)[4]) color);
 196 }
 197
 198
 199 /**
 200  * Called via ctx->Driver.ProgramStringNotify() after a new vertex program
 201  * string has been parsed.
 202  */
 203 void
 204 _tnl_program_string(GLcontext *ctx, GLenum target, struct gl_program *program)
 205 {
 206    /* No-op.
 207     * If we had derived anything from the program that was private to this
 208     * stage we'd recompute/validate it here.
 209     */
 210 }
 211
 212
 213 /**
 214  * Initialize virtual machine state prior to executing vertex program.
 215  */
 216 static void
 217 init_machine(GLcontext *ctx, struct gl_program_machine *machine)
 218 {
 219    /* Input registers get initialized from the current vertex attribs */
 220    MEMCPY(machine->VertAttribs, ctx->Current.Attrib,
 221           MAX_VERTEX_GENERIC_ATTRIBS * 4 * sizeof(GLfloat));
 222
 223    if (ctx->VertexProgram._Current->IsNVProgram) {
 224       GLuint i;
 225       /* Output/result regs are initialized to [0,0,0,1] */
 226       for (i = 0; i < MAX_NV_VERTEX_PROGRAM_OUTPUTS; i++) {
 227          ASSIGN_4V(machine->Outputs[i], 0.0F, 0.0F, 0.0F, 1.0F);
 228       }
 229       /* Temp regs are initialized to [0,0,0,0] */
 230       for (i = 0; i < MAX_NV_VERTEX_PROGRAM_TEMPS; i++) {
 231          ASSIGN_4V(machine->Temporaries[i], 0.0F, 0.0F, 0.0F, 0.0F);
 232       }
 233       for (i = 0; i < MAX_VERTEX_PROGRAM_ADDRESS_REGS; i++) {
 234          ASSIGN_4V(machine->AddressReg[i], 0, 0, 0, 0);
 235       }
 236    }
 237
 238    machine->NumDeriv = 0;
 239
 240    /* init condition codes */
 241    machine->CondCodes[0] = COND_EQ;
 242    machine->CondCodes[1] = COND_EQ;
 243    machine->CondCodes[2] = COND_EQ;
 244    machine->CondCodes[3] = COND_EQ;
 245
 246    /* init call stack */
 247    machine->StackDepth = 0;
 248
 249    machine->FetchTexelLod = vp_fetch_texel;
 250    machine->FetchTexelDeriv = NULL; /* not used by vertex programs */
 251
 252    machine->Samplers = ctx->VertexProgram._Current->Base.SamplerUnits;
 253 }
 254
 255
 256 /**
 257  * Map the texture images which the vertex program will access (if any).
 258  */
 259 static void
 260 map_textures(GLcontext *ctx, const struct gl_vertex_program *vp)
 261 {
 262    GLuint u;
 263
 264    if (!ctx->Driver.MapTexture)
 265       return;
 266
 267    for (u = 0; u < ctx->Const.MaxVertexTextureImageUnits; u++) {
 268       if (vp->Base.TexturesUsed[u]) {
 269          /* Note: _Current *should* correspond to the target indicated
 270           * in TexturesUsed[u].
 271           */
 272          ctx->Driver.MapTexture(ctx, ctx->Texture.Unit[u]._Current);
 273       }
 274    }
 275 }
 276
 277
 278 /**
 279  * Unmap the texture images which were used by the vertex program (if any).
 280  */
 281 static void
 282 unmap_textures(GLcontext *ctx, const struct gl_vertex_program *vp)
 283 {
 284    GLuint u;
 285
 286    if (!ctx->Driver.MapTexture)
 287       return;
 288
 289    for (u = 0; u < ctx->Const.MaxVertexTextureImageUnits; u++) {
 290       if (vp->Base.TexturesUsed[u]) {
 291          /* Note: _Current *should* correspond to the target indicated
 292           * in TexturesUsed[u].
 293           */
 294          ctx->Driver.UnmapTexture(ctx, ctx->Texture.Unit[u]._Current);
 295       }
 296    }
 297 }
 298
 299
 300 /**
 301  * This function executes vertex programs
 302  */
 303 static GLboolean
 304 run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
 305 {
 306    TNLcontext *tnl = TNL_CONTEXT(ctx);
 307    struct vp_stage_data *store = VP_STAGE_DATA(stage);
 308    struct vertex_buffer *VB = &tnl->vb;
 309    struct gl_vertex_program *program = ctx->VertexProgram._Current;
 310    struct gl_program_machine machine;
 311    GLuint outputs[VERT_RESULT_MAX], numOutputs;
 312    GLuint i, j;
 313
 314    if (!program)
 315       return GL_TRUE;
 316
 317    if (program->IsNVProgram) {
 318       _mesa_load_tracked_matrices(ctx);
 319    }
 320    else {
 321       /* ARB program or vertex shader */
 322       _mesa_load_state_parameters(ctx, program->Base.Parameters);
 323    }
 324
 325    /* make list of outputs to save some time below */
 326    numOutputs = 0;
 327    for (i = 0; i < VERT_RESULT_MAX; i++) {
 328       if (program->Base.OutputsWritten & (1 << i)) {
 329          outputs[numOutputs++] = i;
 330       }
 331    }
 332
 333    map_textures(ctx, program);
 334
 335    for (i = 0; i < VB->Count; i++) {
 336       GLuint attr;
 337
 338       init_machine(ctx, &machine);
 339
 340 #if 0
 341       printf("Input  %d: %f, %f, %f, %f\n", i,
 342              VB->AttribPtr[0]->data[i][0],
 343              VB->AttribPtr[0]->data[i][1],
 344              VB->AttribPtr[0]->data[i][2],
 345              VB->AttribPtr[0]->data[i][3]);
 346       printf("   color: %f, %f, %f, %f\n",
 347              VB->AttribPtr[3]->data[i][0],
 348              VB->AttribPtr[3]->data[i][1],
 349              VB->AttribPtr[3]->data[i][2],
 350              VB->AttribPtr[3]->data[i][3]);
 351       printf("  normal: %f, %f, %f, %f\n",
 352              VB->AttribPtr[2]->data[i][0],
 353              VB->AttribPtr[2]->data[i][1],
 354              VB->AttribPtr[2]->data[i][2],
 355              VB->AttribPtr[2]->data[i][3]);
 356 #endif
 357
 358       /* the vertex array case */
 359       for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
 360          if (program->Base.InputsRead & (1 << attr)) {
 361             const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
 362             const GLuint size = VB->AttribPtr[attr]->size;
 363             const GLuint stride = VB->AttribPtr[attr]->stride;
 364             const GLfloat *data = (GLfloat *) (ptr + stride * i);
 365 #ifdef NAN_CHECK
 366             check_float(data[0]);
 367             check_float(data[1]);
 368             check_float(data[2]);
 369             check_float(data[3]);
 370 #endif
 371             COPY_CLEAN_4V(machine.VertAttribs[attr], size, data);
 372          }
 373       }
 374
 375       /* execute the program */
 376       _mesa_execute_program(ctx, &program->Base, &machine);
 377
 378       /* copy the output registers into the VB->attribs arrays */
 379       for (j = 0; j < numOutputs; j++) {
 380          const GLuint attr = outputs[j];
 381 #ifdef NAN_CHECK
 382          check_float(machine.Outputs[attr][0]);
 383          check_float(machine.Outputs[attr][1]);
 384          check_float(machine.Outputs[attr][2]);
 385          check_float(machine.Outputs[attr][3]);
 386 #endif
 387          COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
 388       }
 389 #if 0
 390       printf("HPOS: %f %f %f %f\n",
 391              machine.Outputs[0][0],
 392              machine.Outputs[0][1],
 393              machine.Outputs[0][2],
 394              machine.Outputs[0][3]);
 395 #endif
 396    }
 397
 398    unmap_textures(ctx, program);
 399
 400    /* Fixup fog and point size results if needed */
 401    if (program->IsNVProgram) {
 402       if (ctx->Fog.Enabled &&
 403           (program->Base.OutputsWritten & (1 << VERT_RESULT_FOGC)) == 0) {
 404          for (i = 0; i < VB->Count; i++) {
 405             store->results[VERT_RESULT_FOGC].data[i][0] = 1.0;
 406          }
 407       }
 408
 409       if (ctx->VertexProgram.PointSizeEnabled &&
 410           (program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) == 0) {
 411          for (i = 0; i < VB->Count; i++) {
 412             store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size;
 413          }
 414       }
 415    }
 416
 417    if (program->IsPositionInvariant) {
 418       /* We need the exact same transform as in the fixed function path here
 419        * to guarantee invariance, depending on compiler optimization flags
 420        * results could be different otherwise.
 421        */
 422       VB->ClipPtr = TransformRaw( &store->results[0],
 423                                   &ctx->_ModelProjectMatrix,
 424                                   VB->AttribPtr[0] );
 425
 426       /* Drivers expect this to be clean to element 4...
 427        */
 428       switch (VB->ClipPtr->size) {
 429       case 1:
 430          /* impossible */
 431       case 2:
 432          _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
 433          /* fall-through */
 434       case 3:
 435          _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
 436          /* fall-through */
 437       case 4:
 438          break;
 439       }
 440    }
 441    else {
 442       /* Setup the VB pointers so that the next pipeline stages get
 443        * their data from the right place (the program output arrays).
 444        */
 445       VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
 446       VB->ClipPtr->size = 4;
 447       VB->ClipPtr->count = VB->Count;
 448    }
 449
 450    VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0];
 451    VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0];
 452    VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1];
 453    VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1];
 454    VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC];
 455
 456    VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
 457    VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
 458    VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
 459    VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
 460
 461    for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
 462       VB->TexCoordPtr[i] =
 463       VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
 464          = &store->results[VERT_RESULT_TEX0 + i];
 465    }
 466
 467    for (i = 0; i < ctx->Const.MaxVarying; i++) {
 468       if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
 469          /* Note: varying results get put into the generic attributes */
 470          VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
 471             = &store->results[VERT_RESULT_VAR0 + i];
 472       }
 473    }
 474
 475
 476    /* Perform NDC and cliptest operations:
 477     */
 478    return do_ndc_cliptest(ctx, store);
 479 }
 480
 481
 482 /**
 483  * Called the first time stage->run is called.  In effect, don't
 484  * allocate data until the first time the stage is run.
 485  */
 486 static GLboolean
 487 init_vp(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 488 {
 489    TNLcontext *tnl = TNL_CONTEXT(ctx);
 490    struct vertex_buffer *VB = &(tnl->vb);
 491    struct vp_stage_data *store;
 492    const GLuint size = VB->Size;
 493    GLuint i;
 494
 495    stage->privatePtr = MALLOC(sizeof(*store));
 496    store = VP_STAGE_DATA(stage);
 497    if (!store)
 498       return GL_FALSE;
 499
 500    /* Allocate arrays of vertex output values */
 501    for (i = 0; i < VERT_RESULT_MAX; i++) {
 502       _mesa_vector4f_alloc( &store->results[i], 0, size, 32 );
 503       store->results[i].size = 4;
 504    }
 505
 506    /* a few other misc allocations */
 507    _mesa_vector4f_alloc( &store->ndcCoords, 0, size, 32 );
 508    store->clipmask = (GLubyte *) ALIGN_MALLOC(sizeof(GLubyte)*size, 32 );
 509
 510    return GL_TRUE;
 511 }
 512
 513
 514 /**
 515  * Destructor for this pipeline stage.
 516  */
 517 static void
 518 dtr(struct tnl_pipeline_stage *stage)
 519 {
 520    struct vp_stage_data *store = VP_STAGE_DATA(stage);
 521
 522    if (store) {
 523       GLuint i;
 524
 525       /* free the vertex program result arrays */
 526       for (i = 0; i < VERT_RESULT_MAX; i++)
 527          _mesa_vector4f_free( &store->results[i] );
 528
 529       /* free misc arrays */
 530       _mesa_vector4f_free( &store->ndcCoords );
 531       ALIGN_FREE( store->clipmask );
 532
 533       FREE( store );
 534       stage->privatePtr = NULL;
 535    }
 536 }
 537
 538
 539 static void
 540 validate_vp_stage(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 541 {
 542    if (ctx->VertexProgram._Current) {
 543       _swrast_update_texture_samplers(ctx);
 544    }
 545 }
 546
 547
 548
 549 /**
 550  * Public description of this pipeline stage.
 551  */
 552 const struct tnl_pipeline_stage _tnl_vertex_program_stage =
 553 {
 554    "vertex-program",
 555    NULL,                        /* private_data */
 556    init_vp,                     /* create */
 557    dtr,                         /* destroy */
 558    validate_vp_stage,           /* validate */
 559    run_vp                       /* run -- initially set to ctr */
 560 };