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/macros.h"
  37 #include "main/imports.h"
  38 #include "math/m_xform.h"
  39 #include "program/prog_instruction.h"
  40 #include "program/prog_statevars.h"
  41 #include "program/prog_execute.h"
  42 #include "swrast/s_context.h"
  43
  44 #include "tnl/tnl.h"
  45 #include "tnl/t_context.h"
  46 #include "tnl/t_pipeline.h"
  47
  48
  49 #ifdef NAN_CHECK
  50 /** Check for NaNs and very large values */
  51 static inline void
  52 check_float(float x)
  53 {
  54    assert(!IS_INF_OR_NAN(x));
  55    assert(1.0e-15 <= x && x <= 1.0e15);
  56 }
  57 #endif
  58
  59
  60 /*!
  61  * Private storage for the vertex program pipeline stage.
  62  */
  63 struct vp_stage_data {
  64    /** The results of running the vertex program go into these arrays. */
  65    GLvector4f results[VERT_RESULT_MAX];
  66
  67    GLvector4f ndcCoords;              /**< normalized device coords */
  68    GLubyte *clipmask;                 /**< clip flags */
  69    GLubyte ormask, andmask;           /**< for clipping */
  70
  71    GLboolean vertex_textures;
  72
  73    struct gl_program_machine machine;
  74 };
  75
  76
  77 #define VP_STAGE_DATA(stage) ((struct vp_stage_data *)(stage->privatePtr))
  78
  79
  80 static void
  81 userclip( struct gl_context *ctx,
  82           GLvector4f *clip,
  83           GLubyte *clipmask,
  84           GLubyte *clipormask,
  85           GLubyte *clipandmask )
  86 {
  87    GLuint p;
  88
  89    for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
  90       if (ctx->Transform.ClipPlanesEnabled & (1 << p)) {
  91          GLuint nr, i;
  92          const GLfloat a = ctx->Transform._ClipUserPlane[p][0];
  93          const GLfloat b = ctx->Transform._ClipUserPlane[p][1];
  94          const GLfloat c = ctx->Transform._ClipUserPlane[p][2];
  95          const GLfloat d = ctx->Transform._ClipUserPlane[p][3];
  96          GLfloat *coord = (GLfloat *)clip->data;
  97          GLuint stride = clip->stride;
  98          GLuint count = clip->count;
  99
 100          for (nr = 0, i = 0 ; i < count ; i++) {
 101             GLfloat dp = (coord[0] * a +
 102                           coord[1] * b +
 103                           coord[2] * c +
 104                           coord[3] * d);
 105
 106             if (dp < 0) {
 107                nr++;
 108                clipmask[i] |= CLIP_USER_BIT;
 109             }
 110
 111             STRIDE_F(coord, stride);
 112          }
 113
 114          if (nr > 0) {
 115             *clipormask |= CLIP_USER_BIT;
 116             if (nr == count) {
 117                *clipandmask |= CLIP_USER_BIT;
 118                return;
 119             }
 120          }
 121       }
 122    }
 123 }
 124
 125
 126 static GLboolean
 127 do_ndc_cliptest(struct gl_context *ctx, struct vp_stage_data *store)
 128 {
 129    TNLcontext *tnl = TNL_CONTEXT(ctx);
 130    struct vertex_buffer *VB = &tnl->vb;
 131    /* Cliptest and perspective divide.  Clip functions must clear
 132     * the clipmask.
 133     */
 134    store->ormask = 0;
 135    store->andmask = CLIP_FRUSTUM_BITS;
 136
 137    tnl_clip_prepare(ctx);
 138
 139    if (tnl->NeedNdcCoords) {
 140       VB->NdcPtr =
 141          _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
 142                                             &store->ndcCoords,
 143                                             store->clipmask,
 144                                             &store->ormask,
 145                                             &store->andmask,
 146                                             !ctx->Transform.DepthClamp );
 147    }
 148    else {
 149       VB->NdcPtr = NULL;
 150       _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
 151                                             NULL,
 152                                             store->clipmask,
 153                                             &store->ormask,
 154                                             &store->andmask,
 155                                             !ctx->Transform.DepthClamp );
 156    }
 157
 158    if (store->andmask) {
 159       /* All vertices are outside the frustum */
 160       return GL_FALSE;
 161    }
 162
 163    /* Test userclip planes.  This contributes to VB->ClipMask.
 164     */
 165    /** XXX NEW_SLANG _Enabled ??? */
 166    if (ctx->Transform.ClipPlanesEnabled && (!ctx->VertexProgram._Enabled ||
 167       ctx->VertexProgram.Current->IsPositionInvariant)) {
 168       userclip( ctx,
 169                 VB->ClipPtr,
 170                 store->clipmask,
 171                 &store->ormask,
 172                 &store->andmask );
 173
 174       if (store->andmask) {
 175          return GL_FALSE;
 176       }
 177    }
 178
 179    VB->ClipAndMask = store->andmask;
 180    VB->ClipOrMask = store->ormask;
 181    VB->ClipMask = store->clipmask;
 182
 183    return GL_TRUE;
 184 }
 185
 186
 187 /**
 188  * XXX the texture sampling code in this module is a bit of a hack.
 189  * The texture sampling code is in swrast, though it doesn't have any
 190  * real dependencies on the rest of swrast.  It should probably be
 191  * moved into main/ someday.
 192  */
 193 static void
 194 vp_fetch_texel(struct gl_context *ctx, const GLfloat texcoord[4], GLfloat lambda,
 195                GLuint unit, GLfloat color[4])
 196 {
 197    SWcontext *swrast = SWRAST_CONTEXT(ctx);
 198
 199    /* XXX use a float-valued TextureSample routine here!!! */
 200    swrast->TextureSample[unit](ctx, ctx->Texture.Unit[unit]._Current,
 201                                1, (const GLfloat (*)[4]) texcoord,
 202                                &lambda,  (GLfloat (*)[4]) color);
 203 }
 204
 205
 206 /**
 207  * Called via ctx->Driver.ProgramStringNotify() after a new vertex program
 208  * string has been parsed.
 209  */
 210 GLboolean
 211 _tnl_program_string(struct gl_context *ctx, GLenum target, struct gl_program *program)
 212 {
 213    /* No-op.
 214     * If we had derived anything from the program that was private to this
 215     * stage we'd recompute/validate it here.
 216     */
 217    return GL_TRUE;
 218 }
 219
 220
 221 /**
 222  * Initialize virtual machine state prior to executing vertex program.
 223  */
 224 static void
 225 init_machine(struct gl_context *ctx, struct gl_program_machine *machine,
 226              GLuint instID)
 227 {
 228    /* Input registers get initialized from the current vertex attribs */
 229    memcpy(machine->VertAttribs, ctx->Current.Attrib,
 230           MAX_VERTEX_GENERIC_ATTRIBS * 4 * sizeof(GLfloat));
 231
 232    if (ctx->VertexProgram._Current->IsNVProgram) {
 233       GLuint i;
 234       /* Output/result regs are initialized to [0,0,0,1] */
 235       for (i = 0; i < MAX_NV_VERTEX_PROGRAM_OUTPUTS; i++) {
 236          ASSIGN_4V(machine->Outputs[i], 0.0F, 0.0F, 0.0F, 1.0F);
 237       }
 238       /* Temp regs are initialized to [0,0,0,0] */
 239       for (i = 0; i < MAX_NV_VERTEX_PROGRAM_TEMPS; i++) {
 240          ASSIGN_4V(machine->Temporaries[i], 0.0F, 0.0F, 0.0F, 0.0F);
 241       }
 242       for (i = 0; i < MAX_VERTEX_PROGRAM_ADDRESS_REGS; i++) {
 243          ASSIGN_4V(machine->AddressReg[i], 0, 0, 0, 0);
 244       }
 245    }
 246
 247    machine->NumDeriv = 0;
 248
 249    /* init condition codes */
 250    machine->CondCodes[0] = COND_EQ;
 251    machine->CondCodes[1] = COND_EQ;
 252    machine->CondCodes[2] = COND_EQ;
 253    machine->CondCodes[3] = COND_EQ;
 254
 255    /* init call stack */
 256    machine->StackDepth = 0;
 257
 258    machine->FetchTexelLod = vp_fetch_texel;
 259    machine->FetchTexelDeriv = NULL; /* not used by vertex programs */
 260
 261    machine->Samplers = ctx->VertexProgram._Current->Base.SamplerUnits;
 262
 263    machine->SystemValues[SYSTEM_VALUE_INSTANCE_ID][0] = (GLfloat) instID;
 264 }
 265
 266
 267 /**
 268  * Map the texture images which the vertex program will access (if any).
 269  */
 270 static void
 271 map_textures(struct gl_context *ctx, const struct gl_vertex_program *vp)
 272 {
 273    GLuint u;
 274
 275    if (!ctx->Driver.MapTexture)
 276       return;
 277
 278    for (u = 0; u < ctx->Const.MaxVertexTextureImageUnits; u++) {
 279       if (vp->Base.TexturesUsed[u]) {
 280          /* Note: _Current *should* correspond to the target indicated
 281           * in TexturesUsed[u].
 282           */
 283          ctx->Driver.MapTexture(ctx, ctx->Texture.Unit[u]._Current);
 284       }
 285    }
 286 }
 287
 288
 289 /**
 290  * Unmap the texture images which were used by the vertex program (if any).
 291  */
 292 static void
 293 unmap_textures(struct gl_context *ctx, const struct gl_vertex_program *vp)
 294 {
 295    GLuint u;
 296
 297    if (!ctx->Driver.MapTexture)
 298       return;
 299
 300    for (u = 0; u < ctx->Const.MaxVertexTextureImageUnits; u++) {
 301       if (vp->Base.TexturesUsed[u]) {
 302          /* Note: _Current *should* correspond to the target indicated
 303           * in TexturesUsed[u].
 304           */
 305          ctx->Driver.UnmapTexture(ctx, ctx->Texture.Unit[u]._Current);
 306       }
 307    }
 308 }
 309
 310
 311 /**
 312  * This function executes vertex programs
 313  */
 314 static GLboolean
 315 run_vp( struct gl_context *ctx, struct tnl_pipeline_stage *stage )
 316 {
 317    TNLcontext *tnl = TNL_CONTEXT(ctx);
 318    struct vp_stage_data *store = VP_STAGE_DATA(stage);
 319    struct vertex_buffer *VB = &tnl->vb;
 320    struct gl_vertex_program *program = ctx->VertexProgram._Current;
 321    struct gl_program_machine *machine = &store->machine;
 322    GLuint outputs[VERT_RESULT_MAX], numOutputs;
 323    GLuint i, j;
 324
 325    if (!program)
 326       return GL_TRUE;
 327
 328    if (program->IsNVProgram) {
 329       _mesa_load_tracked_matrices(ctx);
 330    }
 331    else {
 332       /* ARB program or vertex shader */
 333       _mesa_load_state_parameters(ctx, program->Base.Parameters);
 334    }
 335
 336    /* make list of outputs to save some time below */
 337    numOutputs = 0;
 338    for (i = 0; i < VERT_RESULT_MAX; i++) {
 339       if (program->Base.OutputsWritten & BITFIELD64_BIT(i)) {
 340          outputs[numOutputs++] = i;
 341       }
 342    }
 343
 344    /* Allocate result vectors.  We delay this until now to avoid allocating
 345     * memory that would never be used if we don't run the software tnl pipeline.
 346     */
 347    if (!store->results[0].storage) {
 348       for (i = 0; i < VERT_RESULT_MAX; i++) {
 349          assert(!store->results[i].storage);
 350          _mesa_vector4f_alloc( &store->results[i], 0, VB->Size, 32 );
 351          store->results[i].size = 4;
 352       }
 353    }
 354
 355    map_textures(ctx, program);
 356
 357    for (i = 0; i < VB->Count; i++) {
 358       GLuint attr;
 359
 360       init_machine(ctx, machine, tnl->CurInstance);
 361
 362 #if 0
 363       printf("Input  %d: %f, %f, %f, %f\n", i,
 364              VB->AttribPtr[0]->data[i][0],
 365              VB->AttribPtr[0]->data[i][1],
 366              VB->AttribPtr[0]->data[i][2],
 367              VB->AttribPtr[0]->data[i][3]);
 368       printf("   color: %f, %f, %f, %f\n",
 369              VB->AttribPtr[3]->data[i][0],
 370              VB->AttribPtr[3]->data[i][1],
 371              VB->AttribPtr[3]->data[i][2],
 372              VB->AttribPtr[3]->data[i][3]);
 373       printf("  normal: %f, %f, %f, %f\n",
 374              VB->AttribPtr[2]->data[i][0],
 375              VB->AttribPtr[2]->data[i][1],
 376              VB->AttribPtr[2]->data[i][2],
 377              VB->AttribPtr[2]->data[i][3]);
 378 #endif
 379
 380       /* the vertex array case */
 381       for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
 382          if (program->Base.InputsRead & (1 << attr)) {
 383             const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
 384             const GLuint size = VB->AttribPtr[attr]->size;
 385             const GLuint stride = VB->AttribPtr[attr]->stride;
 386             const GLfloat *data = (GLfloat *) (ptr + stride * i);
 387 #ifdef NAN_CHECK
 388             check_float(data[0]);
 389             check_float(data[1]);
 390             check_float(data[2]);
 391             check_float(data[3]);
 392 #endif
 393             COPY_CLEAN_4V(machine->VertAttribs[attr], size, data);
 394          }
 395       }
 396
 397       /* execute the program */
 398       _mesa_execute_program(ctx, &program->Base, machine);
 399
 400       /* copy the output registers into the VB->attribs arrays */
 401       for (j = 0; j < numOutputs; j++) {
 402          const GLuint attr = outputs[j];
 403 #ifdef NAN_CHECK
 404          check_float(machine->Outputs[attr][0]);
 405          check_float(machine->Outputs[attr][1]);
 406          check_float(machine->Outputs[attr][2]);
 407          check_float(machine->Outputs[attr][3]);
 408 #endif
 409          COPY_4V(store->results[attr].data[i], machine->Outputs[attr]);
 410       }
 411
 412       /* FOGC is a special case.  Fragment shader expects (f,0,0,1) */
 413       if (program->Base.OutputsWritten & BITFIELD64_BIT(VERT_RESULT_FOGC)) {
 414          store->results[VERT_RESULT_FOGC].data[i][1] = 0.0;
 415          store->results[VERT_RESULT_FOGC].data[i][2] = 0.0;
 416          store->results[VERT_RESULT_FOGC].data[i][3] = 1.0;
 417       }
 418 #ifdef NAN_CHECK
 419       ASSERT(machine->Outputs[0][3] != 0.0F);
 420 #endif
 421 #if 0
 422       printf("HPOS: %f %f %f %f\n",
 423              machine->Outputs[0][0],
 424              machine->Outputs[0][1],
 425              machine->Outputs[0][2],
 426              machine->Outputs[0][3]);
 427 #endif
 428    }
 429
 430    unmap_textures(ctx, program);
 431
 432    /* Fixup fog and point size results if needed */
 433    if (program->IsNVProgram) {
 434       if (ctx->Fog.Enabled &&
 435           (program->Base.OutputsWritten & BITFIELD64_BIT(VERT_RESULT_FOGC)) == 0) {
 436          for (i = 0; i < VB->Count; i++) {
 437             store->results[VERT_RESULT_FOGC].data[i][0] = 1.0;
 438          }
 439       }
 440
 441       if (ctx->VertexProgram.PointSizeEnabled &&
 442           (program->Base.OutputsWritten & BITFIELD64_BIT(VERT_RESULT_PSIZ)) == 0) {
 443          for (i = 0; i < VB->Count; i++) {
 444             store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size;
 445          }
 446       }
 447    }
 448
 449    if (program->IsPositionInvariant) {
 450       /* We need the exact same transform as in the fixed function path here
 451        * to guarantee invariance, depending on compiler optimization flags
 452        * results could be different otherwise.
 453        */
 454       VB->ClipPtr = TransformRaw( &store->results[0],
 455                                   &ctx->_ModelProjectMatrix,
 456                                   VB->AttribPtr[0] );
 457
 458       /* Drivers expect this to be clean to element 4...
 459        */
 460       switch (VB->ClipPtr->size) {
 461       case 1:
 462          /* impossible */
 463       case 2:
 464          _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
 465          /* fall-through */
 466       case 3:
 467          _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
 468          /* fall-through */
 469       case 4:
 470          break;
 471       }
 472    }
 473    else {
 474       /* Setup the VB pointers so that the next pipeline stages get
 475        * their data from the right place (the program output arrays).
 476        */
 477       VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
 478       VB->ClipPtr->size = 4;
 479       VB->ClipPtr->count = VB->Count;
 480    }
 481
 482    VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
 483    VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
 484    VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
 485    VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
 486    VB->BackfaceColorPtr = &store->results[VERT_RESULT_BFC0];
 487    VB->BackfaceSecondaryColorPtr = &store->results[VERT_RESULT_BFC1];
 488
 489    for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
 490       VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
 491          = &store->results[VERT_RESULT_TEX0 + i];
 492    }
 493
 494    for (i = 0; i < ctx->Const.MaxVarying; i++) {
 495       if (program->Base.OutputsWritten & BITFIELD64_BIT(VERT_RESULT_VAR0 + i)) {
 496          /* Note: varying results get put into the generic attributes */
 497          VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
 498             = &store->results[VERT_RESULT_VAR0 + i];
 499       }
 500    }
 501
 502
 503    /* Perform NDC and cliptest operations:
 504     */
 505    return do_ndc_cliptest(ctx, store);
 506 }
 507
 508
 509 /**
 510  * Called the first time stage->run is called.  In effect, don't
 511  * allocate data until the first time the stage is run.
 512  */
 513 static GLboolean
 514 init_vp(struct gl_context *ctx, struct tnl_pipeline_stage *stage)
 515 {
 516    TNLcontext *tnl = TNL_CONTEXT(ctx);
 517    struct vertex_buffer *VB = &(tnl->vb);
 518    struct vp_stage_data *store;
 519    const GLuint size = VB->Size;
 520
 521    stage->privatePtr = CALLOC(sizeof(*store));
 522    store = VP_STAGE_DATA(stage);
 523    if (!store)
 524       return GL_FALSE;
 525
 526    /* a few other misc allocations */
 527    _mesa_vector4f_alloc( &store->ndcCoords, 0, size, 32 );
 528    store->clipmask = (GLubyte *) _mesa_align_malloc(sizeof(GLubyte)*size, 32 );
 529
 530    return GL_TRUE;
 531 }
 532
 533
 534 /**
 535  * Destructor for this pipeline stage.
 536  */
 537 static void
 538 dtr(struct tnl_pipeline_stage *stage)
 539 {
 540    struct vp_stage_data *store = VP_STAGE_DATA(stage);
 541
 542    if (store) {
 543       GLuint i;
 544
 545       /* free the vertex program result arrays */
 546       for (i = 0; i < VERT_RESULT_MAX; i++)
 547          _mesa_vector4f_free( &store->results[i] );
 548
 549       /* free misc arrays */
 550       _mesa_vector4f_free( &store->ndcCoords );
 551       _mesa_align_free( store->clipmask );
 552
 553       FREE( store );
 554       stage->privatePtr = NULL;
 555    }
 556 }
 557
 558
 559 static void
 560 validate_vp_stage(struct gl_context *ctx, struct tnl_pipeline_stage *stage)
 561 {
 562    if (ctx->VertexProgram._Current) {
 563       _swrast_update_texture_samplers(ctx);
 564    }
 565 }
 566
 567
 568
 569 /**
 570  * Public description of this pipeline stage.
 571  */
 572 const struct tnl_pipeline_stage _tnl_vertex_program_stage =
 573 {
 574    "vertex-program",
 575    NULL,                        /* private_data */
 576    init_vp,                     /* create */
 577    dtr,                         /* destroy */
 578    validate_vp_stage,           /* validate */
 579    run_vp                       /* run -- initially set to ctr */
 580 };