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