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