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