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