src/mesa/tnl/t_draw.c

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  7.1
   4  *
   5  * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  11  * and/or sell copies of the Software, and to permit persons to whom the
  12  * Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included
  15  * in all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  21  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  22  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors:
  25  *    Keith Whitwell <keith@tungstengraphics.com>
  26  */
  27
  28 #include "main/glheader.h"
  29 #include "main/condrender.h"
  30 #include "main/context.h"
  31 #include "main/imports.h"
  32 #include "main/mtypes.h"
  33 #include "main/macros.h"
  34 #include "main/enums.h"
  35
  36 #include "t_context.h"
  37 #include "tnl.h"
  38
  39
  40
  41 static GLubyte *get_space(struct gl_context *ctx, GLuint bytes)
  42 {
  43    TNLcontext *tnl = TNL_CONTEXT(ctx);
  44    GLubyte *space = malloc(bytes);
  45
  46    tnl->block[tnl->nr_blocks++] = space;
  47    return space;
  48 }
  49
  50
  51 static void free_space(struct gl_context *ctx)
  52 {
  53    TNLcontext *tnl = TNL_CONTEXT(ctx);
  54    GLuint i;
  55    for (i = 0; i < tnl->nr_blocks; i++)
  56       free(tnl->block[i]);
  57    tnl->nr_blocks = 0;
  58 }
  59
  60
  61 /* Convert the incoming array to GLfloats.  Understands the
  62  * array->Normalized flag and selects the correct conversion method.
  63  */
  64 #define CONVERT( TYPE, MACRO ) do {             \
  65    GLuint i, j;                                 \
  66    if (input->Normalized) {                     \
  67       for (i = 0; i < count; i++) {             \
  68          const TYPE *in = (TYPE *)ptr;          \
  69          for (j = 0; j < sz; j++) {             \
  70             *fptr++ = MACRO(*in);               \
  71             in++;                               \
  72          }                                      \
  73          ptr += input->StrideB;                 \
  74       }                                         \
  75    } else {                                     \
  76       for (i = 0; i < count; i++) {             \
  77          const TYPE *in = (TYPE *)ptr;          \
  78          for (j = 0; j < sz; j++) {             \
  79             *fptr++ = (GLfloat)(*in);           \
  80             in++;                               \
  81          }                                      \
  82          ptr += input->StrideB;                 \
  83       }                                         \
  84    }                                            \
  85 } while (0)
  86
  87
  88 /**
  89  * Convert array of BGRA/GLubyte[4] values to RGBA/float[4]
  90  * \param ptr  input/ubyte array
  91  * \param fptr  output/float array
  92  */
  93 static void
  94 convert_bgra_to_float(const struct gl_client_array *input,
  95                       const GLubyte *ptr, GLfloat *fptr,
  96                       GLuint count )
  97 {
  98    GLuint i;
  99    assert(input->Normalized);
 100    assert(input->Size == 4);
 101    for (i = 0; i < count; i++) {
 102       const GLubyte *in = (GLubyte *) ptr;  /* in is in BGRA order */
 103       *fptr++ = UBYTE_TO_FLOAT(in[2]);  /* red */
 104       *fptr++ = UBYTE_TO_FLOAT(in[1]);  /* green */
 105       *fptr++ = UBYTE_TO_FLOAT(in[0]);  /* blue */
 106       *fptr++ = UBYTE_TO_FLOAT(in[3]);  /* alpha */
 107       ptr += input->StrideB;
 108    }
 109 }
 110
 111 static void
 112 convert_half_to_float(const struct gl_client_array *input,
 113                       const GLubyte *ptr, GLfloat *fptr,
 114                       GLuint count, GLuint sz)
 115 {
 116    GLuint i, j;
 117
 118    for (i = 0; i < count; i++) {
 119       GLhalfARB *in = (GLhalfARB *)ptr;
 120
 121       for (j = 0; j < sz; j++) {
 122          *fptr++ = _mesa_half_to_float(in[j]);
 123       }
 124       ptr += input->StrideB;
 125    }
 126 }
 127
 128 /**
 129  * \brief Convert fixed-point to floating-point.
 130  *
 131  * In OpenGL, a fixed-point number is a "signed 2's complement 16.16 scaled
 132  * integer" (Table 2.2 of the OpenGL ES 2.0 spec).
 133  *
 134  * If the buffer has the \c normalized flag set, the formula
 135  *     \code normalize(x) := (2*x + 1) / (2^16 - 1) \endcode
 136  * is used to map the fixed-point numbers into the range [-1, 1].
 137  */
 138 static void
 139 convert_fixed_to_float(const struct gl_client_array *input,
 140                        const GLubyte *ptr, GLfloat *fptr,
 141                        GLuint count)
 142 {
 143    GLuint i, j;
 144    const GLint size = input->Size;
 145
 146    if (input->Normalized) {
 147       for (i = 0; i < count; ++i) {
 148          const GLfixed *in = (GLfixed *) ptr;
 149          for (j = 0; j < size; ++j) {
 150             *fptr++ = (GLfloat) (2 * in[j] + 1) / (GLfloat) ((1 << 16) - 1);
 151          }
 152          ptr += input->StrideB;
 153       }
 154    } else {
 155       for (i = 0; i < count; ++i) {
 156          const GLfixed *in = (GLfixed *) ptr;
 157          for (j = 0; j < size; ++j) {
 158             *fptr++ = in[j] / (GLfloat) (1 << 16);
 159          }
 160          ptr += input->StrideB;
 161       }
 162    }
 163 }
 164
 165 /* Adjust pointer to point at first requested element, convert to
 166  * floating point, populate VB->AttribPtr[].
 167  */
 168 static void _tnl_import_array( struct gl_context *ctx,
 169                                GLuint attrib,
 170                                GLuint count,
 171                                const struct gl_client_array *input,
 172                                const GLubyte *ptr )
 173 {
 174    TNLcontext *tnl = TNL_CONTEXT(ctx);
 175    struct vertex_buffer *VB = &tnl->vb;
 176    GLuint stride = input->StrideB;
 177
 178    if (input->Type != GL_FLOAT) {
 179       const GLuint sz = input->Size;
 180       GLubyte *buf = get_space(ctx, count * sz * sizeof(GLfloat));
 181       GLfloat *fptr = (GLfloat *)buf;
 182
 183       switch (input->Type) {
 184       case GL_BYTE:
 185          CONVERT(GLbyte, BYTE_TO_FLOAT);
 186          break;
 187       case GL_UNSIGNED_BYTE:
 188          if (input->Format == GL_BGRA) {
 189             /* See GL_EXT_vertex_array_bgra */
 190             convert_bgra_to_float(input, ptr, fptr, count);
 191          }
 192          else {
 193             CONVERT(GLubyte, UBYTE_TO_FLOAT);
 194          }
 195          break;
 196       case GL_SHORT:
 197          CONVERT(GLshort, SHORT_TO_FLOAT);
 198          break;
 199       case GL_UNSIGNED_SHORT:
 200          CONVERT(GLushort, USHORT_TO_FLOAT);
 201          break;
 202       case GL_INT:
 203          CONVERT(GLint, INT_TO_FLOAT);
 204          break;
 205       case GL_UNSIGNED_INT:
 206          CONVERT(GLuint, UINT_TO_FLOAT);
 207          break;
 208       case GL_DOUBLE:
 209          CONVERT(GLdouble, (GLfloat));
 210          break;
 211       case GL_HALF_FLOAT:
 212          convert_half_to_float(input, ptr, fptr, count, sz);
 213          break;
 214       case GL_FIXED:
 215          convert_fixed_to_float(input, ptr, fptr, count);
 216          break;
 217       default:
 218          assert(0);
 219          break;
 220       }
 221
 222       ptr = buf;
 223       stride = sz * sizeof(GLfloat);
 224    }
 225
 226    VB->AttribPtr[attrib] = &tnl->tmp_inputs[attrib];
 227    VB->AttribPtr[attrib]->data = (GLfloat (*)[4])ptr;
 228    VB->AttribPtr[attrib]->start = (GLfloat *)ptr;
 229    VB->AttribPtr[attrib]->count = count;
 230    VB->AttribPtr[attrib]->stride = stride;
 231    VB->AttribPtr[attrib]->size = input->Size;
 232
 233    /* This should die, but so should the whole GLvector4f concept:
 234     */
 235    VB->AttribPtr[attrib]->flags = (((1<<input->Size)-1) |
 236                                    VEC_NOT_WRITEABLE |
 237                                    (stride == 4*sizeof(GLfloat) ? 0 : VEC_BAD_STRIDE));
 238
 239    VB->AttribPtr[attrib]->storage = NULL;
 240 }
 241
 242 #define CLIPVERTS  ((6 + MAX_CLIP_PLANES) * 2)
 243
 244
 245 static GLboolean *_tnl_import_edgeflag( struct gl_context *ctx,
 246                                         const GLvector4f *input,
 247                                         GLuint count)
 248 {
 249    const GLubyte *ptr = (const GLubyte *)input->data;
 250    const GLuint stride = input->stride;
 251    GLboolean *space = (GLboolean *)get_space(ctx, count + CLIPVERTS);
 252    GLboolean *bptr = space;
 253    GLuint i;
 254
 255    for (i = 0; i < count; i++) {
 256       *bptr++ = ((GLfloat *)ptr)[0] == 1.0;
 257       ptr += stride;
 258    }
 259
 260    return space;
 261 }
 262
 263
 264 static void bind_inputs( struct gl_context *ctx,
 265                          const struct gl_client_array *inputs[],
 266                          GLint count,
 267                          struct gl_buffer_object **bo,
 268                          GLuint *nr_bo )
 269 {
 270    TNLcontext *tnl = TNL_CONTEXT(ctx);
 271    struct vertex_buffer *VB = &tnl->vb;
 272    GLuint i;
 273
 274    /* Map all the VBOs
 275     */
 276    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 277       const void *ptr;
 278
 279       if (inputs[i]->BufferObj->Name) {
 280          if (!inputs[i]->BufferObj->Pointer) {
 281             bo[*nr_bo] = inputs[i]->BufferObj;
 282             (*nr_bo)++;
 283             ctx->Driver.MapBufferRange(ctx, 0, inputs[i]->BufferObj->Size,
 284                                        GL_MAP_READ_BIT,
 285                                        inputs[i]->BufferObj);
 286
 287             assert(inputs[i]->BufferObj->Pointer);
 288          }
 289
 290          ptr = ADD_POINTERS(inputs[i]->BufferObj->Pointer,
 291                             inputs[i]->Ptr);
 292       }
 293       else
 294          ptr = inputs[i]->Ptr;
 295
 296       /* Just make sure the array is floating point, otherwise convert to
 297        * temporary storage.
 298        *
 299        * XXX: remove the GLvector4f type at some stage and just use
 300        * client arrays.
 301        */
 302       _tnl_import_array(ctx, i, count, inputs[i], ptr);
 303    }
 304
 305    /* We process only the vertices between min & max index:
 306     */
 307    VB->Count = count;
 308
 309    /* These should perhaps be part of _TNL_ATTRIB_* */
 310    VB->BackfaceColorPtr = NULL;
 311    VB->BackfaceIndexPtr = NULL;
 312    VB->BackfaceSecondaryColorPtr = NULL;
 313
 314    /* Clipping and drawing code still requires this to be a packed
 315     * array of ubytes which can be written into.  TODO: Fix and
 316     * remove.
 317     */
 318    if (ctx->Polygon.FrontMode != GL_FILL ||
 319        ctx->Polygon.BackMode != GL_FILL)
 320    {
 321       VB->EdgeFlag = _tnl_import_edgeflag( ctx,
 322                                            VB->AttribPtr[_TNL_ATTRIB_EDGEFLAG],
 323                                            VB->Count );
 324    }
 325    else {
 326       /* the data previously pointed to by EdgeFlag may have been freed */
 327       VB->EdgeFlag = NULL;
 328    }
 329 }
 330
 331
 332 /* Translate indices to GLuints and store in VB->Elts.
 333  */
 334 static void bind_indices( struct gl_context *ctx,
 335                           const struct _mesa_index_buffer *ib,
 336                           struct gl_buffer_object **bo,
 337                           GLuint *nr_bo)
 338 {
 339    TNLcontext *tnl = TNL_CONTEXT(ctx);
 340    struct vertex_buffer *VB = &tnl->vb;
 341    GLuint i;
 342    void *ptr;
 343
 344    if (!ib) {
 345       VB->Elts = NULL;
 346       return;
 347    }
 348
 349    if (ib->obj->Name && !ib->obj->Pointer) {
 350       unsigned map_size;
 351
 352       switch (ib->type) {
 353       case GL_UNSIGNED_BYTE:
 354          map_size = ib->count * sizeof(GLubyte);
 355          break;
 356       case GL_UNSIGNED_SHORT:
 357          map_size = ib->count * sizeof(GLushort);
 358          break;
 359       case GL_UNSIGNED_INT:
 360          map_size = ib->count * sizeof(GLuint);
 361          break;
 362       default:
 363          assert(0);
 364          map_size = 0;
 365       }
 366
 367       bo[*nr_bo] = ib->obj;
 368       (*nr_bo)++;
 369       ptr = ctx->Driver.MapBufferRange(ctx, (GLsizeiptr) ib->ptr, map_size,
 370                                        GL_MAP_READ_BIT, ib->obj);
 371       assert(ib->obj->Pointer);
 372    } else {
 373       ptr = ib->ptr;
 374    }
 375
 376    if (ib->type == GL_UNSIGNED_INT && VB->Primitive[0].basevertex == 0) {
 377       VB->Elts = (GLuint *) ptr;
 378    }
 379    else {
 380       GLuint *elts = (GLuint *)get_space(ctx, ib->count * sizeof(GLuint));
 381       VB->Elts = elts;
 382
 383       if (ib->type == GL_UNSIGNED_INT) {
 384          const GLuint *in = (GLuint *)ptr;
 385          for (i = 0; i < ib->count; i++)
 386             *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
 387       }
 388       else if (ib->type == GL_UNSIGNED_SHORT) {
 389          const GLushort *in = (GLushort *)ptr;
 390          for (i = 0; i < ib->count; i++)
 391             *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
 392       }
 393       else {
 394          const GLubyte *in = (GLubyte *)ptr;
 395          for (i = 0; i < ib->count; i++)
 396             *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
 397       }
 398    }
 399 }
 400
 401 static void bind_prims( struct gl_context *ctx,
 402                         const struct _mesa_prim *prim,
 403                         GLuint nr_prims )
 404 {
 405    TNLcontext *tnl = TNL_CONTEXT(ctx);
 406    struct vertex_buffer *VB = &tnl->vb;
 407
 408    VB->Primitive = prim;
 409    VB->PrimitiveCount = nr_prims;
 410 }
 411
 412 static void unmap_vbos( struct gl_context *ctx,
 413                         struct gl_buffer_object **bo,
 414                         GLuint nr_bo )
 415 {
 416    GLuint i;
 417    for (i = 0; i < nr_bo; i++) {
 418       ctx->Driver.UnmapBuffer(ctx, bo[i]);
 419    }
 420 }
 421
 422
 423 void _tnl_vbo_draw_prims(struct gl_context *ctx,
 424                          const struct gl_client_array *arrays[],
 425                          const struct _mesa_prim *prim,
 426                          GLuint nr_prims,
 427                          const struct _mesa_index_buffer *ib,
 428                          GLboolean index_bounds_valid,
 429                          GLuint min_index,
 430                          GLuint max_index)
 431 {
 432    if (!index_bounds_valid)
 433       vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 434
 435    _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 436 }
 437
 438 /* This is the main entrypoint into the slimmed-down software tnl
 439  * module.  In a regular swtnl driver, this can be plugged straight
 440  * into the vbo->Driver.DrawPrims() callback.
 441  */
 442 void _tnl_draw_prims( struct gl_context *ctx,
 443                       const struct gl_client_array *arrays[],
 444                       const struct _mesa_prim *prim,
 445                       GLuint nr_prims,
 446                       const struct _mesa_index_buffer *ib,
 447                       GLuint min_index,
 448                       GLuint max_index)
 449 {
 450    TNLcontext *tnl = TNL_CONTEXT(ctx);
 451    const GLuint TEST_SPLIT = 0;
 452    const GLint max = TEST_SPLIT ? 8 : tnl->vb.Size - MAX_CLIPPED_VERTICES;
 453    GLint max_basevertex = prim->basevertex;
 454    GLuint i;
 455
 456    /* Mesa core state should have been validated already */
 457    assert(ctx->NewState == 0x0);
 458
 459    if (!_mesa_check_conditional_render(ctx))
 460       return; /* don't draw */
 461
 462    for (i = 1; i < nr_prims; i++)
 463       max_basevertex = MAX2(max_basevertex, prim[i].basevertex);
 464
 465    if (0)
 466    {
 467       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 468       for (i = 0; i < nr_prims; i++)
 469          printf("prim %d: %s start %d count %d\n", i,
 470                 _mesa_lookup_enum_by_nr(prim[i].mode),
 471                 prim[i].start,
 472                 prim[i].count);
 473    }
 474
 475    if (min_index) {
 476       /* We always translate away calls with min_index != 0.
 477        */
 478       vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib,
 479                         min_index, max_index,
 480                         _tnl_vbo_draw_prims );
 481       return;
 482    }
 483    else if ((GLint)max_index + max_basevertex > max) {
 484       /* The software TNL pipeline has a fixed amount of storage for
 485        * vertices and it is necessary to split incoming drawing commands
 486        * if they exceed that limit.
 487        */
 488       struct split_limits limits;
 489       limits.max_verts = max;
 490       limits.max_vb_size = ~0;
 491       limits.max_indices = ~0;
 492
 493       /* This will split the buffers one way or another and
 494        * recursively call back into this function.
 495        */
 496       vbo_split_prims( ctx, arrays, prim, nr_prims, ib,
 497                        0, max_index + prim->basevertex,
 498                        _tnl_vbo_draw_prims,
 499                        &limits );
 500    }
 501    else {
 502       /* May need to map a vertex buffer object for every attribute plus
 503        * one for the index buffer.
 504        */
 505       struct gl_buffer_object *bo[VERT_ATTRIB_MAX + 1];
 506       GLuint nr_bo = 0;
 507       GLuint inst;
 508
 509       for (i = 0; i < nr_prims;) {
 510          GLuint this_nr_prims;
 511
 512          /* Our SW TNL pipeline doesn't handle basevertex yet, so bind_indices
 513           * will rebase the elements to the basevertex, and we'll only
 514           * emit strings of prims with the same basevertex in one draw call.
 515           */
 516          for (this_nr_prims = 1; i + this_nr_prims < nr_prims;
 517               this_nr_prims++) {
 518             if (prim[i].basevertex != prim[i + this_nr_prims].basevertex)
 519                break;
 520          }
 521
 522          assert(prim[i].num_instances > 0);
 523
 524          /* Binding inputs may imply mapping some vertex buffer objects.
 525           * They will need to be unmapped below.
 526           */
 527          for (inst = 0; inst < prim[i].num_instances; inst++) {
 528
 529             bind_prims(ctx, &prim[i], this_nr_prims);
 530             bind_inputs(ctx, arrays, max_index + prim[i].basevertex + 1,
 531                         bo, &nr_bo);
 532             bind_indices(ctx, ib, bo, &nr_bo);
 533
 534             tnl->CurInstance = inst;
 535             TNL_CONTEXT(ctx)->Driver.RunPipeline(ctx);
 536
 537             unmap_vbos(ctx, bo, nr_bo);
 538             free_space(ctx);
 539          }
 540
 541          i += this_nr_prims;
 542       }
 543    }
 544 }
 545