src/mesa/tnl/t_draw.c

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