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 /* Adjust pointer to point at first requested element, convert to
 129  * floating point, populate VB->AttribPtr[].
 130  */
 131 static void _tnl_import_array( struct gl_context *ctx,
 132                                GLuint attrib,
 133                                GLuint count,
 134                                const struct gl_client_array *input,
 135                                const GLubyte *ptr )
 136 {
 137    TNLcontext *tnl = TNL_CONTEXT(ctx);
 138    struct vertex_buffer *VB = &tnl->vb;
 139    GLuint stride = input->StrideB;
 140
 141    if (input->Type != GL_FLOAT) {
 142       const GLuint sz = input->Size;
 143       GLubyte *buf = get_space(ctx, count * sz * sizeof(GLfloat));
 144       GLfloat *fptr = (GLfloat *)buf;
 145
 146       switch (input->Type) {
 147       case GL_BYTE:
 148          CONVERT(GLbyte, BYTE_TO_FLOAT);
 149          break;
 150       case GL_UNSIGNED_BYTE:
 151          if (input->Format == GL_BGRA) {
 152             /* See GL_EXT_vertex_array_bgra */
 153             convert_bgra_to_float(input, ptr, fptr, count);
 154          }
 155          else {
 156             CONVERT(GLubyte, UBYTE_TO_FLOAT);
 157          }
 158          break;
 159       case GL_SHORT:
 160          CONVERT(GLshort, SHORT_TO_FLOAT);
 161          break;
 162       case GL_UNSIGNED_SHORT:
 163          CONVERT(GLushort, USHORT_TO_FLOAT);
 164          break;
 165       case GL_INT:
 166          CONVERT(GLint, INT_TO_FLOAT);
 167          break;
 168       case GL_UNSIGNED_INT:
 169          CONVERT(GLuint, UINT_TO_FLOAT);
 170          break;
 171       case GL_DOUBLE:
 172          CONVERT(GLdouble, (GLfloat));
 173          break;
 174       case GL_HALF_FLOAT:
 175          convert_half_to_float(input, ptr, fptr, count, sz);
 176          break;
 177       default:
 178          assert(0);
 179          break;
 180       }
 181
 182       ptr = buf;
 183       stride = sz * sizeof(GLfloat);
 184    }
 185
 186    VB->AttribPtr[attrib] = &tnl->tmp_inputs[attrib];
 187    VB->AttribPtr[attrib]->data = (GLfloat (*)[4])ptr;
 188    VB->AttribPtr[attrib]->start = (GLfloat *)ptr;
 189    VB->AttribPtr[attrib]->count = count;
 190    VB->AttribPtr[attrib]->stride = stride;
 191    VB->AttribPtr[attrib]->size = input->Size;
 192
 193    /* This should die, but so should the whole GLvector4f concept:
 194     */
 195    VB->AttribPtr[attrib]->flags = (((1<<input->Size)-1) |
 196                                    VEC_NOT_WRITEABLE |
 197                                    (stride == 4*sizeof(GLfloat) ? 0 : VEC_BAD_STRIDE));
 198
 199    VB->AttribPtr[attrib]->storage = NULL;
 200 }
 201
 202 #define CLIPVERTS  ((6 + MAX_CLIP_PLANES) * 2)
 203
 204
 205 static GLboolean *_tnl_import_edgeflag( struct gl_context *ctx,
 206                                         const GLvector4f *input,
 207                                         GLuint count)
 208 {
 209    const GLubyte *ptr = (const GLubyte *)input->data;
 210    const GLuint stride = input->stride;
 211    GLboolean *space = (GLboolean *)get_space(ctx, count + CLIPVERTS);
 212    GLboolean *bptr = space;
 213    GLuint i;
 214
 215    for (i = 0; i < count; i++) {
 216       *bptr++ = ((GLfloat *)ptr)[0] == 1.0;
 217       ptr += stride;
 218    }
 219
 220    return space;
 221 }
 222
 223
 224 static void bind_inputs( struct gl_context *ctx,
 225                          const struct gl_client_array *inputs[],
 226                          GLint count,
 227                          struct gl_buffer_object **bo,
 228                          GLuint *nr_bo )
 229 {
 230    TNLcontext *tnl = TNL_CONTEXT(ctx);
 231    struct vertex_buffer *VB = &tnl->vb;
 232    GLuint i;
 233
 234    /* Map all the VBOs
 235     */
 236    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 237       const void *ptr;
 238
 239       if (inputs[i]->BufferObj->Name) {
 240          if (!inputs[i]->BufferObj->Pointer) {
 241             bo[*nr_bo] = inputs[i]->BufferObj;
 242             (*nr_bo)++;
 243             ctx->Driver.MapBuffer(ctx,
 244                                   GL_ARRAY_BUFFER,
 245                                   GL_READ_ONLY_ARB,
 246                                   inputs[i]->BufferObj);
 247
 248             assert(inputs[i]->BufferObj->Pointer);
 249          }
 250
 251          ptr = ADD_POINTERS(inputs[i]->BufferObj->Pointer,
 252                             inputs[i]->Ptr);
 253       }
 254       else
 255          ptr = inputs[i]->Ptr;
 256
 257       /* Just make sure the array is floating point, otherwise convert to
 258        * temporary storage.
 259        *
 260        * XXX: remove the GLvector4f type at some stage and just use
 261        * client arrays.
 262        */
 263       _tnl_import_array(ctx, i, count, inputs[i], ptr);
 264    }
 265
 266    /* We process only the vertices between min & max index:
 267     */
 268    VB->Count = count;
 269
 270    /* These should perhaps be part of _TNL_ATTRIB_* */
 271    VB->BackfaceColorPtr = NULL;
 272    VB->BackfaceIndexPtr = NULL;
 273    VB->BackfaceSecondaryColorPtr = NULL;
 274
 275    /* Clipping and drawing code still requires this to be a packed
 276     * array of ubytes which can be written into.  TODO: Fix and
 277     * remove.
 278     */
 279    if (ctx->Polygon.FrontMode != GL_FILL ||
 280        ctx->Polygon.BackMode != GL_FILL)
 281    {
 282       VB->EdgeFlag = _tnl_import_edgeflag( ctx,
 283                                            VB->AttribPtr[_TNL_ATTRIB_EDGEFLAG],
 284                                            VB->Count );
 285    }
 286    else {
 287       /* the data previously pointed to by EdgeFlag may have been freed */
 288       VB->EdgeFlag = NULL;
 289    }
 290 }
 291
 292
 293 /* Translate indices to GLuints and store in VB->Elts.
 294  */
 295 static void bind_indices( struct gl_context *ctx,
 296                           const struct _mesa_index_buffer *ib,
 297                           struct gl_buffer_object **bo,
 298                           GLuint *nr_bo)
 299 {
 300    TNLcontext *tnl = TNL_CONTEXT(ctx);
 301    struct vertex_buffer *VB = &tnl->vb;
 302    GLuint i;
 303    void *ptr;
 304
 305    if (!ib) {
 306       VB->Elts = NULL;
 307       return;
 308    }
 309
 310    if (ib->obj->Name && !ib->obj->Pointer) {
 311       bo[*nr_bo] = ib->obj;
 312       (*nr_bo)++;
 313       ctx->Driver.MapBuffer(ctx,
 314                             GL_ELEMENT_ARRAY_BUFFER,
 315                             GL_READ_ONLY_ARB,
 316                             ib->obj);
 317
 318       assert(ib->obj->Pointer);
 319    }
 320
 321    ptr = ADD_POINTERS(ib->obj->Pointer, ib->ptr);
 322
 323    if (ib->type == GL_UNSIGNED_INT && VB->Primitive[0].basevertex == 0) {
 324       VB->Elts = (GLuint *) ptr;
 325    }
 326    else {
 327       GLuint *elts = (GLuint *)get_space(ctx, ib->count * sizeof(GLuint));
 328       VB->Elts = elts;
 329
 330       if (ib->type == GL_UNSIGNED_INT) {
 331          const GLuint *in = (GLuint *)ptr;
 332          for (i = 0; i < ib->count; i++)
 333             *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
 334       }
 335       else if (ib->type == GL_UNSIGNED_SHORT) {
 336          const GLushort *in = (GLushort *)ptr;
 337          for (i = 0; i < ib->count; i++)
 338             *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
 339       }
 340       else {
 341          const GLubyte *in = (GLubyte *)ptr;
 342          for (i = 0; i < ib->count; i++)
 343             *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
 344       }
 345    }
 346 }
 347
 348 static void bind_prims( struct gl_context *ctx,
 349                         const struct _mesa_prim *prim,
 350                         GLuint nr_prims )
 351 {
 352    TNLcontext *tnl = TNL_CONTEXT(ctx);
 353    struct vertex_buffer *VB = &tnl->vb;
 354
 355    VB->Primitive = prim;
 356    VB->PrimitiveCount = nr_prims;
 357 }
 358
 359 static void unmap_vbos( struct gl_context *ctx,
 360                         struct gl_buffer_object **bo,
 361                         GLuint nr_bo )
 362 {
 363    GLuint i;
 364    for (i = 0; i < nr_bo; i++) {
 365       ctx->Driver.UnmapBuffer(ctx,
 366                               0, /* target -- I don't see why this would be needed */
 367                               bo[i]);
 368    }
 369 }
 370
 371
 372 void _tnl_vbo_draw_prims(struct gl_context *ctx,
 373                          const struct gl_client_array *arrays[],
 374                          const struct _mesa_prim *prim,
 375                          GLuint nr_prims,
 376                          const struct _mesa_index_buffer *ib,
 377                          GLboolean index_bounds_valid,
 378                          GLuint min_index,
 379                          GLuint max_index)
 380 {
 381    if (!index_bounds_valid)
 382       vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 383
 384    _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 385 }
 386
 387 /* This is the main entrypoint into the slimmed-down software tnl
 388  * module.  In a regular swtnl driver, this can be plugged straight
 389  * into the vbo->Driver.DrawPrims() callback.
 390  */
 391 void _tnl_draw_prims( struct gl_context *ctx,
 392                       const struct gl_client_array *arrays[],
 393                       const struct _mesa_prim *prim,
 394                       GLuint nr_prims,
 395                       const struct _mesa_index_buffer *ib,
 396                       GLuint min_index,
 397                       GLuint max_index)
 398 {
 399    TNLcontext *tnl = TNL_CONTEXT(ctx);
 400    const GLuint TEST_SPLIT = 0;
 401    const GLint max = TEST_SPLIT ? 8 : tnl->vb.Size - MAX_CLIPPED_VERTICES;
 402    GLint max_basevertex = prim->basevertex;
 403    GLuint i;
 404
 405    /* Mesa core state should have been validated already */
 406    assert(ctx->NewState == 0x0);
 407
 408    if (!_mesa_check_conditional_render(ctx))
 409       return; /* don't draw */
 410
 411    for (i = 1; i < nr_prims; i++)
 412       max_basevertex = MAX2(max_basevertex, prim[i].basevertex);
 413
 414    if (0)
 415    {
 416       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 417       for (i = 0; i < nr_prims; i++)
 418          printf("prim %d: %s start %d count %d\n", i,
 419                 _mesa_lookup_enum_by_nr(prim[i].mode),
 420                 prim[i].start,
 421                 prim[i].count);
 422    }
 423
 424    if (min_index) {
 425       /* We always translate away calls with min_index != 0.
 426        */
 427       vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib,
 428                         min_index, max_index,
 429                         _tnl_vbo_draw_prims );
 430       return;
 431    }
 432    else if ((GLint)max_index + max_basevertex > max) {
 433       /* The software TNL pipeline has a fixed amount of storage for
 434        * vertices and it is necessary to split incoming drawing commands
 435        * if they exceed that limit.
 436        */
 437       struct split_limits limits;
 438       limits.max_verts = max;
 439       limits.max_vb_size = ~0;
 440       limits.max_indices = ~0;
 441
 442       /* This will split the buffers one way or another and
 443        * recursively call back into this function.
 444        */
 445       vbo_split_prims( ctx, arrays, prim, nr_prims, ib,
 446                        0, max_index + prim->basevertex,
 447                        _tnl_vbo_draw_prims,
 448                        &limits );
 449    }
 450    else {
 451       /* May need to map a vertex buffer object for every attribute plus
 452        * one for the index buffer.
 453        */
 454       struct gl_buffer_object *bo[VERT_ATTRIB_MAX + 1];
 455       GLuint nr_bo = 0;
 456
 457       for (i = 0; i < nr_prims;) {
 458          GLuint this_nr_prims;
 459
 460          /* Our SW TNL pipeline doesn't handle basevertex yet, so bind_indices
 461           * will rebase the elements to the basevertex, and we'll only
 462           * emit strings of prims with the same basevertex in one draw call.
 463           */
 464          for (this_nr_prims = 1; i + this_nr_prims < nr_prims;
 465               this_nr_prims++) {
 466             if (prim[i].basevertex != prim[i + this_nr_prims].basevertex)
 467                break;
 468          }
 469
 470          /* Binding inputs may imply mapping some vertex buffer objects.
 471           * They will need to be unmapped below.
 472           */
 473          bind_prims(ctx, &prim[i], this_nr_prims);
 474          bind_inputs(ctx, arrays, max_index + prim[i].basevertex + 1,
 475                      bo, &nr_bo);
 476          bind_indices(ctx, ib, bo, &nr_bo);
 477
 478          TNL_CONTEXT(ctx)->Driver.RunPipeline(ctx);
 479
 480          unmap_vbos(ctx, bo, nr_bo);
 481          free_space(ctx);
 482
 483          i += this_nr_prims;
 484       }
 485    }
 486 }
 487