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