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