src/mesa/tnl/t_vertex.c

   1 /*
   2  * Copyright 2003 Tungsten Graphics, inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
  19  * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors:
  25  *    Keith Whitwell <keithw@tungstengraphics.com>
  26  */
  27
  28 #include "main/glheader.h"
  29 #include "main/context.h"
  30 #include "main/colormac.h"
  31
  32 #include "t_context.h"
  33 #include "t_vertex.h"
  34
  35 #define DBG 0
  36
  37 /* Build and manage clipspace/ndc/window vertices.
  38  */
  39
  40 static GLboolean match_fastpath( struct tnl_clipspace *vtx,
  41                                  const struct tnl_clipspace_fastpath *fp)
  42 {
  43    GLuint j;
  44
  45    if (vtx->attr_count != fp->attr_count)
  46       return GL_FALSE;
  47
  48    for (j = 0; j < vtx->attr_count; j++)
  49       if (vtx->attr[j].format != fp->attr[j].format ||
  50           vtx->attr[j].inputsize != fp->attr[j].size ||
  51           vtx->attr[j].vertoffset != fp->attr[j].offset)
  52          return GL_FALSE;
  53
  54    if (fp->match_strides) {
  55       if (vtx->vertex_size != fp->vertex_size)
  56          return GL_FALSE;
  57
  58       for (j = 0; j < vtx->attr_count; j++)
  59          if (vtx->attr[j].inputstride != fp->attr[j].stride)
  60             return GL_FALSE;
  61    }
  62
  63    return GL_TRUE;
  64 }
  65
  66 static GLboolean search_fastpath_emit( struct tnl_clipspace *vtx )
  67 {
  68    struct tnl_clipspace_fastpath *fp = vtx->fastpath;
  69
  70    for ( ; fp ; fp = fp->next) {
  71       if (match_fastpath(vtx, fp)) {
  72          vtx->emit = fp->func;
  73          return GL_TRUE;
  74       }
  75    }
  76
  77    return GL_FALSE;
  78 }
  79
  80 void _tnl_register_fastpath( struct tnl_clipspace *vtx,
  81                              GLboolean match_strides )
  82 {
  83    struct tnl_clipspace_fastpath *fastpath = CALLOC_STRUCT(tnl_clipspace_fastpath);
  84    GLuint i;
  85
  86    fastpath->vertex_size = vtx->vertex_size;
  87    fastpath->attr_count = vtx->attr_count;
  88    fastpath->match_strides = match_strides;
  89    fastpath->func = vtx->emit;
  90    fastpath->attr = (struct tnl_attr_type *)
  91       _mesa_malloc(vtx->attr_count * sizeof(fastpath->attr[0]));
  92
  93    for (i = 0; i < vtx->attr_count; i++) {
  94       fastpath->attr[i].format = vtx->attr[i].format;
  95       fastpath->attr[i].stride = vtx->attr[i].inputstride;
  96       fastpath->attr[i].size = vtx->attr[i].inputsize;
  97       fastpath->attr[i].offset = vtx->attr[i].vertoffset;
  98    }
  99
 100    fastpath->next = vtx->fastpath;
 101    vtx->fastpath = fastpath;
 102 }
 103
 104
 105
 106 /***********************************************************************
 107  * Build codegen functions or return generic ones:
 108  */
 109 static void choose_emit_func( GLcontext *ctx, GLuint count, GLubyte *dest)
 110 {
 111    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
 112    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 113    struct tnl_clipspace_attr *a = vtx->attr;
 114    const GLuint attr_count = vtx->attr_count;
 115    GLuint j;
 116
 117    for (j = 0; j < attr_count; j++) {
 118       GLvector4f *vptr = VB->AttribPtr[a[j].attrib];
 119       a[j].inputstride = vptr->stride;
 120       a[j].inputsize = vptr->size;
 121       a[j].emit = a[j].insert[vptr->size - 1]; /* not always used */
 122    }
 123
 124    vtx->emit = NULL;
 125
 126    /* Does this match an existing (hardwired, codegen or known-bad)
 127     * fastpath?
 128     */
 129    if (search_fastpath_emit(vtx)) {
 130       /* Use this result.  If it is null, then it is already known
 131        * that the current state will fail for codegen and there is no
 132        * point trying again.
 133        */
 134    }
 135    else if (vtx->codegen_emit) {
 136       vtx->codegen_emit(ctx);
 137    }
 138
 139    if (!vtx->emit) {
 140       _tnl_generate_hardwired_emit(ctx);
 141    }
 142
 143    /* Otherwise use the generic version:
 144     */
 145    if (!vtx->emit)
 146       vtx->emit = _tnl_generic_emit;
 147
 148    vtx->emit( ctx, count, dest );
 149 }
 150
 151
 152
 153 static void choose_interp_func( GLcontext *ctx,
 154                                 GLfloat t,
 155                                 GLuint edst, GLuint eout, GLuint ein,
 156                                 GLboolean force_boundary )
 157 {
 158    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 159
 160    if (vtx->need_extras &&
 161        (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
 162       vtx->interp = _tnl_generic_interp_extras;
 163    } else {
 164       vtx->interp = _tnl_generic_interp;
 165    }
 166
 167    vtx->interp( ctx, t, edst, eout, ein, force_boundary );
 168 }
 169
 170
 171 static void choose_copy_pv_func(  GLcontext *ctx, GLuint edst, GLuint esrc )
 172 {
 173    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 174
 175    if (vtx->need_extras &&
 176        (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
 177       vtx->copy_pv = _tnl_generic_copy_pv_extras;
 178    } else {
 179       vtx->copy_pv = _tnl_generic_copy_pv;
 180    }
 181
 182    vtx->copy_pv( ctx, edst, esrc );
 183 }
 184
 185
 186 /***********************************************************************
 187  * Public entrypoints, mostly dispatch to the above:
 188  */
 189
 190
 191 /* Interpolate between two vertices to produce a third:
 192  */
 193 void _tnl_interp( GLcontext *ctx,
 194                   GLfloat t,
 195                   GLuint edst, GLuint eout, GLuint ein,
 196                   GLboolean force_boundary )
 197 {
 198    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 199    vtx->interp( ctx, t, edst, eout, ein, force_boundary );
 200 }
 201
 202 /* Copy colors from one vertex to another:
 203  */
 204 void _tnl_copy_pv(  GLcontext *ctx, GLuint edst, GLuint esrc )
 205 {
 206    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 207    vtx->copy_pv( ctx, edst, esrc );
 208 }
 209
 210
 211 /* Extract a named attribute from a hardware vertex.  Will have to
 212  * reverse any viewport transformation, swizzling or other conversions
 213  * which may have been applied:
 214  */
 215 void _tnl_get_attr( GLcontext *ctx, const void *vin,
 216                               GLenum attr, GLfloat *dest )
 217 {
 218    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 219    const struct tnl_clipspace_attr *a = vtx->attr;
 220    const GLuint attr_count = vtx->attr_count;
 221    GLuint j;
 222
 223    for (j = 0; j < attr_count; j++) {
 224       if (a[j].attrib == attr) {
 225          a[j].extract( &a[j], dest, (GLubyte *)vin + a[j].vertoffset );
 226          return;
 227       }
 228    }
 229
 230    /* Else return the value from ctx->Current.
 231     */
 232    if (attr == _TNL_ATTRIB_POINTSIZE) {
 233       /* If the hardware vertex doesn't have point size then use size from
 234        * GLcontext.  XXX this will be wrong if drawing attenuated points!
 235        */
 236       dest[0] = ctx->Point.Size;
 237    }
 238    else {
 239       _mesa_memcpy( dest, ctx->Current.Attrib[attr], 4*sizeof(GLfloat));
 240    }
 241 }
 242
 243
 244 /* Complementary operation to the above.
 245  */
 246 void _tnl_set_attr( GLcontext *ctx, void *vout,
 247                     GLenum attr, const GLfloat *src )
 248 {
 249    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 250    const struct tnl_clipspace_attr *a = vtx->attr;
 251    const GLuint attr_count = vtx->attr_count;
 252    GLuint j;
 253
 254    for (j = 0; j < attr_count; j++) {
 255       if (a[j].attrib == attr) {
 256          a[j].insert[4-1]( &a[j], (GLubyte *)vout + a[j].vertoffset, src );
 257          return;
 258       }
 259    }
 260 }
 261
 262
 263 void *_tnl_get_vertex( GLcontext *ctx, GLuint nr )
 264 {
 265    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 266
 267    return vtx->vertex_buf + nr * vtx->vertex_size;
 268 }
 269
 270 void _tnl_invalidate_vertex_state( GLcontext *ctx, GLuint new_state )
 271 {
 272    if (new_state & (_DD_NEW_TRI_LIGHT_TWOSIDE|_DD_NEW_TRI_UNFILLED) ) {
 273       struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 274       vtx->new_inputs = ~0;
 275       vtx->interp = choose_interp_func;
 276       vtx->copy_pv = choose_copy_pv_func;
 277    }
 278 }
 279
 280 static void invalidate_funcs( struct tnl_clipspace *vtx )
 281 {
 282    vtx->emit = choose_emit_func;
 283    vtx->interp = choose_interp_func;
 284    vtx->copy_pv = choose_copy_pv_func;
 285    vtx->new_inputs = ~0;
 286 }
 287
 288 GLuint _tnl_install_attrs( GLcontext *ctx, const struct tnl_attr_map *map,
 289                            GLuint nr, const GLfloat *vp,
 290                            GLuint unpacked_size )
 291 {
 292    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 293    GLuint offset = 0;
 294    GLuint i, j;
 295
 296    assert(nr < _TNL_ATTRIB_MAX);
 297    assert(nr == 0 || map[0].attrib == VERT_ATTRIB_POS);
 298
 299    vtx->new_inputs = ~0;
 300    vtx->need_viewport = GL_FALSE;
 301
 302    if (vp) {
 303       vtx->need_viewport = GL_TRUE;
 304    }
 305
 306    for (j = 0, i = 0; i < nr; i++) {
 307       const GLuint format = map[i].format;
 308       if (format == EMIT_PAD) {
 309          if (DBG)
 310             _mesa_printf("%d: pad %d, offset %d\n", i,
 311                          map[i].offset, offset);
 312
 313          offset += map[i].offset;
 314
 315       }
 316       else {
 317          GLuint tmpoffset;
 318
 319          if (unpacked_size)
 320             tmpoffset = map[i].offset;
 321          else
 322             tmpoffset = offset;
 323
 324          if (vtx->attr_count != j ||
 325              vtx->attr[j].attrib != map[i].attrib ||
 326              vtx->attr[j].format != format ||
 327              vtx->attr[j].vertoffset != tmpoffset) {
 328             invalidate_funcs(vtx);
 329
 330             vtx->attr[j].attrib = map[i].attrib;
 331             vtx->attr[j].format = format;
 332             vtx->attr[j].vp = vp;
 333             vtx->attr[j].insert = _tnl_format_info[format].insert;
 334             vtx->attr[j].extract = _tnl_format_info[format].extract;
 335             vtx->attr[j].vertattrsize = _tnl_format_info[format].attrsize;
 336             vtx->attr[j].vertoffset = tmpoffset;
 337          }
 338
 339
 340          if (DBG)
 341             _mesa_printf("%d: %s, vp %p, offset %d\n", i,
 342                          _tnl_format_info[format].name, (void *)vp,
 343                          vtx->attr[j].vertoffset);
 344
 345          offset += _tnl_format_info[format].attrsize;
 346          j++;
 347       }
 348    }
 349
 350    vtx->attr_count = j;
 351
 352    if (unpacked_size)
 353       vtx->vertex_size = unpacked_size;
 354    else
 355       vtx->vertex_size = offset;
 356
 357    assert(vtx->vertex_size <= vtx->max_vertex_size);
 358    return vtx->vertex_size;
 359 }
 360
 361
 362
 363 void _tnl_invalidate_vertices( GLcontext *ctx, GLuint newinputs )
 364 {
 365    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 366    vtx->new_inputs |= newinputs;
 367 }
 368
 369
 370 /* This event has broader use beyond this file - will move elsewhere
 371  * and probably invoke a driver callback.
 372  */
 373 void _tnl_notify_pipeline_output_change( GLcontext *ctx )
 374 {
 375    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 376    invalidate_funcs(vtx);
 377 }
 378
 379
 380 static void adjust_input_ptrs( GLcontext *ctx, GLint diff)
 381 {
 382    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
 383    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 384    struct tnl_clipspace_attr *a = vtx->attr;
 385    const GLuint count = vtx->attr_count;
 386    int j;
 387
 388    diff -= 1;
 389    for (j=0; j<count; ++j) {
 390            register GLvector4f *vptr = VB->AttribPtr[a->attrib];
 391            (a++)->inputptr += diff*vptr->stride;
 392    }
 393 }
 394
 395 static void update_input_ptrs( GLcontext *ctx, GLuint start )
 396 {
 397    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
 398    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 399    struct tnl_clipspace_attr *a = vtx->attr;
 400    const GLuint count = vtx->attr_count;
 401    GLuint j;
 402
 403    for (j = 0; j < count; j++) {
 404       GLvector4f *vptr = VB->AttribPtr[a[j].attrib];
 405
 406       if (vtx->emit != choose_emit_func) {
 407          assert(a[j].inputstride == vptr->stride);
 408          assert(a[j].inputsize == vptr->size);
 409       }
 410
 411       a[j].inputptr = ((GLubyte *)vptr->data) + start * vptr->stride;
 412    }
 413
 414    if (a->vp) {
 415       vtx->vp_scale[0] = a->vp[MAT_SX];
 416       vtx->vp_scale[1] = a->vp[MAT_SY];
 417       vtx->vp_scale[2] = a->vp[MAT_SZ];
 418       vtx->vp_scale[3] = 1.0;
 419       vtx->vp_xlate[0] = a->vp[MAT_TX];
 420       vtx->vp_xlate[1] = a->vp[MAT_TY];
 421       vtx->vp_xlate[2] = a->vp[MAT_TZ];
 422       vtx->vp_xlate[3] = 0.0;
 423    }
 424 }
 425
 426
 427 void _tnl_build_vertices( GLcontext *ctx,
 428                           GLuint start,
 429                           GLuint end,
 430                           GLuint newinputs )
 431 {
 432    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 433    update_input_ptrs( ctx, start );
 434    vtx->emit( ctx, end - start,
 435               (GLubyte *)(vtx->vertex_buf +
 436                           start * vtx->vertex_size));
 437 }
 438
 439 /* Emit VB vertices start..end to dest.  Note that VB vertex at
 440  * postion start will be emitted to dest at position zero.
 441  */
 442 void *_tnl_emit_vertices_to_buffer( GLcontext *ctx,
 443                                     GLuint start,
 444                                     GLuint end,
 445                                     void *dest )
 446 {
 447    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 448
 449    update_input_ptrs(ctx, start);
 450    /* Note: dest should not be adjusted for non-zero 'start' values:
 451     */
 452    vtx->emit( ctx, end - start, (GLubyte*) dest );
 453    return (void *)((GLubyte *)dest + vtx->vertex_size * (end - start));
 454 }
 455
 456 /* Emit indexed VB vertices start..end to dest.  Note that VB vertex at
 457  * postion start will be emitted to dest at position zero.
 458  */
 459
 460 void *_tnl_emit_indexed_vertices_to_buffer( GLcontext *ctx,
 461                                             const GLuint *elts,
 462                                             GLuint start,
 463                                             GLuint end,
 464                                             void *dest )
 465 {
 466    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 467    GLuint oldIndex;
 468    GLubyte *cdest = dest;
 469
 470    update_input_ptrs(ctx, oldIndex = elts[start++]);
 471    vtx->emit( ctx, 1, cdest );
 472    cdest += vtx->vertex_size;
 473
 474    for (; start < end; ++start) {
 475       adjust_input_ptrs(ctx, elts[start] - oldIndex);
 476       oldIndex = elts[start];
 477       vtx->emit( ctx, 1, cdest);
 478       cdest += vtx->vertex_size;
 479    }
 480
 481    return (void *) cdest;
 482 }
 483
 484
 485 void _tnl_init_vertices( GLcontext *ctx,
 486                         GLuint vb_size,
 487                         GLuint max_vertex_size )
 488 {
 489    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 490
 491    _tnl_install_attrs( ctx, NULL, 0, NULL, 0 );
 492
 493    vtx->need_extras = GL_TRUE;
 494    if (max_vertex_size > vtx->max_vertex_size) {
 495       _tnl_free_vertices( ctx );
 496       vtx->max_vertex_size = max_vertex_size;
 497       vtx->vertex_buf = (GLubyte *)ALIGN_CALLOC(vb_size * max_vertex_size, 32 );
 498       invalidate_funcs(vtx);
 499    }
 500
 501    switch(CHAN_TYPE) {
 502    case GL_UNSIGNED_BYTE:
 503       vtx->chan_scale[0] = 255.0;
 504       vtx->chan_scale[1] = 255.0;
 505       vtx->chan_scale[2] = 255.0;
 506       vtx->chan_scale[3] = 255.0;
 507       break;
 508    case GL_UNSIGNED_SHORT:
 509       vtx->chan_scale[0] = 65535.0;
 510       vtx->chan_scale[1] = 65535.0;
 511       vtx->chan_scale[2] = 65535.0;
 512       vtx->chan_scale[3] = 65535.0;
 513       break;
 514    default:
 515       vtx->chan_scale[0] = 1.0;
 516       vtx->chan_scale[1] = 1.0;
 517       vtx->chan_scale[2] = 1.0;
 518       vtx->chan_scale[3] = 1.0;
 519       break;
 520    }
 521
 522    vtx->identity[0] = 0.0;
 523    vtx->identity[1] = 0.0;
 524    vtx->identity[2] = 0.0;
 525    vtx->identity[3] = 1.0;
 526
 527    vtx->codegen_emit = NULL;
 528
 529 #ifdef USE_SSE_ASM
 530    if (!_mesa_getenv("MESA_NO_CODEGEN"))
 531       vtx->codegen_emit = _tnl_generate_sse_emit;
 532 #endif
 533 }
 534
 535
 536 void _tnl_free_vertices( GLcontext *ctx )
 537 {
 538    TNLcontext *tnl = TNL_CONTEXT(ctx);
 539    if (tnl) {
 540       struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 541       struct tnl_clipspace_fastpath *fp, *tmp;
 542
 543       if (vtx->vertex_buf) {
 544          ALIGN_FREE(vtx->vertex_buf);
 545          vtx->vertex_buf = NULL;
 546       }
 547
 548       for (fp = vtx->fastpath ; fp ; fp = tmp) {
 549          tmp = fp->next;
 550          FREE(fp->attr);
 551
 552          /* KW: At the moment, fp->func is constrained to be allocated by
 553           * _mesa_exec_alloc(), as the hardwired fastpaths in
 554           * t_vertex_generic.c are handled specially.  It would be nice
 555           * to unify them, but this probably won't change until this
 556           * module gets another overhaul.
 557           */
 558          _mesa_exec_free((void *) fp->func);
 559          FREE(fp);
 560       }
 561
 562       vtx->fastpath = NULL;
 563    }
 564 }