src/mesa/state_tracker/st_draw.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * 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
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 /*
  29  * This file implements the st_draw_vbo() function which is called from
  30  * Mesa's VBO module.  All point/line/triangle rendering is done through
  31  * this function whether the user called glBegin/End, glDrawArrays,
  32  * glDrawElements, glEvalMesh, or glCalList, etc.
  33  *
  34  * We basically convert the VBO's vertex attribute/array information into
  35  * Gallium vertex state, bind the vertex buffer objects and call
  36  * pipe->draw_vbo().
  37  *
  38  * Authors:
  39  *   Keith Whitwell <keith@tungstengraphics.com>
  40  */
  41
  42
  43 #include "main/imports.h"
  44 #include "main/image.h"
  45 #include "main/bufferobj.h"
  46 #include "main/macros.h"
  47 #include "main/mfeatures.h"
  48
  49 #include "vbo/vbo.h"
  50
  51 #include "st_context.h"
  52 #include "st_atom.h"
  53 #include "st_cb_bufferobjects.h"
  54 #include "st_cb_xformfb.h"
  55 #include "st_draw.h"
  56 #include "st_program.h"
  57
  58 #include "pipe/p_context.h"
  59 #include "pipe/p_defines.h"
  60 #include "util/u_inlines.h"
  61 #include "util/u_format.h"
  62 #include "util/u_prim.h"
  63 #include "util/u_draw_quad.h"
  64 #include "draw/draw_context.h"
  65 #include "cso_cache/cso_context.h"
  66
  67 #include "../glsl/ir_uniform.h"
  68
  69
  70 static GLuint double_types[4] = {
  71    PIPE_FORMAT_R64_FLOAT,
  72    PIPE_FORMAT_R64G64_FLOAT,
  73    PIPE_FORMAT_R64G64B64_FLOAT,
  74    PIPE_FORMAT_R64G64B64A64_FLOAT
  75 };
  76
  77 static GLuint float_types[4] = {
  78    PIPE_FORMAT_R32_FLOAT,
  79    PIPE_FORMAT_R32G32_FLOAT,
  80    PIPE_FORMAT_R32G32B32_FLOAT,
  81    PIPE_FORMAT_R32G32B32A32_FLOAT
  82 };
  83
  84 static GLuint half_float_types[4] = {
  85    PIPE_FORMAT_R16_FLOAT,
  86    PIPE_FORMAT_R16G16_FLOAT,
  87    PIPE_FORMAT_R16G16B16_FLOAT,
  88    PIPE_FORMAT_R16G16B16A16_FLOAT
  89 };
  90
  91 static GLuint uint_types_norm[4] = {
  92    PIPE_FORMAT_R32_UNORM,
  93    PIPE_FORMAT_R32G32_UNORM,
  94    PIPE_FORMAT_R32G32B32_UNORM,
  95    PIPE_FORMAT_R32G32B32A32_UNORM
  96 };
  97
  98 static GLuint uint_types_scale[4] = {
  99    PIPE_FORMAT_R32_USCALED,
 100    PIPE_FORMAT_R32G32_USCALED,
 101    PIPE_FORMAT_R32G32B32_USCALED,
 102    PIPE_FORMAT_R32G32B32A32_USCALED
 103 };
 104
 105 static GLuint uint_types_int[4] = {
 106    PIPE_FORMAT_R32_UINT,
 107    PIPE_FORMAT_R32G32_UINT,
 108    PIPE_FORMAT_R32G32B32_UINT,
 109    PIPE_FORMAT_R32G32B32A32_UINT
 110 };
 111
 112 static GLuint int_types_norm[4] = {
 113    PIPE_FORMAT_R32_SNORM,
 114    PIPE_FORMAT_R32G32_SNORM,
 115    PIPE_FORMAT_R32G32B32_SNORM,
 116    PIPE_FORMAT_R32G32B32A32_SNORM
 117 };
 118
 119 static GLuint int_types_scale[4] = {
 120    PIPE_FORMAT_R32_SSCALED,
 121    PIPE_FORMAT_R32G32_SSCALED,
 122    PIPE_FORMAT_R32G32B32_SSCALED,
 123    PIPE_FORMAT_R32G32B32A32_SSCALED
 124 };
 125
 126 static GLuint int_types_int[4] = {
 127    PIPE_FORMAT_R32_SINT,
 128    PIPE_FORMAT_R32G32_SINT,
 129    PIPE_FORMAT_R32G32B32_SINT,
 130    PIPE_FORMAT_R32G32B32A32_SINT
 131 };
 132
 133 static GLuint ushort_types_norm[4] = {
 134    PIPE_FORMAT_R16_UNORM,
 135    PIPE_FORMAT_R16G16_UNORM,
 136    PIPE_FORMAT_R16G16B16_UNORM,
 137    PIPE_FORMAT_R16G16B16A16_UNORM
 138 };
 139
 140 static GLuint ushort_types_scale[4] = {
 141    PIPE_FORMAT_R16_USCALED,
 142    PIPE_FORMAT_R16G16_USCALED,
 143    PIPE_FORMAT_R16G16B16_USCALED,
 144    PIPE_FORMAT_R16G16B16A16_USCALED
 145 };
 146
 147 static GLuint ushort_types_int[4] = {
 148    PIPE_FORMAT_R16_UINT,
 149    PIPE_FORMAT_R16G16_UINT,
 150    PIPE_FORMAT_R16G16B16_UINT,
 151    PIPE_FORMAT_R16G16B16A16_UINT
 152 };
 153
 154 static GLuint short_types_norm[4] = {
 155    PIPE_FORMAT_R16_SNORM,
 156    PIPE_FORMAT_R16G16_SNORM,
 157    PIPE_FORMAT_R16G16B16_SNORM,
 158    PIPE_FORMAT_R16G16B16A16_SNORM
 159 };
 160
 161 static GLuint short_types_scale[4] = {
 162    PIPE_FORMAT_R16_SSCALED,
 163    PIPE_FORMAT_R16G16_SSCALED,
 164    PIPE_FORMAT_R16G16B16_SSCALED,
 165    PIPE_FORMAT_R16G16B16A16_SSCALED
 166 };
 167
 168 static GLuint short_types_int[4] = {
 169    PIPE_FORMAT_R16_SINT,
 170    PIPE_FORMAT_R16G16_SINT,
 171    PIPE_FORMAT_R16G16B16_SINT,
 172    PIPE_FORMAT_R16G16B16A16_SINT
 173 };
 174
 175 static GLuint ubyte_types_norm[4] = {
 176    PIPE_FORMAT_R8_UNORM,
 177    PIPE_FORMAT_R8G8_UNORM,
 178    PIPE_FORMAT_R8G8B8_UNORM,
 179    PIPE_FORMAT_R8G8B8A8_UNORM
 180 };
 181
 182 static GLuint ubyte_types_scale[4] = {
 183    PIPE_FORMAT_R8_USCALED,
 184    PIPE_FORMAT_R8G8_USCALED,
 185    PIPE_FORMAT_R8G8B8_USCALED,
 186    PIPE_FORMAT_R8G8B8A8_USCALED
 187 };
 188
 189 static GLuint ubyte_types_int[4] = {
 190    PIPE_FORMAT_R8_UINT,
 191    PIPE_FORMAT_R8G8_UINT,
 192    PIPE_FORMAT_R8G8B8_UINT,
 193    PIPE_FORMAT_R8G8B8A8_UINT
 194 };
 195
 196 static GLuint byte_types_norm[4] = {
 197    PIPE_FORMAT_R8_SNORM,
 198    PIPE_FORMAT_R8G8_SNORM,
 199    PIPE_FORMAT_R8G8B8_SNORM,
 200    PIPE_FORMAT_R8G8B8A8_SNORM
 201 };
 202
 203 static GLuint byte_types_scale[4] = {
 204    PIPE_FORMAT_R8_SSCALED,
 205    PIPE_FORMAT_R8G8_SSCALED,
 206    PIPE_FORMAT_R8G8B8_SSCALED,
 207    PIPE_FORMAT_R8G8B8A8_SSCALED
 208 };
 209
 210 static GLuint byte_types_int[4] = {
 211    PIPE_FORMAT_R8_SINT,
 212    PIPE_FORMAT_R8G8_SINT,
 213    PIPE_FORMAT_R8G8B8_SINT,
 214    PIPE_FORMAT_R8G8B8A8_SINT
 215 };
 216
 217 static GLuint fixed_types[4] = {
 218    PIPE_FORMAT_R32_FIXED,
 219    PIPE_FORMAT_R32G32_FIXED,
 220    PIPE_FORMAT_R32G32B32_FIXED,
 221    PIPE_FORMAT_R32G32B32A32_FIXED
 222 };
 223
 224
 225
 226 /**
 227  * Return a PIPE_FORMAT_x for the given GL datatype and size.
 228  */
 229 enum pipe_format
 230 st_pipe_vertex_format(GLenum type, GLuint size, GLenum format,
 231                       GLboolean normalized, GLboolean integer)
 232 {
 233    assert((type >= GL_BYTE && type <= GL_DOUBLE) ||
 234           type == GL_FIXED || type == GL_HALF_FLOAT ||
 235           type == GL_INT_2_10_10_10_REV ||
 236           type == GL_UNSIGNED_INT_2_10_10_10_REV);
 237    assert(size >= 1);
 238    assert(size <= 4);
 239    assert(format == GL_RGBA || format == GL_BGRA);
 240
 241    if (type == GL_INT_2_10_10_10_REV ||
 242        type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 243       assert(size == 4);
 244       assert(!integer);
 245
 246       if (format == GL_BGRA) {
 247          if (type == GL_INT_2_10_10_10_REV) {
 248             if (normalized)
 249                return PIPE_FORMAT_B10G10R10A2_SNORM;
 250             else
 251                return PIPE_FORMAT_B10G10R10A2_SSCALED;
 252          } else {
 253             if (normalized)
 254                return PIPE_FORMAT_B10G10R10A2_UNORM;
 255             else
 256                return PIPE_FORMAT_B10G10R10A2_USCALED;
 257          }
 258       } else {
 259          if (type == GL_INT_2_10_10_10_REV) {
 260             if (normalized)
 261                return PIPE_FORMAT_R10G10B10A2_SNORM;
 262             else
 263                return PIPE_FORMAT_R10G10B10A2_SSCALED;
 264          } else {
 265             if (normalized)
 266                return PIPE_FORMAT_R10G10B10A2_UNORM;
 267             else
 268                return PIPE_FORMAT_R10G10B10A2_USCALED;
 269          }
 270       }
 271    }
 272
 273    if (format == GL_BGRA) {
 274       /* this is an odd-ball case */
 275       assert(type == GL_UNSIGNED_BYTE);
 276       assert(normalized);
 277       return PIPE_FORMAT_B8G8R8A8_UNORM;
 278    }
 279
 280    if (integer) {
 281       switch (type) {
 282       case GL_INT: return int_types_int[size-1];
 283       case GL_SHORT: return short_types_int[size-1];
 284       case GL_BYTE: return byte_types_int[size-1];
 285       case GL_UNSIGNED_INT: return uint_types_int[size-1];
 286       case GL_UNSIGNED_SHORT: return ushort_types_int[size-1];
 287       case GL_UNSIGNED_BYTE: return ubyte_types_int[size-1];
 288       default: assert(0); return 0;
 289       }
 290    }
 291    else if (normalized) {
 292       switch (type) {
 293       case GL_DOUBLE: return double_types[size-1];
 294       case GL_FLOAT: return float_types[size-1];
 295       case GL_HALF_FLOAT: return half_float_types[size-1];
 296       case GL_INT: return int_types_norm[size-1];
 297       case GL_SHORT: return short_types_norm[size-1];
 298       case GL_BYTE: return byte_types_norm[size-1];
 299       case GL_UNSIGNED_INT: return uint_types_norm[size-1];
 300       case GL_UNSIGNED_SHORT: return ushort_types_norm[size-1];
 301       case GL_UNSIGNED_BYTE: return ubyte_types_norm[size-1];
 302       case GL_FIXED: return fixed_types[size-1];
 303       default: assert(0); return 0;
 304       }
 305    }
 306    else {
 307       switch (type) {
 308       case GL_DOUBLE: return double_types[size-1];
 309       case GL_FLOAT: return float_types[size-1];
 310       case GL_HALF_FLOAT: return half_float_types[size-1];
 311       case GL_INT: return int_types_scale[size-1];
 312       case GL_SHORT: return short_types_scale[size-1];
 313       case GL_BYTE: return byte_types_scale[size-1];
 314       case GL_UNSIGNED_INT: return uint_types_scale[size-1];
 315       case GL_UNSIGNED_SHORT: return ushort_types_scale[size-1];
 316       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size-1];
 317       case GL_FIXED: return fixed_types[size-1];
 318       default: assert(0); return 0;
 319       }
 320    }
 321    return PIPE_FORMAT_NONE; /* silence compiler warning */
 322 }
 323
 324
 325 /**
 326  * This is very similar to vbo_all_varyings_in_vbos() but we are
 327  * only interested in per-vertex data.  See bug 38626.
 328  */
 329 static GLboolean
 330 all_varyings_in_vbos(const struct gl_client_array *arrays[])
 331 {
 332    GLuint i;
 333
 334    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 335       if (arrays[i]->StrideB &&
 336           !arrays[i]->InstanceDivisor &&
 337           !_mesa_is_bufferobj(arrays[i]->BufferObj))
 338          return GL_FALSE;
 339
 340    return GL_TRUE;
 341 }
 342
 343
 344 /**
 345  * Examine the active arrays to determine if we have interleaved
 346  * vertex arrays all living in one VBO, or all living in user space.
 347  */
 348 static GLboolean
 349 is_interleaved_arrays(const struct st_vertex_program *vp,
 350                       const struct st_vp_variant *vpv,
 351                       const struct gl_client_array **arrays)
 352 {
 353    GLuint attr;
 354    const struct gl_buffer_object *firstBufObj = NULL;
 355    GLint firstStride = -1;
 356    const GLubyte *firstPtr = NULL;
 357    GLboolean userSpaceBuffer = GL_FALSE;
 358
 359    for (attr = 0; attr < vpv->num_inputs; attr++) {
 360       const GLuint mesaAttr = vp->index_to_input[attr];
 361       const struct gl_client_array *array = arrays[mesaAttr];
 362       const struct gl_buffer_object *bufObj = array->BufferObj;
 363       const GLsizei stride = array->StrideB; /* in bytes */
 364
 365       if (attr == 0) {
 366          /* save info about the first array */
 367          firstStride = stride;
 368          firstPtr = array->Ptr;
 369          firstBufObj = bufObj;
 370          userSpaceBuffer = !bufObj || !bufObj->Name;
 371       }
 372       else {
 373          /* check if other arrays interleave with the first, in same buffer */
 374          if (stride != firstStride)
 375             return GL_FALSE; /* strides don't match */
 376
 377          if (bufObj != firstBufObj)
 378             return GL_FALSE; /* arrays in different VBOs */
 379
 380          if (abs(array->Ptr - firstPtr) > firstStride)
 381             return GL_FALSE; /* arrays start too far apart */
 382
 383          if ((!_mesa_is_bufferobj(bufObj)) != userSpaceBuffer)
 384             return GL_FALSE; /* mix of VBO and user-space arrays */
 385       }
 386    }
 387
 388    return GL_TRUE;
 389 }
 390
 391
 392 /**
 393  * Set up for drawing interleaved arrays that all live in one VBO
 394  * or all live in user space.
 395  * \param vbuffer  returns vertex buffer info
 396  * \param velements  returns vertex element info
 397  * \return GL_TRUE for success, GL_FALSE otherwise (probably out of memory)
 398  */
 399 static GLboolean
 400 setup_interleaved_attribs(struct gl_context *ctx,
 401                           const struct st_vertex_program *vp,
 402                           const struct st_vp_variant *vpv,
 403                           const struct gl_client_array **arrays,
 404                           struct pipe_vertex_buffer *vbuffer,
 405                           struct pipe_vertex_element velements[],
 406                           unsigned max_index,
 407                           unsigned num_instances)
 408 {
 409    struct st_context *st = st_context(ctx);
 410    struct pipe_context *pipe = st->pipe;
 411    GLuint attr;
 412    const GLubyte *low_addr = NULL;
 413    GLboolean usingVBO;      /* all arrays in a VBO? */
 414    struct gl_buffer_object *bufobj;
 415    GLuint user_buffer_size = 0;
 416    GLuint vertex_size = 0;  /* bytes per vertex, in bytes */
 417    GLsizei stride;
 418
 419    /* Find the lowest address of the arrays we're drawing,
 420     * Init bufobj and stride.
 421     */
 422    if (vpv->num_inputs) {
 423       const GLuint mesaAttr0 = vp->index_to_input[0];
 424       const struct gl_client_array *array = arrays[mesaAttr0];
 425
 426       /* Since we're doing interleaved arrays, we know there'll be at most
 427        * one buffer object and the stride will be the same for all arrays.
 428        * Grab them now.
 429        */
 430       bufobj = array->BufferObj;
 431       stride = array->StrideB;
 432
 433       low_addr = arrays[vp->index_to_input[0]]->Ptr;
 434
 435       for (attr = 1; attr < vpv->num_inputs; attr++) {
 436          const GLubyte *start = arrays[vp->index_to_input[attr]]->Ptr;
 437          low_addr = MIN2(low_addr, start);
 438       }
 439    }
 440    else {
 441       /* not sure we'll ever have zero inputs, but play it safe */
 442       bufobj = NULL;
 443       stride = 0;
 444       low_addr = 0;
 445    }
 446
 447    /* are the arrays in user space? */
 448    usingVBO = _mesa_is_bufferobj(bufobj);
 449
 450    for (attr = 0; attr < vpv->num_inputs; attr++) {
 451       const GLuint mesaAttr = vp->index_to_input[attr];
 452       const struct gl_client_array *array = arrays[mesaAttr];
 453       unsigned src_offset = (unsigned) (array->Ptr - low_addr);
 454       GLuint element_size = array->_ElementSize;
 455
 456       assert(element_size == array->Size * _mesa_sizeof_type(array->Type));
 457
 458       velements[attr].src_offset = src_offset;
 459       velements[attr].instance_divisor = array->InstanceDivisor;
 460       velements[attr].vertex_buffer_index = 0;
 461       velements[attr].src_format = st_pipe_vertex_format(array->Type,
 462                                                          array->Size,
 463                                                          array->Format,
 464                                                          array->Normalized,
 465                                                          array->Integer);
 466       assert(velements[attr].src_format);
 467
 468       if (!usingVBO) {
 469          /* how many bytes referenced by this attribute array? */
 470          uint divisor = array->InstanceDivisor;
 471          uint last_index = divisor ? num_instances / divisor : max_index;
 472          uint bytes = src_offset + stride * last_index + element_size;
 473
 474          user_buffer_size = MAX2(user_buffer_size, bytes);
 475
 476          /* update vertex size */
 477          vertex_size = MAX2(vertex_size, src_offset + element_size);
 478       }
 479    }
 480
 481    /*
 482     * Return the vbuffer info and setup user-space attrib info, if needed.
 483     */
 484    if (vpv->num_inputs == 0) {
 485       /* just defensive coding here */
 486       vbuffer->buffer = NULL;
 487       vbuffer->buffer_offset = 0;
 488       vbuffer->stride = 0;
 489       st->num_user_attribs = 0;
 490    }
 491    else if (usingVBO) {
 492       /* all interleaved arrays in a VBO */
 493       struct st_buffer_object *stobj = st_buffer_object(bufobj);
 494
 495       if (!stobj || !stobj->buffer) {
 496          /* probably out of memory (or zero-sized buffer) */
 497          return GL_FALSE;
 498       }
 499
 500       vbuffer->buffer = NULL;
 501       pipe_resource_reference(&vbuffer->buffer, stobj->buffer);
 502       vbuffer->buffer_offset = pointer_to_offset(low_addr);
 503       vbuffer->stride = stride;
 504       st->num_user_attribs = 0;
 505    }
 506    else {
 507       /* all interleaved arrays in user memory */
 508       vbuffer->buffer = pipe_user_buffer_create(pipe->screen,
 509                                                 (void*) low_addr,
 510                                                 user_buffer_size,
 511                                                 PIPE_BIND_VERTEX_BUFFER);
 512       vbuffer->buffer_offset = 0;
 513       vbuffer->stride = stride;
 514
 515       /* Track user vertex buffers. */
 516       pipe_resource_reference(&st->user_attrib[0].buffer, vbuffer->buffer);
 517       st->user_attrib[0].element_size = vertex_size;
 518       st->user_attrib[0].stride = stride;
 519       st->num_user_attribs = 1;
 520    }
 521
 522    return GL_TRUE;
 523 }
 524
 525
 526 /**
 527  * Set up a separate pipe_vertex_buffer and pipe_vertex_element for each
 528  * vertex attribute.
 529  * \param vbuffer  returns vertex buffer info
 530  * \param velements  returns vertex element info
 531  * \return GL_TRUE for success, GL_FALSE otherwise (probably out of memory)
 532  */
 533 static GLboolean
 534 setup_non_interleaved_attribs(struct gl_context *ctx,
 535                               const struct st_vertex_program *vp,
 536                               const struct st_vp_variant *vpv,
 537                               const struct gl_client_array **arrays,
 538                               struct pipe_vertex_buffer vbuffer[],
 539                               struct pipe_vertex_element velements[],
 540                               unsigned max_index,
 541                               unsigned num_instances)
 542 {
 543    struct st_context *st = st_context(ctx);
 544    struct pipe_context *pipe = st->pipe;
 545    GLuint attr;
 546
 547    for (attr = 0; attr < vpv->num_inputs; attr++) {
 548       const GLuint mesaAttr = vp->index_to_input[attr];
 549       const struct gl_client_array *array = arrays[mesaAttr];
 550       struct gl_buffer_object *bufobj = array->BufferObj;
 551       GLuint element_size = array->_ElementSize;
 552       GLsizei stride = array->StrideB;
 553
 554       assert(element_size == array->Size * _mesa_sizeof_type(array->Type));
 555
 556       if (_mesa_is_bufferobj(bufobj)) {
 557          /* Attribute data is in a VBO.
 558           * Recall that for VBOs, the gl_client_array->Ptr field is
 559           * really an offset from the start of the VBO, not a pointer.
 560           */
 561          struct st_buffer_object *stobj = st_buffer_object(bufobj);
 562
 563          if (!stobj || !stobj->buffer) {
 564             /* probably out of memory (or zero-sized buffer) */
 565             return GL_FALSE;
 566          }
 567
 568          vbuffer[attr].buffer = NULL;
 569          pipe_resource_reference(&vbuffer[attr].buffer, stobj->buffer);
 570          vbuffer[attr].buffer_offset = pointer_to_offset(array->Ptr);
 571       }
 572       else {
 573          /* wrap user data */
 574          uint bytes;
 575          void *ptr;
 576
 577          if (array->Ptr) {
 578             uint divisor = array->InstanceDivisor;
 579             uint last_index = divisor ? num_instances / divisor : max_index;
 580
 581             bytes = stride * last_index + element_size;
 582
 583             ptr = (void *) array->Ptr;
 584          }
 585          else {
 586             /* no array, use ctx->Current.Attrib[] value */
 587             bytes = element_size = sizeof(ctx->Current.Attrib[0]);
 588             ptr = (void *) ctx->Current.Attrib[mesaAttr];
 589             stride = 0;
 590          }
 591
 592          assert(ptr);
 593          assert(bytes);
 594
 595          vbuffer[attr].buffer =
 596             pipe_user_buffer_create(pipe->screen, ptr, bytes,
 597                                     PIPE_BIND_VERTEX_BUFFER);
 598
 599          vbuffer[attr].buffer_offset = 0;
 600
 601          /* Track user vertex buffers. */
 602          pipe_resource_reference(&st->user_attrib[attr].buffer, vbuffer[attr].buffer);
 603          st->user_attrib[attr].element_size = element_size;
 604          st->user_attrib[attr].stride = stride;
 605          st->num_user_attribs = MAX2(st->num_user_attribs, attr + 1);
 606
 607          if (!vbuffer[attr].buffer) {
 608             /* probably ran out of memory */
 609             return GL_FALSE;
 610          }
 611       }
 612
 613       /* common-case setup */
 614       vbuffer[attr].stride = stride; /* in bytes */
 615
 616       velements[attr].src_offset = 0;
 617       velements[attr].instance_divisor = array->InstanceDivisor;
 618       velements[attr].vertex_buffer_index = attr;
 619       velements[attr].src_format = st_pipe_vertex_format(array->Type,
 620                                                          array->Size,
 621                                                          array->Format,
 622                                                          array->Normalized,
 623                                                          array->Integer);
 624       assert(velements[attr].src_format);
 625    }
 626
 627    return GL_TRUE;
 628 }
 629
 630
 631 static void
 632 setup_index_buffer(struct gl_context *ctx,
 633                    const struct _mesa_index_buffer *ib,
 634                    struct pipe_index_buffer *ibuffer)
 635 {
 636    struct st_context *st = st_context(ctx);
 637    struct pipe_context *pipe = st->pipe;
 638
 639    memset(ibuffer, 0, sizeof(*ibuffer));
 640    if (ib) {
 641       struct gl_buffer_object *bufobj = ib->obj;
 642
 643       ibuffer->index_size = vbo_sizeof_ib_type(ib->type);
 644
 645       /* get/create the index buffer object */
 646       if (_mesa_is_bufferobj(bufobj)) {
 647          /* elements/indexes are in a real VBO */
 648          struct st_buffer_object *stobj = st_buffer_object(bufobj);
 649          pipe_resource_reference(&ibuffer->buffer, stobj->buffer);
 650          ibuffer->offset = pointer_to_offset(ib->ptr);
 651       }
 652       else {
 653          /* element/indicies are in user space memory */
 654          ibuffer->buffer =
 655             pipe_user_buffer_create(pipe->screen, (void *) ib->ptr,
 656                                     ib->count * ibuffer->index_size,
 657                                     PIPE_BIND_INDEX_BUFFER);
 658       }
 659    }
 660 }
 661
 662
 663 /**
 664  * Prior to drawing, check that any uniforms referenced by the
 665  * current shader have been set.  If a uniform has not been set,
 666  * issue a warning.
 667  */
 668 static void
 669 check_uniforms(struct gl_context *ctx)
 670 {
 671    struct gl_shader_program *shProg[3] = {
 672       ctx->Shader.CurrentVertexProgram,
 673       ctx->Shader.CurrentGeometryProgram,
 674       ctx->Shader.CurrentFragmentProgram,
 675    };
 676    unsigned j;
 677
 678    for (j = 0; j < 3; j++) {
 679       unsigned i;
 680
 681       if (shProg[j] == NULL || !shProg[j]->LinkStatus)
 682          continue;
 683
 684       for (i = 0; i < shProg[j]->NumUserUniformStorage; i++) {
 685          const struct gl_uniform_storage *u = &shProg[j]->UniformStorage[i];
 686          if (!u->initialized) {
 687             _mesa_warning(ctx,
 688                           "Using shader with uninitialized uniform: %s",
 689                           u->name);
 690          }
 691       }
 692    }
 693 }
 694
 695
 696 /*
 697  * Notes on primitive restart:
 698  * The code below is used when the gallium driver does not support primitive
 699  * restart itself.  We map the index buffer, find the restart indexes, unmap
 700  * the index buffer then draw the sub-primitives delineated by the restarts.
 701  * A couple possible optimizations:
 702  * 1. Save the list of sub-primitive (start, count) values in a list attached
 703  *    to the index buffer for re-use in subsequent draws.  The list would be
 704  *    invalidated when the contents of the buffer changed.
 705  * 2. If drawing triangle strips or quad strips, create a new index buffer
 706  *    that uses duplicated vertices to render the disjoint strips as one
 707  *    long strip.  We'd have to be careful to avoid using too much memory
 708  *    for this.
 709  * Finally, some apps might perform better if they don't use primitive restart
 710  * at all rather than this fallback path.  Set MESA_EXTENSION_OVERRIDE to
 711  * "-GL_NV_primitive_restart" to test that.
 712  */
 713
 714
 715 struct sub_primitive
 716 {
 717    unsigned start, count;
 718 };
 719
 720
 721 /**
 722  * Scan the elements array to find restart indexes.  Return a list
 723  * of primitive (start,count) pairs to indicate how to draw the sub-
 724  * primitives delineated by the restart index.
 725  */
 726 static struct sub_primitive *
 727 find_sub_primitives(const void *elements, unsigned element_size,
 728                     unsigned start, unsigned end, unsigned restart_index,
 729                     unsigned *num_sub_prims)
 730 {
 731    const unsigned max_prims = end - start;
 732    struct sub_primitive *sub_prims;
 733    unsigned i, cur_start, cur_count, num;
 734
 735    sub_prims = (struct sub_primitive *)
 736       malloc(max_prims * sizeof(struct sub_primitive));
 737
 738    if (!sub_prims) {
 739       *num_sub_prims = 0;
 740       return NULL;
 741    }
 742
 743    cur_start = start;
 744    cur_count = 0;
 745    num = 0;
 746
 747 #define SCAN_ELEMENTS(TYPE) \
 748    for (i = start; i < end; i++) { \
 749       if (((const TYPE *) elements)[i] == restart_index) { \
 750          if (cur_count > 0) { \
 751             assert(num < max_prims); \
 752             sub_prims[num].start = cur_start; \
 753             sub_prims[num].count = cur_count; \
 754             num++; \
 755          } \
 756          cur_start = i + 1; \
 757          cur_count = 0; \
 758       } \
 759       else { \
 760          cur_count++; \
 761       } \
 762    } \
 763    if (cur_count > 0) { \
 764       assert(num < max_prims); \
 765       sub_prims[num].start = cur_start; \
 766       sub_prims[num].count = cur_count; \
 767       num++; \
 768    }
 769
 770    switch (element_size) {
 771    case 1:
 772       SCAN_ELEMENTS(ubyte);
 773       break;
 774    case 2:
 775       SCAN_ELEMENTS(ushort);
 776       break;
 777    case 4:
 778       SCAN_ELEMENTS(uint);
 779       break;
 780    default:
 781       assert(0 && "bad index_size in find_sub_primitives()");
 782    }
 783
 784 #undef SCAN_ELEMENTS
 785
 786    *num_sub_prims = num;
 787
 788    return sub_prims;
 789 }
 790
 791
 792 /**
 793  * For gallium drivers that don't support the primitive restart
 794  * feature, handle it here by breaking up the indexed primitive into
 795  * sub-primitives.
 796  */
 797 static void
 798 handle_fallback_primitive_restart(struct pipe_context *pipe,
 799                                   const struct _mesa_index_buffer *ib,
 800                                   struct pipe_index_buffer *ibuffer,
 801                                   struct pipe_draw_info *orig_info)
 802 {
 803    const unsigned start = orig_info->start;
 804    const unsigned count = orig_info->count;
 805    struct pipe_draw_info info = *orig_info;
 806    struct pipe_transfer *transfer = NULL;
 807    unsigned instance, i;
 808    const void *ptr = NULL;
 809    struct sub_primitive *sub_prims;
 810    unsigned num_sub_prims;
 811
 812    assert(info.indexed);
 813    assert(ibuffer->buffer);
 814    assert(ib);
 815
 816    if (!ibuffer->buffer || !ib)
 817       return;
 818
 819    info.primitive_restart = FALSE;
 820    info.instance_count = 1;
 821
 822    if (_mesa_is_bufferobj(ib->obj)) {
 823       ptr = pipe_buffer_map_range(pipe, ibuffer->buffer,
 824                                   start * ibuffer->index_size, /* start */
 825                                   count * ibuffer->index_size, /* length */
 826                                   PIPE_TRANSFER_READ, &transfer);
 827    }
 828    else {
 829       ptr = ib->ptr;
 830    }
 831
 832    if (!ptr)
 833       return;
 834
 835    ptr = ADD_POINTERS(ptr, ibuffer->offset);
 836
 837    sub_prims = find_sub_primitives(ptr, ibuffer->index_size,
 838                                    0, count, orig_info->restart_index,
 839                                    &num_sub_prims);
 840
 841    if (transfer)
 842       pipe_buffer_unmap(pipe, transfer);
 843
 844    /* Now draw the sub primitives.
 845     * Need to loop over instances as well to preserve draw order.
 846     */
 847    for (instance = 0; instance < orig_info->instance_count; instance++) {
 848       info.start_instance = instance + orig_info->start_instance;
 849       for (i = 0; i < num_sub_prims; i++) {
 850          info.start = sub_prims[i].start;
 851          info.count = sub_prims[i].count;
 852          if (u_trim_pipe_prim(info.mode, &info.count)) {
 853             pipe->draw_vbo(pipe, &info);
 854          }
 855       }
 856    }
 857
 858    if (sub_prims)
 859       free(sub_prims);
 860 }
 861
 862
 863 /**
 864  * Translate OpenGL primtive type (GL_POINTS, GL_TRIANGLE_STRIP, etc) to
 865  * the corresponding Gallium type.
 866  */
 867 static unsigned
 868 translate_prim(const struct gl_context *ctx, unsigned prim)
 869 {
 870    /* GL prims should match Gallium prims, spot-check a few */
 871    assert(GL_POINTS == PIPE_PRIM_POINTS);
 872    assert(GL_QUADS == PIPE_PRIM_QUADS);
 873    assert(GL_TRIANGLE_STRIP_ADJACENCY == PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY);
 874
 875    /* Avoid quadstrips if it's easy to do so:
 876     * Note: it's important to do the correct trimming if we change the
 877     * prim type!  We do that wherever this function is called.
 878     */
 879    if (prim == GL_QUAD_STRIP &&
 880        ctx->Light.ShadeModel != GL_FLAT &&
 881        ctx->Polygon.FrontMode == GL_FILL &&
 882        ctx->Polygon.BackMode == GL_FILL)
 883       prim = GL_TRIANGLE_STRIP;
 884
 885    return prim;
 886 }
 887
 888
 889 /**
 890  * Setup vertex arrays and buffers prior to drawing.
 891  * \return GL_TRUE for success, GL_FALSE otherwise (probably out of memory)
 892  */
 893 static GLboolean
 894 st_validate_varrays(struct gl_context *ctx,
 895                     const struct gl_client_array **arrays,
 896                     unsigned max_index,
 897                     unsigned num_instances)
 898 {
 899    struct st_context *st = st_context(ctx);
 900    const struct st_vertex_program *vp;
 901    const struct st_vp_variant *vpv;
 902    struct pipe_vertex_buffer vbuffer[PIPE_MAX_SHADER_INPUTS];
 903    struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];
 904    unsigned num_vbuffers, num_velements;
 905    GLuint attr;
 906    unsigned i;
 907
 908    /* must get these after state validation! */
 909    vp = st->vp;
 910    vpv = st->vp_variant;
 911
 912    memset(velements, 0, sizeof(struct pipe_vertex_element) * vpv->num_inputs);
 913
 914    /* Unreference any user vertex buffers. */
 915    for (i = 0; i < st->num_user_attribs; i++) {
 916       pipe_resource_reference(&st->user_attrib[i].buffer, NULL);
 917    }
 918    st->num_user_attribs = 0;
 919
 920    /*
 921     * Setup the vbuffer[] and velements[] arrays.
 922     */
 923    if (is_interleaved_arrays(vp, vpv, arrays)) {
 924       if (!setup_interleaved_attribs(ctx, vp, vpv, arrays, vbuffer, velements,
 925                                      max_index, num_instances)) {
 926          return GL_FALSE;
 927       }
 928
 929       num_vbuffers = 1;
 930       num_velements = vpv->num_inputs;
 931       if (num_velements == 0)
 932          num_vbuffers = 0;
 933    }
 934    else {
 935       if (!setup_non_interleaved_attribs(ctx, vp, vpv, arrays,
 936                                          vbuffer, velements, max_index,
 937                                          num_instances)) {
 938          return GL_FALSE;
 939       }
 940
 941       num_vbuffers = vpv->num_inputs;
 942       num_velements = vpv->num_inputs;
 943    }
 944
 945    cso_set_vertex_buffers(st->cso_context, num_vbuffers, vbuffer);
 946    cso_set_vertex_elements(st->cso_context, num_velements, velements);
 947
 948    /* unreference buffers (frees wrapped user-space buffer objects)
 949     * This is OK, because the pipe driver should reference buffers by itself
 950     * in set_vertex_buffers. */
 951    for (attr = 0; attr < num_vbuffers; attr++) {
 952       pipe_resource_reference(&vbuffer[attr].buffer, NULL);
 953       assert(!vbuffer[attr].buffer);
 954    }
 955
 956    return GL_TRUE;
 957 }
 958
 959
 960 /**
 961  * This function gets plugged into the VBO module and is called when
 962  * we have something to render.
 963  * Basically, translate the information into the format expected by gallium.
 964  */
 965 void
 966 st_draw_vbo(struct gl_context *ctx,
 967             const struct gl_client_array **arrays,
 968             const struct _mesa_prim *prims,
 969             GLuint nr_prims,
 970             const struct _mesa_index_buffer *ib,
 971             GLboolean index_bounds_valid,
 972             GLuint min_index,
 973             GLuint max_index,
 974             struct gl_transform_feedback_object *tfb_vertcount)
 975 {
 976    struct st_context *st = st_context(ctx);
 977    struct pipe_context *pipe = st->pipe;
 978    struct pipe_index_buffer ibuffer;
 979    struct pipe_draw_info info;
 980    unsigned i, num_instances = 1;
 981    GLboolean new_array =
 982       st->dirty.st &&
 983       (st->dirty.mesa & (_NEW_ARRAY | _NEW_PROGRAM | _NEW_BUFFER_OBJECT)) != 0;
 984
 985    /* Mesa core state should have been validated already */
 986    assert(ctx->NewState == 0x0);
 987
 988    if (ib) {
 989       /* Gallium probably doesn't want this in some cases. */
 990       if (!index_bounds_valid)
 991          if (!all_varyings_in_vbos(arrays))
 992             vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index);
 993
 994       for (i = 0; i < nr_prims; i++) {
 995          num_instances = MAX2(num_instances, prims[i].num_instances);
 996       }
 997    }
 998    else {
 999       /* Get min/max index for non-indexed drawing. */
1000       min_index = ~0;
1001       max_index = 0;
1002
1003       for (i = 0; i < nr_prims; i++) {
1004          min_index = MIN2(min_index, prims[i].start);
1005          max_index = MAX2(max_index, prims[i].start + prims[i].count - 1);
1006          num_instances = MAX2(num_instances, prims[i].num_instances);
1007       }
1008    }
1009
1010    /* Validate state. */
1011    if (st->dirty.st) {
1012       GLboolean vertDataEdgeFlags;
1013
1014       /* sanity check for pointer arithmetic below */
1015       assert(sizeof(arrays[0]->Ptr[0]) == 1);
1016
1017       vertDataEdgeFlags = arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj &&
1018                           arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj->Name;
1019       if (vertDataEdgeFlags != st->vertdata_edgeflags) {
1020          st->vertdata_edgeflags = vertDataEdgeFlags;
1021          st->dirty.st |= ST_NEW_EDGEFLAGS_DATA;
1022       }
1023
1024       st_validate_state(st);
1025
1026       if (new_array) {
1027          if (!st_validate_varrays(ctx, arrays, max_index, num_instances)) {
1028             /* probably out of memory, no-op the draw call */
1029             return;
1030          }
1031       }
1032
1033 #if 0
1034       if (MESA_VERBOSE & VERBOSE_GLSL) {
1035          check_uniforms(ctx);
1036       }
1037 #else
1038       (void) check_uniforms;
1039 #endif
1040    }
1041
1042    /* Notify the driver that the content of user buffers may have been
1043     * changed. */
1044    assert(max_index >= min_index);
1045    if (!new_array && st->num_user_attribs) {
1046       for (i = 0; i < st->num_user_attribs; i++) {
1047          if (st->user_attrib[i].buffer) {
1048             unsigned element_size = st->user_attrib[i].element_size;
1049             unsigned stride = st->user_attrib[i].stride;
1050             unsigned min_offset = min_index * stride;
1051             unsigned max_offset = max_index * stride + element_size;
1052
1053             assert(max_offset > min_offset);
1054
1055             pipe->redefine_user_buffer(pipe, st->user_attrib[i].buffer,
1056                                        min_offset,
1057                                        max_offset - min_offset);
1058          }
1059       }
1060    }
1061
1062    setup_index_buffer(ctx, ib, &ibuffer);
1063    pipe->set_index_buffer(pipe, &ibuffer);
1064
1065    util_draw_init_info(&info);
1066    if (ib) {
1067       info.indexed = TRUE;
1068       if (min_index != ~0 && max_index != ~0) {
1069          info.min_index = min_index;
1070          info.max_index = max_index;
1071       }
1072
1073       /* The VBO module handles restart for the non-indexed GLDrawArrays
1074        * so we only set these fields for indexed drawing:
1075        */
1076       info.primitive_restart = ctx->Array.PrimitiveRestart;
1077       info.restart_index = ctx->Array.RestartIndex;
1078    }
1079
1080    /* Set info.count_from_stream_output. */
1081    if (tfb_vertcount) {
1082       st_transform_feedback_draw_init(tfb_vertcount, &info);
1083    }
1084
1085    /* do actual drawing */
1086    for (i = 0; i < nr_prims; i++) {
1087       info.mode = translate_prim( ctx, prims[i].mode );
1088       info.start = prims[i].start;
1089       info.count = prims[i].count;
1090       info.instance_count = prims[i].num_instances;
1091       info.index_bias = prims[i].basevertex;
1092       if (!ib) {
1093          info.min_index = info.start;
1094          info.max_index = info.start + info.count - 1;
1095       }
1096
1097       if (info.count_from_stream_output) {
1098          pipe->draw_vbo(pipe, &info);
1099       }
1100       else if (info.primitive_restart) {
1101          if (st->sw_primitive_restart) {
1102             /* Handle primitive restart for drivers that doesn't support it */
1103             handle_fallback_primitive_restart(pipe, ib, &ibuffer, &info);
1104          }
1105          else {
1106             /* don't trim, restarts might be inside index list */
1107             pipe->draw_vbo(pipe, &info);
1108          }
1109       }
1110       else if (u_trim_pipe_prim(info.mode, &info.count))
1111          pipe->draw_vbo(pipe, &info);
1112    }
1113
1114    pipe_resource_reference(&ibuffer.buffer, NULL);
1115 }
1116
1117
1118 void
1119 st_init_draw(struct st_context *st)
1120 {
1121    struct gl_context *ctx = st->ctx;
1122
1123    vbo_set_draw_func(ctx, st_draw_vbo);
1124
1125 #if FEATURE_feedback || FEATURE_rastpos
1126    st->draw = draw_create(st->pipe); /* for selection/feedback */
1127
1128    /* Disable draw options that might convert points/lines to tris, etc.
1129     * as that would foul-up feedback/selection mode.
1130     */
1131    draw_wide_line_threshold(st->draw, 1000.0f);
1132    draw_wide_point_threshold(st->draw, 1000.0f);
1133    draw_enable_line_stipple(st->draw, FALSE);
1134    draw_enable_point_sprites(st->draw, FALSE);
1135 #endif
1136 }
1137
1138
1139 void
1140 st_destroy_draw(struct st_context *st)
1141 {
1142 #if FEATURE_feedback || FEATURE_rastpos
1143    draw_destroy(st->draw);
1144 #endif
1145 }