src/mesa/drivers/dri/i965/brw_draw_upload.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 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 #undef NDEBUG
  29
  30 #include "main/glheader.h"
  31 #include "main/bufferobj.h"
  32 #include "main/context.h"
  33 #include "main/enums.h"
  34
  35 #include "brw_draw.h"
  36 #include "brw_defines.h"
  37 #include "brw_context.h"
  38 #include "brw_state.h"
  39
  40 #include "intel_batchbuffer.h"
  41 #include "intel_buffer_objects.h"
  42
  43 static GLuint double_types[5] = {
  44    0,
  45    BRW_SURFACEFORMAT_R64_FLOAT,
  46    BRW_SURFACEFORMAT_R64G64_FLOAT,
  47    BRW_SURFACEFORMAT_R64G64B64_FLOAT,
  48    BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
  49 };
  50
  51 static GLuint float_types[5] = {
  52    0,
  53    BRW_SURFACEFORMAT_R32_FLOAT,
  54    BRW_SURFACEFORMAT_R32G32_FLOAT,
  55    BRW_SURFACEFORMAT_R32G32B32_FLOAT,
  56    BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
  57 };
  58
  59 static GLuint half_float_types[5] = {
  60    0,
  61    BRW_SURFACEFORMAT_R16_FLOAT,
  62    BRW_SURFACEFORMAT_R16G16_FLOAT,
  63    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT,
  64    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
  65 };
  66
  67 static GLuint uint_types_norm[5] = {
  68    0,
  69    BRW_SURFACEFORMAT_R32_UNORM,
  70    BRW_SURFACEFORMAT_R32G32_UNORM,
  71    BRW_SURFACEFORMAT_R32G32B32_UNORM,
  72    BRW_SURFACEFORMAT_R32G32B32A32_UNORM
  73 };
  74
  75 static GLuint uint_types_scale[5] = {
  76    0,
  77    BRW_SURFACEFORMAT_R32_USCALED,
  78    BRW_SURFACEFORMAT_R32G32_USCALED,
  79    BRW_SURFACEFORMAT_R32G32B32_USCALED,
  80    BRW_SURFACEFORMAT_R32G32B32A32_USCALED
  81 };
  82
  83 static GLuint int_types_norm[5] = {
  84    0,
  85    BRW_SURFACEFORMAT_R32_SNORM,
  86    BRW_SURFACEFORMAT_R32G32_SNORM,
  87    BRW_SURFACEFORMAT_R32G32B32_SNORM,
  88    BRW_SURFACEFORMAT_R32G32B32A32_SNORM
  89 };
  90
  91 static GLuint int_types_scale[5] = {
  92    0,
  93    BRW_SURFACEFORMAT_R32_SSCALED,
  94    BRW_SURFACEFORMAT_R32G32_SSCALED,
  95    BRW_SURFACEFORMAT_R32G32B32_SSCALED,
  96    BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
  97 };
  98
  99 static GLuint ushort_types_norm[5] = {
 100    0,
 101    BRW_SURFACEFORMAT_R16_UNORM,
 102    BRW_SURFACEFORMAT_R16G16_UNORM,
 103    BRW_SURFACEFORMAT_R16G16B16_UNORM,
 104    BRW_SURFACEFORMAT_R16G16B16A16_UNORM
 105 };
 106
 107 static GLuint ushort_types_scale[5] = {
 108    0,
 109    BRW_SURFACEFORMAT_R16_USCALED,
 110    BRW_SURFACEFORMAT_R16G16_USCALED,
 111    BRW_SURFACEFORMAT_R16G16B16_USCALED,
 112    BRW_SURFACEFORMAT_R16G16B16A16_USCALED
 113 };
 114
 115 static GLuint short_types_norm[5] = {
 116    0,
 117    BRW_SURFACEFORMAT_R16_SNORM,
 118    BRW_SURFACEFORMAT_R16G16_SNORM,
 119    BRW_SURFACEFORMAT_R16G16B16_SNORM,
 120    BRW_SURFACEFORMAT_R16G16B16A16_SNORM
 121 };
 122
 123 static GLuint short_types_scale[5] = {
 124    0,
 125    BRW_SURFACEFORMAT_R16_SSCALED,
 126    BRW_SURFACEFORMAT_R16G16_SSCALED,
 127    BRW_SURFACEFORMAT_R16G16B16_SSCALED,
 128    BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
 129 };
 130
 131 static GLuint ubyte_types_norm[5] = {
 132    0,
 133    BRW_SURFACEFORMAT_R8_UNORM,
 134    BRW_SURFACEFORMAT_R8G8_UNORM,
 135    BRW_SURFACEFORMAT_R8G8B8_UNORM,
 136    BRW_SURFACEFORMAT_R8G8B8A8_UNORM
 137 };
 138
 139 static GLuint ubyte_types_scale[5] = {
 140    0,
 141    BRW_SURFACEFORMAT_R8_USCALED,
 142    BRW_SURFACEFORMAT_R8G8_USCALED,
 143    BRW_SURFACEFORMAT_R8G8B8_USCALED,
 144    BRW_SURFACEFORMAT_R8G8B8A8_USCALED
 145 };
 146
 147 static GLuint byte_types_norm[5] = {
 148    0,
 149    BRW_SURFACEFORMAT_R8_SNORM,
 150    BRW_SURFACEFORMAT_R8G8_SNORM,
 151    BRW_SURFACEFORMAT_R8G8B8_SNORM,
 152    BRW_SURFACEFORMAT_R8G8B8A8_SNORM
 153 };
 154
 155 static GLuint byte_types_scale[5] = {
 156    0,
 157    BRW_SURFACEFORMAT_R8_SSCALED,
 158    BRW_SURFACEFORMAT_R8G8_SSCALED,
 159    BRW_SURFACEFORMAT_R8G8B8_SSCALED,
 160    BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
 161 };
 162
 163
 164 /**
 165  * Given vertex array type/size/format/normalized info, return
 166  * the appopriate hardware surface type.
 167  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 168  */
 169 static GLuint get_surface_type( GLenum type, GLuint size,
 170                                 GLenum format, GLboolean normalized )
 171 {
 172    if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
 173       printf("type %s size %d normalized %d\n",
 174                    _mesa_lookup_enum_by_nr(type), size, normalized);
 175
 176    if (normalized) {
 177       switch (type) {
 178       case GL_DOUBLE: return double_types[size];
 179       case GL_FLOAT: return float_types[size];
 180       case GL_HALF_FLOAT: return half_float_types[size];
 181       case GL_INT: return int_types_norm[size];
 182       case GL_SHORT: return short_types_norm[size];
 183       case GL_BYTE: return byte_types_norm[size];
 184       case GL_UNSIGNED_INT: return uint_types_norm[size];
 185       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 186       case GL_UNSIGNED_BYTE:
 187          if (format == GL_BGRA) {
 188             /* See GL_EXT_vertex_array_bgra */
 189             assert(size == 4);
 190             return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
 191          }
 192          else {
 193             return ubyte_types_norm[size];
 194          }
 195       default: assert(0); return 0;
 196       }
 197    }
 198    else {
 199       assert(format == GL_RGBA); /* sanity check */
 200       switch (type) {
 201       case GL_DOUBLE: return double_types[size];
 202       case GL_FLOAT: return float_types[size];
 203       case GL_HALF_FLOAT: return half_float_types[size];
 204       case GL_INT: return int_types_scale[size];
 205       case GL_SHORT: return short_types_scale[size];
 206       case GL_BYTE: return byte_types_scale[size];
 207       case GL_UNSIGNED_INT: return uint_types_scale[size];
 208       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 209       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 210       default: assert(0); return 0;
 211       }
 212    }
 213 }
 214
 215
 216 static GLuint get_size( GLenum type )
 217 {
 218    switch (type) {
 219    case GL_DOUBLE: return sizeof(GLdouble);
 220    case GL_FLOAT: return sizeof(GLfloat);
 221    case GL_HALF_FLOAT: return sizeof(GLhalfARB);
 222    case GL_INT: return sizeof(GLint);
 223    case GL_SHORT: return sizeof(GLshort);
 224    case GL_BYTE: return sizeof(GLbyte);
 225    case GL_UNSIGNED_INT: return sizeof(GLuint);
 226    case GL_UNSIGNED_SHORT: return sizeof(GLushort);
 227    case GL_UNSIGNED_BYTE: return sizeof(GLubyte);
 228    default: return 0;
 229    }
 230 }
 231
 232 static GLuint get_index_type(GLenum type)
 233 {
 234    switch (type) {
 235    case GL_UNSIGNED_BYTE:  return BRW_INDEX_BYTE;
 236    case GL_UNSIGNED_SHORT: return BRW_INDEX_WORD;
 237    case GL_UNSIGNED_INT:   return BRW_INDEX_DWORD;
 238    default: assert(0); return 0;
 239    }
 240 }
 241
 242 static void
 243 copy_array_to_vbo_array( struct brw_context *brw,
 244                          struct brw_vertex_element *element,
 245                          struct brw_vertex_buffer *buffer,
 246                          GLuint dst_stride)
 247 {
 248    GLuint size = element->count * dst_stride;
 249
 250    buffer->stride = dst_stride;
 251    if (dst_stride == element->glarray->StrideB) {
 252       intel_upload_data(&brw->intel, element->glarray->Ptr, size, dst_stride,
 253                         &buffer->bo, &buffer->offset);
 254    } else {
 255       const unsigned char *src = element->glarray->Ptr;
 256       char *map = intel_upload_map(&brw->intel, size, dst_stride);
 257       char *dst = map;
 258       int i;
 259
 260       for (i = 0; i < element->count; i++) {
 261          memcpy(dst, src, dst_stride);
 262          src += element->glarray->StrideB;
 263          dst += dst_stride;
 264       }
 265       intel_upload_unmap(&brw->intel, map, size, dst_stride,
 266                          &buffer->bo, &buffer->offset);
 267    }
 268 }
 269
 270 static void brw_prepare_vertices(struct brw_context *brw)
 271 {
 272    struct gl_context *ctx = &brw->intel.ctx;
 273    struct intel_context *intel = intel_context(ctx);
 274    GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
 275    const unsigned char *ptr = NULL;
 276    GLuint interleaved = 0, total_size = 0, count = -1;
 277    unsigned int min_index = brw->vb.min_index;
 278    unsigned int max_index = brw->vb.max_index;
 279    int i, j;
 280
 281    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 282    GLuint nr_uploads = 0;
 283
 284    /* First build an array of pointers to ve's in vb.inputs_read
 285     */
 286    if (0)
 287       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 288
 289    /* Accumulate the list of enabled arrays. */
 290    brw->vb.nr_enabled = 0;
 291    while (vs_inputs) {
 292       GLuint i = _mesa_ffsll(vs_inputs) - 1;
 293       struct brw_vertex_element *input = &brw->vb.inputs[i];
 294
 295       vs_inputs &= ~(1 << i);
 296       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 297    }
 298
 299    if (brw->vb.nr_enabled == 0)
 300       return;
 301
 302    if (brw->vb.nr_buffers)
 303       goto validate;
 304
 305    /* XXX: In the rare cases where this happens we fallback all
 306     * the way to software rasterization, although a tnl fallback
 307     * would be sufficient.  I don't know of *any* real world
 308     * cases with > 17 vertex attributes enabled, so it probably
 309     * isn't an issue at this point.
 310     */
 311    if (brw->vb.nr_enabled >= BRW_VEP_MAX) {
 312       intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 313       return;
 314    }
 315
 316    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 317       struct brw_vertex_element *input = brw->vb.enabled[i];
 318       const struct gl_client_array *glarray = input->glarray;
 319       int type_size = get_size(glarray->Type);
 320
 321       input->element_size = type_size * glarray->Size;
 322
 323       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 324          struct intel_buffer_object *intel_buffer =
 325             intel_buffer_object(glarray->BufferObj);
 326          int k;
 327
 328          for (k = 0; k < i; k++) {
 329             const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
 330             if (glarray->BufferObj == other->BufferObj &&
 331                 glarray->StrideB == other->StrideB &&
 332                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 333             {
 334                input->buffer = brw->vb.enabled[k]->buffer;
 335                input->offset = glarray->Ptr - other->Ptr;
 336                break;
 337             }
 338          }
 339          if (k == i) {
 340             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 341             /* Named buffer object: Just reference its contents directly. */
 342             buffer->bo = intel_bufferobj_source(intel, intel_buffer,
 343                                                 &buffer->offset);
 344             drm_intel_bo_reference(buffer->bo);
 345             buffer->offset += (uintptr_t)glarray->Ptr;
 346             buffer->stride = glarray->StrideB;
 347
 348             input->buffer = j++;
 349             input->offset = 0;
 350          }
 351          input->count = glarray->_MaxElement;
 352
 353          /* This is a common place to reach if the user mistakenly supplies
 354           * a pointer in place of a VBO offset.  If we just let it go through,
 355           * we may end up dereferencing a pointer beyond the bounds of the
 356           * GTT.  We would hope that the VBO's max_index would save us, but
 357           * Mesa appears to hand us min/max values not clipped to the
 358           * array object's _MaxElement, and _MaxElement frequently appears
 359           * to be wrong anyway.
 360           *
 361           * The VBO spec allows application termination in this case, and it's
 362           * probably a service to the poor programmer to do so rather than
 363           * trying to just not render.
 364           */
 365          assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
 366       } else {
 367          input->count = glarray->StrideB ? max_index + 1 : 1;
 368
 369          /* Queue the buffer object up to be uploaded in the next pass,
 370           * when we've decided if we're doing interleaved or not.
 371           */
 372          if (nr_uploads == 0) {
 373             /* Position array not properly enabled:
 374              */
 375             if (input->attrib == VERT_ATTRIB_POS && glarray->StrideB == 0) {
 376                intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 377                return;
 378             }
 379
 380             interleaved = glarray->StrideB;
 381             ptr = glarray->Ptr;
 382          }
 383          else if (interleaved != glarray->StrideB ||
 384                   (uintptr_t)(glarray->Ptr - ptr) > interleaved)
 385          {
 386             interleaved = 0;
 387          }
 388          else if (total_size & (type_size -1))
 389          {
 390             /* enforce natural alignment (for doubles) */
 391             interleaved = 0;
 392          }
 393
 394          if (count > input->count)
 395             count = input->count;
 396
 397          upload[nr_uploads++] = input;
 398          total_size += input->element_size;
 399       }
 400    }
 401
 402    /* Handle any arrays to be uploaded. */
 403    if (nr_uploads > 1) {
 404       if (interleaved && interleaved <= 2*total_size) {
 405          /* All uploads are interleaved, so upload the arrays together as
 406           * interleaved.  First, upload the contents and set up upload[0].
 407           */
 408          upload[0]->count = count; /* trim the upload over all arrays */
 409          copy_array_to_vbo_array(brw,
 410                                  upload[0], &brw->vb.buffers[j],
 411                                  interleaved);
 412
 413          for (i = 0; i < nr_uploads; i++) {
 414             /* Then, just point upload[i] at upload[0]'s buffer. */
 415             upload[i]->offset =
 416                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 417             upload[i]->buffer = j;
 418          }
 419          j++;
 420
 421          nr_uploads = 0;
 422       }
 423       else if (total_size < 2048) {
 424          /* Upload non-interleaved arrays into a single interleaved array */
 425          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 426          int count = upload[0]->count, offset;
 427          char *map;
 428
 429          map = intel_upload_map(&brw->intel, total_size * count, total_size);
 430          for (i = offset = 0; i < nr_uploads; i++) {
 431             const unsigned char *src = upload[i]->glarray->Ptr;
 432             int size = upload[i]->element_size;
 433             int stride = upload[i]->glarray->StrideB;
 434             char *dst = map + offset;
 435             int n;
 436
 437             for (n = 0; n < count; n++) {
 438                memcpy(dst, src, size);
 439                src += stride;
 440                dst += total_size;
 441             }
 442
 443             upload[i]->offset = offset;
 444             upload[i]->buffer = j;
 445
 446             offset += size;
 447          }
 448          intel_upload_unmap(&brw->intel, map, total_size * count, total_size,
 449                             &buffer->bo, &buffer->offset);
 450          buffer->stride = offset;
 451          j++;
 452
 453          nr_uploads = 0;
 454       }
 455    }
 456    /* Upload non-interleaved arrays */
 457    for (i = 0; i < nr_uploads; i++) {
 458       copy_array_to_vbo_array(brw,
 459                               upload[i], &brw->vb.buffers[j],
 460                               upload[i]->element_size);
 461       upload[i]->buffer = j++;
 462       upload[i]->offset = 0;
 463    }
 464
 465    /* can we simply extend the current vb? */
 466    brw->vb.start_vertex_bias = 0;
 467    if (j == brw->vb.nr_current_buffers) {
 468       int delta = 0;
 469       for (i = 0; i < j; i++) {
 470          int d;
 471
 472          if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle ||
 473              brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride)
 474             break;
 475
 476          d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset;
 477          if (delta == 0)
 478             delta = d / brw->vb.current_buffers[i].stride;
 479          if (delta * brw->vb.current_buffers[i].stride != d)
 480             break;
 481       }
 482
 483       if (i == j) {
 484          brw->vb.start_vertex_bias = delta;
 485          while (--j >= 0)
 486             drm_intel_bo_unreference(brw->vb.buffers[j].bo);
 487          j = 0;
 488       }
 489    }
 490
 491    brw->vb.nr_buffers = j;
 492
 493 validate:
 494    brw_prepare_query_begin(brw);
 495    for (i = 0; i < brw->vb.nr_buffers; i++) {
 496       brw_add_validated_bo(brw, brw->vb.buffers[i].bo);
 497    }
 498 }
 499
 500 static void brw_emit_vertices(struct brw_context *brw)
 501 {
 502    struct gl_context *ctx = &brw->intel.ctx;
 503    struct intel_context *intel = intel_context(ctx);
 504    GLuint i;
 505
 506    brw_emit_query_begin(brw);
 507
 508    /* If the VS doesn't read any inputs (calculating vertex position from
 509     * a state variable for some reason, for example), emit a single pad
 510     * VERTEX_ELEMENT struct and bail.
 511     *
 512     * The stale VB state stays in place, but they don't do anything unless
 513     * a VE loads from them.
 514     */
 515    if (brw->vb.nr_enabled == 0) {
 516       BEGIN_BATCH(3);
 517       OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
 518       if (intel->gen >= 6) {
 519          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 520                    GEN6_VE0_VALID |
 521                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 522                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 523       } else {
 524          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 525                    BRW_VE0_VALID |
 526                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 527                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 528       }
 529       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 530                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 531                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 532                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 533       CACHED_BATCH();
 534       return;
 535    }
 536
 537    /* Now emit VB and VEP state packets.
 538     */
 539
 540    if (brw->vb.nr_buffers) {
 541       BEGIN_BATCH(1 + 4*brw->vb.nr_buffers);
 542       OUT_BATCH((CMD_VERTEX_BUFFER << 16) | (4*brw->vb.nr_buffers - 1));
 543       for (i = 0; i < brw->vb.nr_buffers; i++) {
 544          struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 545          uint32_t dw0;
 546
 547          if (intel->gen >= 6) {
 548             dw0 = GEN6_VB0_ACCESS_VERTEXDATA | (i << GEN6_VB0_INDEX_SHIFT);
 549          } else {
 550             dw0 = BRW_VB0_ACCESS_VERTEXDATA | (i << BRW_VB0_INDEX_SHIFT);
 551          }
 552
 553          OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT));
 554          OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset);
 555          if (intel->gen >= 5) {
 556             OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1);
 557          } else
 558             OUT_BATCH(buffer->bo->size / buffer->stride);
 559          OUT_BATCH(0); /* Instance data step rate */
 560
 561          brw->vb.current_buffers[i].handle = buffer->bo->handle;
 562          brw->vb.current_buffers[i].offset = buffer->offset;
 563          brw->vb.current_buffers[i].stride = buffer->stride;
 564       }
 565       brw->vb.nr_current_buffers = i;
 566       ADVANCE_BATCH();
 567    }
 568
 569    BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
 570    OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | (2*brw->vb.nr_enabled - 1));
 571    for (i = 0; i < brw->vb.nr_enabled; i++) {
 572       struct brw_vertex_element *input = brw->vb.enabled[i];
 573       uint32_t format = get_surface_type(input->glarray->Type,
 574                                          input->glarray->Size,
 575                                          input->glarray->Format,
 576                                          input->glarray->Normalized);
 577       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 578       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 579       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 580       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 581
 582       switch (input->glarray->Size) {
 583       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 584       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 585       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 586       case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
 587          break;
 588       }
 589
 590       if (intel->gen >= 6) {
 591          OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
 592                    GEN6_VE0_VALID |
 593                    (format << BRW_VE0_FORMAT_SHIFT) |
 594                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 595       } else {
 596          OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) |
 597                    BRW_VE0_VALID |
 598                    (format << BRW_VE0_FORMAT_SHIFT) |
 599                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 600       }
 601
 602       if (intel->gen >= 5)
 603           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 604                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 605                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 606                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 607       else
 608           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 609                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 610                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 611                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 612                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 613    }
 614    CACHED_BATCH();
 615 }
 616
 617 const struct brw_tracked_state brw_vertices = {
 618    .dirty = {
 619       .mesa = 0,
 620       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 621       .cache = 0,
 622    },
 623    .prepare = brw_prepare_vertices,
 624    .emit = brw_emit_vertices,
 625 };
 626
 627 static void brw_prepare_indices(struct brw_context *brw)
 628 {
 629    struct gl_context *ctx = &brw->intel.ctx;
 630    struct intel_context *intel = &brw->intel;
 631    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 632    GLuint ib_size;
 633    drm_intel_bo *bo = NULL;
 634    struct gl_buffer_object *bufferobj;
 635    GLuint offset;
 636    GLuint ib_type_size;
 637
 638    if (index_buffer == NULL)
 639       return;
 640
 641    ib_type_size = get_size(index_buffer->type);
 642    ib_size = ib_type_size * index_buffer->count;
 643    bufferobj = index_buffer->obj;
 644
 645    /* Turn into a proper VBO:
 646     */
 647    if (!_mesa_is_bufferobj(bufferobj)) {
 648
 649       /* Get new bufferobj, offset:
 650        */
 651       intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size,
 652                         &bo, &offset);
 653       brw->ib.start_vertex_offset = offset / ib_type_size;
 654       offset = 0;
 655    } else {
 656       offset = (GLuint) (unsigned long) index_buffer->ptr;
 657
 658       /* If the index buffer isn't aligned to its element size, we have to
 659        * rebase it into a temporary.
 660        */
 661        if ((get_size(index_buffer->type) - 1) & offset) {
 662            GLubyte *map = ctx->Driver.MapBuffer(ctx,
 663                                                 GL_ELEMENT_ARRAY_BUFFER_ARB,
 664                                                 GL_DYNAMIC_DRAW_ARB,
 665                                                 bufferobj);
 666            map += offset;
 667
 668            intel_upload_data(&brw->intel, map, ib_size, ib_type_size,
 669                              &bo, &offset);
 670            brw->ib.start_vertex_offset = offset / ib_type_size;
 671            offset = 0;
 672
 673            ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
 674        } else {
 675           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 676            * the index buffer state when we're just moving the start index
 677            * of our drawing.
 678            */
 679           brw->ib.start_vertex_offset = offset / ib_type_size;
 680
 681           bo = intel_bufferobj_source(intel, intel_buffer_object(bufferobj),
 682                                       &offset);
 683           drm_intel_bo_reference(bo);
 684        }
 685    }
 686
 687    if (brw->ib.bo != bo || brw->ib.offset != offset) {
 688       drm_intel_bo_unreference(brw->ib.bo);
 689       brw->ib.bo = bo;
 690       brw->ib.offset = offset;
 691
 692       brw_add_validated_bo(brw, brw->ib.bo);
 693       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 694    } else {
 695       drm_intel_bo_unreference(bo);
 696    }
 697 }
 698
 699 const struct brw_tracked_state brw_indices = {
 700    .dirty = {
 701       .mesa = 0,
 702       .brw = BRW_NEW_INDICES,
 703       .cache = 0,
 704    },
 705    .prepare = brw_prepare_indices,
 706 };
 707
 708 static void brw_emit_index_buffer(struct brw_context *brw)
 709 {
 710    struct intel_context *intel = &brw->intel;
 711    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 712
 713    if (index_buffer == NULL)
 714       return;
 715
 716    BEGIN_BATCH(3);
 717    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 718              /* cut index enable << 10 */
 719              get_index_type(index_buffer->type) << 8 |
 720              1);
 721    OUT_RELOC(brw->ib.bo,
 722              I915_GEM_DOMAIN_VERTEX, 0,
 723              brw->ib.offset);
 724    OUT_RELOC(brw->ib.bo,
 725              I915_GEM_DOMAIN_VERTEX, 0,
 726              brw->ib.bo->size - 1);
 727    ADVANCE_BATCH();
 728 }
 729
 730 const struct brw_tracked_state brw_index_buffer = {
 731    .dirty = {
 732       .mesa = 0,
 733       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 734       .cache = 0,
 735    },
 736    .emit = brw_emit_index_buffer,
 737 };