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
  29 #include "main/glheader.h"
  30 #include "main/bufferobj.h"
  31 #include "main/context.h"
  32 #include "main/state.h"
  33 #include "main/api_validate.h"
  34 #include "main/enums.h"
  35
  36 #include "brw_draw.h"
  37 #include "brw_defines.h"
  38 #include "brw_context.h"
  39 #include "brw_state.h"
  40 #include "brw_fallback.h"
  41
  42 #include "intel_batchbuffer.h"
  43 #include "intel_buffer_objects.h"
  44 #include "intel_tex.h"
  45
  46 static GLuint double_types[5] = {
  47    0,
  48    BRW_SURFACEFORMAT_R64_FLOAT,
  49    BRW_SURFACEFORMAT_R64G64_FLOAT,
  50    BRW_SURFACEFORMAT_R64G64B64_FLOAT,
  51    BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
  52 };
  53
  54 static GLuint float_types[5] = {
  55    0,
  56    BRW_SURFACEFORMAT_R32_FLOAT,
  57    BRW_SURFACEFORMAT_R32G32_FLOAT,
  58    BRW_SURFACEFORMAT_R32G32B32_FLOAT,
  59    BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
  60 };
  61
  62 static GLuint half_float_types[5] = {
  63    0,
  64    BRW_SURFACEFORMAT_R16_FLOAT,
  65    BRW_SURFACEFORMAT_R16G16_FLOAT,
  66    0, /* can't seem to render this one */
  67    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
  68 };
  69
  70 static GLuint uint_types_norm[5] = {
  71    0,
  72    BRW_SURFACEFORMAT_R32_UNORM,
  73    BRW_SURFACEFORMAT_R32G32_UNORM,
  74    BRW_SURFACEFORMAT_R32G32B32_UNORM,
  75    BRW_SURFACEFORMAT_R32G32B32A32_UNORM
  76 };
  77
  78 static GLuint uint_types_scale[5] = {
  79    0,
  80    BRW_SURFACEFORMAT_R32_USCALED,
  81    BRW_SURFACEFORMAT_R32G32_USCALED,
  82    BRW_SURFACEFORMAT_R32G32B32_USCALED,
  83    BRW_SURFACEFORMAT_R32G32B32A32_USCALED
  84 };
  85
  86 static GLuint int_types_norm[5] = {
  87    0,
  88    BRW_SURFACEFORMAT_R32_SNORM,
  89    BRW_SURFACEFORMAT_R32G32_SNORM,
  90    BRW_SURFACEFORMAT_R32G32B32_SNORM,
  91    BRW_SURFACEFORMAT_R32G32B32A32_SNORM
  92 };
  93
  94 static GLuint int_types_scale[5] = {
  95    0,
  96    BRW_SURFACEFORMAT_R32_SSCALED,
  97    BRW_SURFACEFORMAT_R32G32_SSCALED,
  98    BRW_SURFACEFORMAT_R32G32B32_SSCALED,
  99    BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
 100 };
 101
 102 static GLuint ushort_types_norm[5] = {
 103    0,
 104    BRW_SURFACEFORMAT_R16_UNORM,
 105    BRW_SURFACEFORMAT_R16G16_UNORM,
 106    BRW_SURFACEFORMAT_R16G16B16_UNORM,
 107    BRW_SURFACEFORMAT_R16G16B16A16_UNORM
 108 };
 109
 110 static GLuint ushort_types_scale[5] = {
 111    0,
 112    BRW_SURFACEFORMAT_R16_USCALED,
 113    BRW_SURFACEFORMAT_R16G16_USCALED,
 114    BRW_SURFACEFORMAT_R16G16B16_USCALED,
 115    BRW_SURFACEFORMAT_R16G16B16A16_USCALED
 116 };
 117
 118 static GLuint short_types_norm[5] = {
 119    0,
 120    BRW_SURFACEFORMAT_R16_SNORM,
 121    BRW_SURFACEFORMAT_R16G16_SNORM,
 122    BRW_SURFACEFORMAT_R16G16B16_SNORM,
 123    BRW_SURFACEFORMAT_R16G16B16A16_SNORM
 124 };
 125
 126 static GLuint short_types_scale[5] = {
 127    0,
 128    BRW_SURFACEFORMAT_R16_SSCALED,
 129    BRW_SURFACEFORMAT_R16G16_SSCALED,
 130    BRW_SURFACEFORMAT_R16G16B16_SSCALED,
 131    BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
 132 };
 133
 134 static GLuint ubyte_types_norm[5] = {
 135    0,
 136    BRW_SURFACEFORMAT_R8_UNORM,
 137    BRW_SURFACEFORMAT_R8G8_UNORM,
 138    BRW_SURFACEFORMAT_R8G8B8_UNORM,
 139    BRW_SURFACEFORMAT_R8G8B8A8_UNORM
 140 };
 141
 142 static GLuint ubyte_types_scale[5] = {
 143    0,
 144    BRW_SURFACEFORMAT_R8_USCALED,
 145    BRW_SURFACEFORMAT_R8G8_USCALED,
 146    BRW_SURFACEFORMAT_R8G8B8_USCALED,
 147    BRW_SURFACEFORMAT_R8G8B8A8_USCALED
 148 };
 149
 150 static GLuint byte_types_norm[5] = {
 151    0,
 152    BRW_SURFACEFORMAT_R8_SNORM,
 153    BRW_SURFACEFORMAT_R8G8_SNORM,
 154    BRW_SURFACEFORMAT_R8G8B8_SNORM,
 155    BRW_SURFACEFORMAT_R8G8B8A8_SNORM
 156 };
 157
 158 static GLuint byte_types_scale[5] = {
 159    0,
 160    BRW_SURFACEFORMAT_R8_SSCALED,
 161    BRW_SURFACEFORMAT_R8G8_SSCALED,
 162    BRW_SURFACEFORMAT_R8G8B8_SSCALED,
 163    BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
 164 };
 165
 166
 167 /**
 168  * Given vertex array type/size/format/normalized info, return
 169  * the appopriate hardware surface type.
 170  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 171  */
 172 static GLuint get_surface_type( GLenum type, GLuint size,
 173                                 GLenum format, GLboolean normalized )
 174 {
 175    if (INTEL_DEBUG & DEBUG_VERTS)
 176       _mesa_printf("type %s size %d normalized %d\n",
 177                    _mesa_lookup_enum_by_nr(type), size, normalized);
 178
 179    if (normalized) {
 180       switch (type) {
 181       case GL_DOUBLE: return double_types[size];
 182       case GL_FLOAT: return float_types[size];
 183       case GL_HALF_FLOAT: return half_float_types[size];
 184       case GL_INT: return int_types_norm[size];
 185       case GL_SHORT: return short_types_norm[size];
 186       case GL_BYTE: return byte_types_norm[size];
 187       case GL_UNSIGNED_INT: return uint_types_norm[size];
 188       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 189       case GL_UNSIGNED_BYTE:
 190          if (format == GL_BGRA) {
 191             /* See GL_EXT_vertex_array_bgra */
 192             assert(size == 4);
 193             return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
 194          }
 195          else {
 196             return ubyte_types_norm[size];
 197          }
 198       default: assert(0); return 0;
 199       }
 200    }
 201    else {
 202       assert(format == GL_RGBA); /* sanity check */
 203       switch (type) {
 204       case GL_DOUBLE: return double_types[size];
 205       case GL_FLOAT: return float_types[size];
 206       case GL_HALF_FLOAT: return half_float_types[size];
 207       case GL_INT: return int_types_scale[size];
 208       case GL_SHORT: return short_types_scale[size];
 209       case GL_BYTE: return byte_types_scale[size];
 210       case GL_UNSIGNED_INT: return uint_types_scale[size];
 211       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 212       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 213       default: assert(0); return 0;
 214       }
 215    }
 216 }
 217
 218
 219 static GLuint get_size( GLenum type )
 220 {
 221    switch (type) {
 222    case GL_DOUBLE: return sizeof(GLdouble);
 223    case GL_FLOAT: return sizeof(GLfloat);
 224    case GL_HALF_FLOAT: return sizeof(GLhalfARB);
 225    case GL_INT: return sizeof(GLint);
 226    case GL_SHORT: return sizeof(GLshort);
 227    case GL_BYTE: return sizeof(GLbyte);
 228    case GL_UNSIGNED_INT: return sizeof(GLuint);
 229    case GL_UNSIGNED_SHORT: return sizeof(GLushort);
 230    case GL_UNSIGNED_BYTE: return sizeof(GLubyte);
 231    default: return 0;
 232    }
 233 }
 234
 235 static GLuint get_index_type(GLenum type)
 236 {
 237    switch (type) {
 238    case GL_UNSIGNED_BYTE:  return BRW_INDEX_BYTE;
 239    case GL_UNSIGNED_SHORT: return BRW_INDEX_WORD;
 240    case GL_UNSIGNED_INT:   return BRW_INDEX_DWORD;
 241    default: assert(0); return 0;
 242    }
 243 }
 244
 245 static void wrap_buffers( struct brw_context *brw,
 246                           GLuint size )
 247 {
 248    if (size < BRW_UPLOAD_INIT_SIZE)
 249       size = BRW_UPLOAD_INIT_SIZE;
 250
 251    brw->vb.upload.offset = 0;
 252
 253    if (brw->vb.upload.bo != NULL)
 254       dri_bo_unreference(brw->vb.upload.bo);
 255    brw->vb.upload.bo = dri_bo_alloc(brw->intel.bufmgr, "temporary VBO",
 256                                     size, 1);
 257 }
 258
 259 static void get_space( struct brw_context *brw,
 260                        GLuint size,
 261                        dri_bo **bo_return,
 262                        GLuint *offset_return )
 263 {
 264    size = ALIGN(size, 64);
 265
 266    if (brw->vb.upload.bo == NULL ||
 267        brw->vb.upload.offset + size > brw->vb.upload.bo->size) {
 268       wrap_buffers(brw, size);
 269    }
 270
 271    assert(*bo_return == NULL);
 272    dri_bo_reference(brw->vb.upload.bo);
 273    *bo_return = brw->vb.upload.bo;
 274    *offset_return = brw->vb.upload.offset;
 275    brw->vb.upload.offset += size;
 276 }
 277
 278 static void
 279 copy_array_to_vbo_array( struct brw_context *brw,
 280                          struct brw_vertex_element *element,
 281                          GLuint dst_stride)
 282 {
 283    struct intel_context *intel = &brw->intel;
 284    GLuint size = element->count * dst_stride;
 285
 286    get_space(brw, size, &element->bo, &element->offset);
 287
 288    if (element->glarray->StrideB == 0) {
 289       assert(element->count == 1);
 290       element->stride = 0;
 291    } else {
 292       element->stride = dst_stride;
 293    }
 294
 295    if (dst_stride == element->glarray->StrideB) {
 296       if (intel->intelScreen->kernel_exec_fencing) {
 297          drm_intel_gem_bo_map_gtt(element->bo);
 298          memcpy((char *)element->bo->virtual + element->offset,
 299                 element->glarray->Ptr, size);
 300          drm_intel_gem_bo_unmap_gtt(element->bo);
 301       } else {
 302          dri_bo_subdata(element->bo,
 303                         element->offset,
 304                         size,
 305                         element->glarray->Ptr);
 306       }
 307    } else {
 308       char *dest;
 309       const unsigned char *src = element->glarray->Ptr;
 310       int i;
 311
 312       if (intel->intelScreen->kernel_exec_fencing) {
 313          drm_intel_gem_bo_map_gtt(element->bo);
 314          dest = element->bo->virtual;
 315          dest += element->offset;
 316
 317          for (i = 0; i < element->count; i++) {
 318             memcpy(dest, src, dst_stride);
 319             src += element->glarray->StrideB;
 320             dest += dst_stride;
 321          }
 322
 323          drm_intel_gem_bo_unmap_gtt(element->bo);
 324       } else {
 325          void *data;
 326
 327          data = _mesa_malloc(dst_stride * element->count);
 328          dest = data;
 329          for (i = 0; i < element->count; i++) {
 330             memcpy(dest, src, dst_stride);
 331             src += element->glarray->StrideB;
 332             dest += dst_stride;
 333          }
 334
 335          dri_bo_subdata(element->bo,
 336                         element->offset,
 337                         size,
 338                         data);
 339
 340          _mesa_free(data);
 341       }
 342    }
 343 }
 344
 345 static void brw_prepare_vertices(struct brw_context *brw)
 346 {
 347    GLcontext *ctx = &brw->intel.ctx;
 348    struct intel_context *intel = intel_context(ctx);
 349    GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
 350    GLuint i;
 351    const unsigned char *ptr = NULL;
 352    GLuint interleave = 0;
 353    unsigned int min_index = brw->vb.min_index;
 354    unsigned int max_index = brw->vb.max_index;
 355
 356    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 357    GLuint nr_uploads = 0;
 358
 359    /* First build an array of pointers to ve's in vb.inputs_read
 360     */
 361    if (0)
 362       _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 363
 364    /* Accumulate the list of enabled arrays. */
 365    brw->vb.nr_enabled = 0;
 366    while (vs_inputs) {
 367       GLuint i = _mesa_ffsll(vs_inputs) - 1;
 368       struct brw_vertex_element *input = &brw->vb.inputs[i];
 369
 370       vs_inputs &= ~(1 << i);
 371       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 372    }
 373
 374    /* XXX: In the rare cases where this happens we fallback all
 375     * the way to software rasterization, although a tnl fallback
 376     * would be sufficient.  I don't know of *any* real world
 377     * cases with > 17 vertex attributes enabled, so it probably
 378     * isn't an issue at this point.
 379     */
 380    if (brw->vb.nr_enabled >= BRW_VEP_MAX) {
 381       intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 382       return;
 383    }
 384
 385    for (i = 0; i < brw->vb.nr_enabled; i++) {
 386       struct brw_vertex_element *input = brw->vb.enabled[i];
 387
 388       input->element_size = get_size(input->glarray->Type) * input->glarray->Size;
 389
 390       if (_mesa_is_bufferobj(input->glarray->BufferObj)) {
 391          struct intel_buffer_object *intel_buffer =
 392             intel_buffer_object(input->glarray->BufferObj);
 393
 394          /* Named buffer object: Just reference its contents directly. */
 395          dri_bo_unreference(input->bo);
 396          input->bo = intel_bufferobj_buffer(intel, intel_buffer,
 397                                             INTEL_READ);
 398          dri_bo_reference(input->bo);
 399          input->offset = (unsigned long)input->glarray->Ptr;
 400          input->stride = input->glarray->StrideB;
 401          input->count = input->glarray->_MaxElement;
 402
 403          /* This is a common place to reach if the user mistakenly supplies
 404           * a pointer in place of a VBO offset.  If we just let it go through,
 405           * we may end up dereferencing a pointer beyond the bounds of the
 406           * GTT.  We would hope that the VBO's max_index would save us, but
 407           * Mesa appears to hand us min/max values not clipped to the
 408           * array object's _MaxElement, and _MaxElement frequently appears
 409           * to be wrong anyway.
 410           *
 411           * The VBO spec allows application termination in this case, and it's
 412           * probably a service to the poor programmer to do so rather than
 413           * trying to just not render.
 414           */
 415          assert(input->offset < input->bo->size);
 416       } else {
 417          input->count = input->glarray->StrideB ? max_index + 1 - min_index : 1;
 418          if (input->bo != NULL) {
 419             /* Already-uploaded vertex data is present from a previous
 420              * prepare_vertices, but we had to re-validate state due to
 421              * check_aperture failing and a new batch being produced.
 422              */
 423             continue;
 424          }
 425
 426          /* Queue the buffer object up to be uploaded in the next pass,
 427           * when we've decided if we're doing interleaved or not.
 428           */
 429          if (input->attrib == VERT_ATTRIB_POS) {
 430             /* Position array not properly enabled:
 431              */
 432             if (input->glarray->StrideB == 0) {
 433                intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 434                return;
 435             }
 436
 437             interleave = input->glarray->StrideB;
 438             ptr = input->glarray->Ptr;
 439          }
 440          else if (interleave != input->glarray->StrideB ||
 441                   (const unsigned char *)input->glarray->Ptr - ptr < 0 ||
 442                   (const unsigned char *)input->glarray->Ptr - ptr > interleave)
 443          {
 444             interleave = 0;
 445          }
 446
 447          upload[nr_uploads++] = input;
 448
 449          /* We rebase drawing to start at element zero only when
 450           * varyings are not in vbos, which means we can end up
 451           * uploading non-varying arrays (stride != 0) when min_index
 452           * is zero.  This doesn't matter as the amount to upload is
 453           * the same for these arrays whether the draw call is rebased
 454           * or not - we just have to upload the one element.
 455           */
 456          assert(min_index == 0 || input->glarray->StrideB == 0);
 457       }
 458    }
 459
 460    /* Handle any arrays to be uploaded. */
 461    if (nr_uploads > 1 && interleave && interleave <= 256) {
 462       /* All uploads are interleaved, so upload the arrays together as
 463        * interleaved.  First, upload the contents and set up upload[0].
 464        */
 465       copy_array_to_vbo_array(brw, upload[0], interleave);
 466
 467       for (i = 1; i < nr_uploads; i++) {
 468          /* Then, just point upload[i] at upload[0]'s buffer. */
 469          upload[i]->stride = interleave;
 470          upload[i]->offset = upload[0]->offset +
 471             ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 472          upload[i]->bo = upload[0]->bo;
 473          dri_bo_reference(upload[i]->bo);
 474       }
 475    }
 476    else {
 477       /* Upload non-interleaved arrays */
 478       for (i = 0; i < nr_uploads; i++) {
 479           copy_array_to_vbo_array(brw, upload[i], upload[i]->element_size);
 480       }
 481    }
 482
 483    brw_prepare_query_begin(brw);
 484
 485    for (i = 0; i < brw->vb.nr_enabled; i++) {
 486       struct brw_vertex_element *input = brw->vb.enabled[i];
 487
 488       brw_add_validated_bo(brw, input->bo);
 489    }
 490 }
 491
 492 static void brw_emit_vertices(struct brw_context *brw)
 493 {
 494    GLcontext *ctx = &brw->intel.ctx;
 495    struct intel_context *intel = intel_context(ctx);
 496    GLuint i;
 497
 498    brw_emit_query_begin(brw);
 499
 500    /* If the VS doesn't read any inputs (calculating vertex position from
 501     * a state variable for some reason, for example), emit a single pad
 502     * VERTEX_ELEMENT struct and bail.
 503     *
 504     * The stale VB state stays in place, but they don't do anything unless
 505     * a VE loads from them.
 506     */
 507    if (brw->vb.nr_enabled == 0) {
 508       BEGIN_BATCH(3);
 509       OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
 510       OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 511                 BRW_VE0_VALID |
 512                 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 513                 (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 514       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 515                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 516                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 517                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 518       ADVANCE_BATCH();
 519       return;
 520    }
 521
 522    /* Now emit VB and VEP state packets.
 523     *
 524     * This still defines a hardware VB for each input, even if they
 525     * are interleaved or from the same VBO.  TBD if this makes a
 526     * performance difference.
 527     */
 528    BEGIN_BATCH(1 + brw->vb.nr_enabled * 4);
 529    OUT_BATCH((CMD_VERTEX_BUFFER << 16) |
 530              ((1 + brw->vb.nr_enabled * 4) - 2));
 531
 532    for (i = 0; i < brw->vb.nr_enabled; i++) {
 533       struct brw_vertex_element *input = brw->vb.enabled[i];
 534
 535       OUT_BATCH((i << BRW_VB0_INDEX_SHIFT) |
 536                 BRW_VB0_ACCESS_VERTEXDATA |
 537                 (input->stride << BRW_VB0_PITCH_SHIFT));
 538       OUT_RELOC(input->bo,
 539                 I915_GEM_DOMAIN_VERTEX, 0,
 540                 input->offset);
 541       if (intel->is_ironlake) {
 542          OUT_RELOC(input->bo,
 543                    I915_GEM_DOMAIN_VERTEX, 0,
 544                    input->bo->size - 1);
 545       } else
 546           OUT_BATCH(input->stride ? input->count : 0);
 547       OUT_BATCH(0); /* Instance data step rate */
 548    }
 549    ADVANCE_BATCH();
 550
 551    BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
 552    OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_enabled * 2) - 2));
 553    for (i = 0; i < brw->vb.nr_enabled; i++) {
 554       struct brw_vertex_element *input = brw->vb.enabled[i];
 555       uint32_t format = get_surface_type(input->glarray->Type,
 556                                          input->glarray->Size,
 557                                          input->glarray->Format,
 558                                          input->glarray->Normalized);
 559       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 560       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 561       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 562       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 563
 564       switch (input->glarray->Size) {
 565       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 566       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 567       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 568       case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
 569          break;
 570       }
 571
 572       OUT_BATCH((i << BRW_VE0_INDEX_SHIFT) |
 573                 BRW_VE0_VALID |
 574                 (format << BRW_VE0_FORMAT_SHIFT) |
 575                 (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 576
 577       if (intel->is_ironlake)
 578           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 579                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 580                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 581                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 582       else
 583           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 584                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 585                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 586                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 587                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 588    }
 589    ADVANCE_BATCH();
 590 }
 591
 592 const struct brw_tracked_state brw_vertices = {
 593    .dirty = {
 594       .mesa = 0,
 595       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 596       .cache = 0,
 597    },
 598    .prepare = brw_prepare_vertices,
 599    .emit = brw_emit_vertices,
 600 };
 601
 602 static void brw_prepare_indices(struct brw_context *brw)
 603 {
 604    GLcontext *ctx = &brw->intel.ctx;
 605    struct intel_context *intel = &brw->intel;
 606    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 607    GLuint ib_size;
 608    dri_bo *bo = NULL;
 609    struct gl_buffer_object *bufferobj;
 610    GLuint offset;
 611    GLuint ib_type_size;
 612
 613    if (index_buffer == NULL)
 614       return;
 615
 616    ib_type_size = get_size(index_buffer->type);
 617    ib_size = ib_type_size * index_buffer->count;
 618    bufferobj = index_buffer->obj;;
 619
 620    /* Turn into a proper VBO:
 621     */
 622    if (!_mesa_is_bufferobj(bufferobj)) {
 623       brw->ib.start_vertex_offset = 0;
 624
 625       /* Get new bufferobj, offset:
 626        */
 627       get_space(brw, ib_size, &bo, &offset);
 628
 629       /* Straight upload
 630        */
 631       if (intel->intelScreen->kernel_exec_fencing) {
 632          drm_intel_gem_bo_map_gtt(bo);
 633          memcpy((char *)bo->virtual + offset, index_buffer->ptr, ib_size);
 634          drm_intel_gem_bo_unmap_gtt(bo);
 635       } else {
 636          dri_bo_subdata(bo, offset, ib_size, index_buffer->ptr);
 637       }
 638    } else {
 639       offset = (GLuint) (unsigned long) index_buffer->ptr;
 640       brw->ib.start_vertex_offset = 0;
 641
 642       /* If the index buffer isn't aligned to its element size, we have to
 643        * rebase it into a temporary.
 644        */
 645        if ((get_size(index_buffer->type) - 1) & offset) {
 646            GLubyte *map = ctx->Driver.MapBuffer(ctx,
 647                                                 GL_ELEMENT_ARRAY_BUFFER_ARB,
 648                                                 GL_DYNAMIC_DRAW_ARB,
 649                                                 bufferobj);
 650            map += offset;
 651
 652            get_space(brw, ib_size, &bo, &offset);
 653
 654            dri_bo_subdata(bo, offset, ib_size, map);
 655
 656            ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
 657        } else {
 658           bo = intel_bufferobj_buffer(intel, intel_buffer_object(bufferobj),
 659                                       INTEL_READ);
 660           dri_bo_reference(bo);
 661
 662           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 663            * the index buffer state when we're just moving the start index
 664            * of our drawing.
 665            */
 666           brw->ib.start_vertex_offset = offset / ib_type_size;
 667           offset = 0;
 668           ib_size = bo->size;
 669        }
 670    }
 671
 672    if (brw->ib.bo != bo ||
 673        brw->ib.offset != offset ||
 674        brw->ib.size != ib_size)
 675    {
 676       drm_intel_bo_unreference(brw->ib.bo);
 677       brw->ib.bo = bo;
 678       brw->ib.offset = offset;
 679       brw->ib.size = ib_size;
 680
 681       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 682    } else {
 683       drm_intel_bo_unreference(bo);
 684    }
 685
 686    brw_add_validated_bo(brw, brw->ib.bo);
 687 }
 688
 689 const struct brw_tracked_state brw_indices = {
 690    .dirty = {
 691       .mesa = 0,
 692       .brw = BRW_NEW_INDICES,
 693       .cache = 0,
 694    },
 695    .prepare = brw_prepare_indices,
 696 };
 697
 698 static void brw_emit_index_buffer(struct brw_context *brw)
 699 {
 700    struct intel_context *intel = &brw->intel;
 701    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 702
 703    if (index_buffer == NULL)
 704       return;
 705
 706    /* Emit the indexbuffer packet:
 707     */
 708    {
 709       struct brw_indexbuffer ib;
 710
 711       memset(&ib, 0, sizeof(ib));
 712
 713       ib.header.bits.opcode = CMD_INDEX_BUFFER;
 714       ib.header.bits.length = sizeof(ib)/4 - 2;
 715       ib.header.bits.index_format = get_index_type(index_buffer->type);
 716       ib.header.bits.cut_index_enable = 0;
 717
 718       BEGIN_BATCH(4);
 719       OUT_BATCH( ib.header.dword );
 720       OUT_RELOC(brw->ib.bo,
 721                 I915_GEM_DOMAIN_VERTEX, 0,
 722                 brw->ib.offset);
 723       OUT_RELOC(brw->ib.bo,
 724                 I915_GEM_DOMAIN_VERTEX, 0,
 725                 brw->ib.offset + brw->ib.size - 1);
 726       OUT_BATCH( 0 );
 727       ADVANCE_BATCH();
 728    }
 729 }
 730
 731 const struct brw_tracked_state brw_index_buffer = {
 732    .dirty = {
 733       .mesa = 0,
 734       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 735       .cache = 0,
 736    },
 737    .emit = brw_emit_index_buffer,
 738 };