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                          GLuint dst_stride)
 246 {
 247    GLuint size = element->count * dst_stride;
 248
 249    if (element->glarray->StrideB == 0) {
 250       assert(element->count == 1);
 251       element->stride = 0;
 252    } else {
 253       element->stride = dst_stride;
 254    }
 255
 256    if (dst_stride == element->glarray->StrideB) {
 257       intel_upload_data(&brw->intel, element->glarray->Ptr, size,
 258                         &element->bo, &element->offset);
 259    } else {
 260       const unsigned char *src = element->glarray->Ptr;
 261       char *dst = intel_upload_map(&brw->intel, size,
 262                                    &element->bo, &element->offset);
 263       int i;
 264
 265       for (i = 0; i < element->count; i++) {
 266          memcpy(dst, src, dst_stride);
 267          src += element->glarray->StrideB;
 268          dst += dst_stride;
 269       }
 270    }
 271 }
 272
 273 static void brw_prepare_vertices(struct brw_context *brw)
 274 {
 275    struct gl_context *ctx = &brw->intel.ctx;
 276    struct intel_context *intel = intel_context(ctx);
 277    GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
 278    GLuint i;
 279    const unsigned char *ptr = NULL;
 280    GLuint interleave = 0;
 281    unsigned int min_index = brw->vb.min_index;
 282    unsigned int max_index = brw->vb.max_index;
 283
 284    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 285    GLuint nr_uploads = 0;
 286
 287    /* First build an array of pointers to ve's in vb.inputs_read
 288     */
 289    if (0)
 290       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 291
 292    /* Accumulate the list of enabled arrays. */
 293    brw->vb.nr_enabled = 0;
 294    while (vs_inputs) {
 295       GLuint i = _mesa_ffsll(vs_inputs) - 1;
 296       struct brw_vertex_element *input = &brw->vb.inputs[i];
 297
 298       vs_inputs &= ~(1 << i);
 299       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 300    }
 301
 302    if (brw->vb.nr_enabled == 0)
 303       return;
 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 = 0; i < brw->vb.nr_enabled; i++) {
 317       struct brw_vertex_element *input = brw->vb.enabled[i];
 318
 319       input->element_size = get_size(input->glarray->Type) * input->glarray->Size;
 320
 321       if (_mesa_is_bufferobj(input->glarray->BufferObj)) {
 322          struct intel_buffer_object *intel_buffer =
 323             intel_buffer_object(input->glarray->BufferObj);
 324          GLuint offset;
 325
 326          /* Named buffer object: Just reference its contents directly. */
 327          drm_intel_bo_unreference(input->bo);
 328          input->bo = intel_bufferobj_source(intel, intel_buffer, &offset);
 329          drm_intel_bo_reference(input->bo);
 330          input->offset = offset + (unsigned long)input->glarray->Ptr;
 331          input->stride = input->glarray->StrideB;
 332          input->count = input->glarray->_MaxElement;
 333
 334          /* This is a common place to reach if the user mistakenly supplies
 335           * a pointer in place of a VBO offset.  If we just let it go through,
 336           * we may end up dereferencing a pointer beyond the bounds of the
 337           * GTT.  We would hope that the VBO's max_index would save us, but
 338           * Mesa appears to hand us min/max values not clipped to the
 339           * array object's _MaxElement, and _MaxElement frequently appears
 340           * to be wrong anyway.
 341           *
 342           * The VBO spec allows application termination in this case, and it's
 343           * probably a service to the poor programmer to do so rather than
 344           * trying to just not render.
 345           */
 346          assert(input->offset < input->bo->size);
 347       } else {
 348          input->count = input->glarray->StrideB ? max_index + 1 : 1;
 349          if (input->bo != NULL) {
 350             /* Already-uploaded vertex data is present from a previous
 351              * prepare_vertices, but we had to re-validate state due to
 352              * check_aperture failing and a new batch being produced.
 353              */
 354             continue;
 355          }
 356
 357          /* Queue the buffer object up to be uploaded in the next pass,
 358           * when we've decided if we're doing interleaved or not.
 359           */
 360          if (input->attrib == VERT_ATTRIB_POS) {
 361             /* Position array not properly enabled:
 362              */
 363             if (input->glarray->StrideB == 0) {
 364                intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 365                return;
 366             }
 367
 368             interleave = input->glarray->StrideB;
 369             ptr = input->glarray->Ptr;
 370          }
 371          else if (interleave != input->glarray->StrideB ||
 372                   (const unsigned char *)input->glarray->Ptr - ptr < 0 ||
 373                   (const unsigned char *)input->glarray->Ptr - ptr > interleave)
 374          {
 375             interleave = 0;
 376          }
 377
 378          upload[nr_uploads++] = input;
 379       }
 380    }
 381
 382    /* Handle any arrays to be uploaded. */
 383    if (nr_uploads > 1 && interleave && interleave <= 256) {
 384       /* All uploads are interleaved, so upload the arrays together as
 385        * interleaved.  First, upload the contents and set up upload[0].
 386        */
 387       copy_array_to_vbo_array(brw, upload[0], interleave);
 388
 389       for (i = 1; i < nr_uploads; i++) {
 390          /* Then, just point upload[i] at upload[0]'s buffer. */
 391          upload[i]->stride = interleave;
 392          upload[i]->offset = upload[0]->offset +
 393             ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 394          upload[i]->bo = upload[0]->bo;
 395          drm_intel_bo_reference(upload[i]->bo);
 396       }
 397    }
 398    else {
 399       /* Upload non-interleaved arrays */
 400       for (i = 0; i < nr_uploads; i++) {
 401           copy_array_to_vbo_array(brw, upload[i], upload[i]->element_size);
 402       }
 403    }
 404
 405    brw_prepare_query_begin(brw);
 406
 407    for (i = 0; i < brw->vb.nr_enabled; i++) {
 408       struct brw_vertex_element *input = brw->vb.enabled[i];
 409
 410       brw_add_validated_bo(brw, input->bo);
 411    }
 412 }
 413
 414 static void brw_emit_vertices(struct brw_context *brw)
 415 {
 416    struct gl_context *ctx = &brw->intel.ctx;
 417    struct intel_context *intel = intel_context(ctx);
 418    GLuint i;
 419
 420    brw_emit_query_begin(brw);
 421
 422    /* If the VS doesn't read any inputs (calculating vertex position from
 423     * a state variable for some reason, for example), emit a single pad
 424     * VERTEX_ELEMENT struct and bail.
 425     *
 426     * The stale VB state stays in place, but they don't do anything unless
 427     * a VE loads from them.
 428     */
 429    if (brw->vb.nr_enabled == 0) {
 430       BEGIN_BATCH(3);
 431       OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
 432       if (intel->gen >= 6) {
 433          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 434                    GEN6_VE0_VALID |
 435                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 436                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 437       } else {
 438          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 439                    BRW_VE0_VALID |
 440                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 441                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 442       }
 443       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 444                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 445                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 446                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 447       CACHED_BATCH();
 448       return;
 449    }
 450
 451    /* Now emit VB and VEP state packets.
 452     *
 453     * This still defines a hardware VB for each input, even if they
 454     * are interleaved or from the same VBO.  TBD if this makes a
 455     * performance difference.
 456     */
 457    BEGIN_BATCH(1 + brw->vb.nr_enabled * 4);
 458    OUT_BATCH((CMD_VERTEX_BUFFER << 16) |
 459              ((1 + brw->vb.nr_enabled * 4) - 2));
 460
 461    for (i = 0; i < brw->vb.nr_enabled; i++) {
 462       struct brw_vertex_element *input = brw->vb.enabled[i];
 463       uint32_t dw0;
 464
 465       if (intel->gen >= 6) {
 466          dw0 = GEN6_VB0_ACCESS_VERTEXDATA |
 467             (i << GEN6_VB0_INDEX_SHIFT);
 468       } else {
 469          dw0 = BRW_VB0_ACCESS_VERTEXDATA |
 470             (i << BRW_VB0_INDEX_SHIFT);
 471       }
 472
 473       OUT_BATCH(dw0 |
 474                 (input->stride << BRW_VB0_PITCH_SHIFT));
 475       OUT_RELOC(input->bo,
 476                 I915_GEM_DOMAIN_VERTEX, 0,
 477                 input->offset);
 478       if (intel->gen >= 5) {
 479          OUT_RELOC(input->bo,
 480                    I915_GEM_DOMAIN_VERTEX, 0,
 481                    input->bo->size - 1);
 482       } else
 483           OUT_BATCH(input->stride ? input->count : 0);
 484       OUT_BATCH(0); /* Instance data step rate */
 485    }
 486    ADVANCE_BATCH();
 487
 488    BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
 489    OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_enabled * 2) - 2));
 490    for (i = 0; i < brw->vb.nr_enabled; i++) {
 491       struct brw_vertex_element *input = brw->vb.enabled[i];
 492       uint32_t format = get_surface_type(input->glarray->Type,
 493                                          input->glarray->Size,
 494                                          input->glarray->Format,
 495                                          input->glarray->Normalized);
 496       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 497       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 498       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 499       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 500
 501       switch (input->glarray->Size) {
 502       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 503       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 504       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 505       case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
 506          break;
 507       }
 508
 509       if (intel->gen >= 6) {
 510          OUT_BATCH((i << GEN6_VE0_INDEX_SHIFT) |
 511                    GEN6_VE0_VALID |
 512                    (format << BRW_VE0_FORMAT_SHIFT) |
 513                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 514       } else {
 515          OUT_BATCH((i << BRW_VE0_INDEX_SHIFT) |
 516                    BRW_VE0_VALID |
 517                    (format << BRW_VE0_FORMAT_SHIFT) |
 518                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 519       }
 520
 521       if (intel->gen >= 5)
 522           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 523                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 524                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 525                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 526       else
 527           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 528                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 529                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 530                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 531                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 532    }
 533    CACHED_BATCH();
 534 }
 535
 536 const struct brw_tracked_state brw_vertices = {
 537    .dirty = {
 538       .mesa = 0,
 539       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 540       .cache = 0,
 541    },
 542    .prepare = brw_prepare_vertices,
 543    .emit = brw_emit_vertices,
 544 };
 545
 546 static void brw_prepare_indices(struct brw_context *brw)
 547 {
 548    struct gl_context *ctx = &brw->intel.ctx;
 549    struct intel_context *intel = &brw->intel;
 550    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 551    GLuint ib_size;
 552    drm_intel_bo *bo = NULL;
 553    struct gl_buffer_object *bufferobj;
 554    GLuint offset;
 555    GLuint ib_type_size;
 556
 557    if (index_buffer == NULL)
 558       return;
 559
 560    ib_type_size = get_size(index_buffer->type);
 561    ib_size = ib_type_size * index_buffer->count;
 562    bufferobj = index_buffer->obj;
 563
 564    /* Turn into a proper VBO:
 565     */
 566    if (!_mesa_is_bufferobj(bufferobj)) {
 567
 568       /* Get new bufferobj, offset:
 569        */
 570       intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, &bo, &offset);
 571       brw->ib.start_vertex_offset = offset / ib_type_size;
 572       offset = 0;
 573    } else {
 574       offset = (GLuint) (unsigned long) index_buffer->ptr;
 575
 576       /* If the index buffer isn't aligned to its element size, we have to
 577        * rebase it into a temporary.
 578        */
 579        if ((get_size(index_buffer->type) - 1) & offset) {
 580            GLubyte *map = ctx->Driver.MapBuffer(ctx,
 581                                                 GL_ELEMENT_ARRAY_BUFFER_ARB,
 582                                                 GL_DYNAMIC_DRAW_ARB,
 583                                                 bufferobj);
 584            map += offset;
 585
 586            intel_upload_data(&brw->intel, map, ib_size, &bo, &offset);
 587            brw->ib.start_vertex_offset = offset / ib_type_size;
 588            offset = 0;
 589
 590            ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
 591        } else {
 592           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 593            * the index buffer state when we're just moving the start index
 594            * of our drawing.
 595            */
 596           brw->ib.start_vertex_offset = offset / ib_type_size;
 597
 598           bo = intel_bufferobj_source(intel, intel_buffer_object(bufferobj),
 599                                       &offset);
 600           drm_intel_bo_reference(bo);
 601        }
 602    }
 603
 604    if (brw->ib.bo != bo || brw->ib.offset != offset) {
 605       drm_intel_bo_unreference(brw->ib.bo);
 606       brw->ib.bo = bo;
 607       brw->ib.offset = offset;
 608
 609       brw_add_validated_bo(brw, brw->ib.bo);
 610       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 611    } else {
 612       drm_intel_bo_unreference(bo);
 613    }
 614 }
 615
 616 const struct brw_tracked_state brw_indices = {
 617    .dirty = {
 618       .mesa = 0,
 619       .brw = BRW_NEW_INDICES,
 620       .cache = 0,
 621    },
 622    .prepare = brw_prepare_indices,
 623 };
 624
 625 static void brw_emit_index_buffer(struct brw_context *brw)
 626 {
 627    struct intel_context *intel = &brw->intel;
 628    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 629
 630    if (index_buffer == NULL)
 631       return;
 632
 633    BEGIN_BATCH(3);
 634    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 635              /* cut index enable << 10 */
 636              get_index_type(index_buffer->type) << 8 |
 637              1);
 638    OUT_RELOC(brw->ib.bo,
 639              I915_GEM_DOMAIN_VERTEX, 0,
 640              brw->ib.offset);
 641    OUT_RELOC(brw->ib.bo,
 642              I915_GEM_DOMAIN_VERTEX, 0,
 643              brw->ib.bo->size - 1);
 644    ADVANCE_BATCH();
 645 }
 646
 647 const struct brw_tracked_state brw_index_buffer = {
 648    .dirty = {
 649       .mesa = 0,
 650       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 651       .cache = 0,
 652    },
 653    .emit = brw_emit_index_buffer,
 654 };