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                         int count,
 246                         struct brw_vertex_buffer *buffer,
 247                         GLuint dst_stride)
 248 {
 249    GLuint size = count * dst_stride;
 250
 251    buffer->stride = dst_stride;
 252    if (dst_stride == element->glarray->StrideB) {
 253       intel_upload_data(&brw->intel, element->glarray->Ptr, size, dst_stride,
 254                         &buffer->bo, &buffer->offset);
 255    } else {
 256       const unsigned char *src = element->glarray->Ptr;
 257       char *map = intel_upload_map(&brw->intel, size, dst_stride);
 258       char *dst = map;
 259       int i;
 260
 261       for (i = 0; i < count; i++) {
 262          memcpy(dst, src, dst_stride);
 263          src += element->glarray->StrideB;
 264          dst += dst_stride;
 265       }
 266       intel_upload_unmap(&brw->intel, map, size, dst_stride,
 267                          &buffer->bo, &buffer->offset);
 268    }
 269 }
 270
 271 static void brw_prepare_vertices(struct brw_context *brw)
 272 {
 273    struct gl_context *ctx = &brw->intel.ctx;
 274    struct intel_context *intel = intel_context(ctx);
 275    GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
 276    const unsigned char *ptr = NULL;
 277    GLuint interleaved = 0, total_size = 0;
 278    unsigned int min_index = brw->vb.min_index;
 279    unsigned int max_index = brw->vb.max_index;
 280    int i, j;
 281
 282    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 283    GLuint nr_uploads = 0;
 284
 285    /* First build an array of pointers to ve's in vb.inputs_read
 286     */
 287    if (0)
 288       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 289
 290    /* Accumulate the list of enabled arrays. */
 291    brw->vb.nr_enabled = 0;
 292    while (vs_inputs) {
 293       GLuint i = _mesa_ffsll(vs_inputs) - 1;
 294       struct brw_vertex_element *input = &brw->vb.inputs[i];
 295
 296       vs_inputs &= ~(1 << i);
 297       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 298    }
 299
 300    if (brw->vb.nr_enabled == 0)
 301       return;
 302
 303    if (brw->vb.nr_buffers)
 304       goto validate;
 305
 306    /* XXX: In the rare cases where this happens we fallback all
 307     * the way to software rasterization, although a tnl fallback
 308     * would be sufficient.  I don't know of *any* real world
 309     * cases with > 17 vertex attributes enabled, so it probably
 310     * isn't an issue at this point.
 311     */
 312    if (brw->vb.nr_enabled >= BRW_VEP_MAX) {
 313       intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 314       return;
 315    }
 316
 317    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 318       struct brw_vertex_element *input = brw->vb.enabled[i];
 319       const struct gl_client_array *glarray = input->glarray;
 320       int type_size = get_size(glarray->Type);
 321
 322       input->element_size = type_size * glarray->Size;
 323
 324       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 325          struct intel_buffer_object *intel_buffer =
 326             intel_buffer_object(glarray->BufferObj);
 327          int k;
 328
 329          for (k = 0; k < i; k++) {
 330             const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
 331             if (glarray->BufferObj == other->BufferObj &&
 332                 glarray->StrideB == other->StrideB &&
 333                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 334             {
 335                input->buffer = brw->vb.enabled[k]->buffer;
 336                input->offset = glarray->Ptr - other->Ptr;
 337                break;
 338             }
 339          }
 340          if (k == i) {
 341             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 342
 343             /* Named buffer object: Just reference its contents directly. */
 344             buffer->bo = intel_bufferobj_source(intel, intel_buffer,
 345                                                 &buffer->offset);
 346             drm_intel_bo_reference(buffer->bo);
 347             buffer->offset += (uintptr_t)glarray->Ptr;
 348             buffer->stride = glarray->StrideB;
 349
 350             input->buffer = j++;
 351             input->offset = 0;
 352          }
 353
 354          /* This is a common place to reach if the user mistakenly supplies
 355           * a pointer in place of a VBO offset.  If we just let it go through,
 356           * we may end up dereferencing a pointer beyond the bounds of the
 357           * GTT.  We would hope that the VBO's max_index would save us, but
 358           * Mesa appears to hand us min/max values not clipped to the
 359           * array object's _MaxElement, and _MaxElement frequently appears
 360           * to be wrong anyway.
 361           *
 362           * The VBO spec allows application termination in this case, and it's
 363           * probably a service to the poor programmer to do so rather than
 364           * trying to just not render.
 365           */
 366          assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
 367       } else {
 368          /* Queue the buffer object up to be uploaded in the next pass,
 369           * when we've decided if we're doing interleaved or not.
 370           */
 371          if (nr_uploads == 0) {
 372             /* Position array not properly enabled:
 373              */
 374             if (input->attrib == VERT_ATTRIB_POS && glarray->StrideB == 0) {
 375                intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 376                return;
 377             }
 378
 379             interleaved = glarray->StrideB;
 380             ptr = glarray->Ptr;
 381          }
 382          else if (interleaved != glarray->StrideB ||
 383                   (uintptr_t)(glarray->Ptr - ptr) > interleaved)
 384          {
 385             interleaved = 0;
 386          }
 387          else if ((uintptr_t)(glarray->Ptr - ptr) & (type_size -1))
 388          {
 389             /* enforce natural alignment (for doubles) */
 390             interleaved = 0;
 391          }
 392
 393          upload[nr_uploads++] = input;
 394          total_size = ALIGN(total_size, type_size);
 395          total_size += input->element_size;
 396       }
 397    }
 398
 399    /* If we need to upload all the arrays, then we can trim those arrays to
 400     * only the used elements [min_index, max_index] so long as we adjust all
 401     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 402     */
 403    if (nr_uploads == brw->vb.nr_enabled) {
 404       brw->vb.start_vertex_bias = min_index;
 405    } else {
 406       brw->vb.start_vertex_bias = 0;
 407       min_index = 0;
 408    }
 409
 410    /* Handle any arrays to be uploaded. */
 411    if (nr_uploads > 1) {
 412       if (interleaved && interleaved <= 2*total_size) {
 413          /* All uploads are interleaved, so upload the arrays together as
 414           * interleaved.  First, upload the contents and set up upload[0].
 415           */
 416          copy_array_to_vbo_array(brw, upload[0], max_index - min_index + 1,
 417                                  &brw->vb.buffers[j], interleaved);
 418
 419          for (i = 0; i < nr_uploads; i++) {
 420             /* Then, just point upload[i] at upload[0]'s buffer. */
 421             upload[i]->offset =
 422                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 423             upload[i]->buffer = j;
 424          }
 425          j++;
 426
 427          nr_uploads = 0;
 428       }
 429       else if (total_size < 2048) {
 430          /* Upload non-interleaved arrays into a single interleaved array */
 431          struct brw_vertex_buffer *buffer;
 432          int count = max_index - min_index + 1;
 433          int offset;
 434          char *map;
 435
 436          map = intel_upload_map(&brw->intel, total_size * count, total_size);
 437          for (i = offset = 0; i < nr_uploads; i++) {
 438             const unsigned char *src = upload[i]->glarray->Ptr;
 439             int size = upload[i]->element_size;
 440             int stride = upload[i]->glarray->StrideB;
 441             char *dst;
 442             int n;
 443
 444             offset = ALIGN(offset, get_size(upload[i]->glarray->Type));
 445             dst = map + offset;
 446             src += min_index * size;
 447
 448             for (n = 0; n < count; n++) {
 449                memcpy(dst, src, size);
 450                src += stride;
 451                dst += total_size;
 452             }
 453
 454             upload[i]->offset = offset;
 455             upload[i]->buffer = j;
 456
 457             offset += size;
 458          }
 459          assert(offset == total_size);
 460          buffer = &brw->vb.buffers[j++];
 461          intel_upload_unmap(&brw->intel, map, offset * count, offset,
 462                             &buffer->bo, &buffer->offset);
 463          buffer->stride = offset;
 464
 465          nr_uploads = 0;
 466       }
 467    }
 468    /* Upload non-interleaved arrays */
 469    for (i = 0; i < nr_uploads; i++) {
 470       copy_array_to_vbo_array(brw, upload[i], max_index - min_index + 1,
 471                               &brw->vb.buffers[j], upload[i]->element_size);
 472       upload[i]->buffer = j++;
 473       upload[i]->offset = 0;
 474    }
 475
 476    /* can we simply extend the current vb? */
 477    if (j == brw->vb.nr_current_buffers) {
 478       int delta = 0;
 479       for (i = 0; i < j; i++) {
 480          int d;
 481
 482          if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle ||
 483              brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride)
 484             break;
 485
 486          d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset;
 487          if (delta == 0)
 488             delta = d / brw->vb.current_buffers[i].stride;
 489          if (delta * brw->vb.current_buffers[i].stride != d)
 490             break;
 491       }
 492
 493       if (i == j) {
 494          brw->vb.start_vertex_bias += delta;
 495          while (--j >= 0)
 496             drm_intel_bo_unreference(brw->vb.buffers[j].bo);
 497          j = 0;
 498       }
 499    }
 500
 501    brw->vb.nr_buffers = j;
 502
 503 validate:
 504    brw_prepare_query_begin(brw);
 505    for (i = 0; i < brw->vb.nr_buffers; i++) {
 506       brw_add_validated_bo(brw, brw->vb.buffers[i].bo);
 507    }
 508 }
 509
 510 static void brw_emit_vertices(struct brw_context *brw)
 511 {
 512    struct gl_context *ctx = &brw->intel.ctx;
 513    struct intel_context *intel = intel_context(ctx);
 514    GLuint i;
 515
 516    brw_emit_query_begin(brw);
 517
 518    /* If the VS doesn't read any inputs (calculating vertex position from
 519     * a state variable for some reason, for example), emit a single pad
 520     * VERTEX_ELEMENT struct and bail.
 521     *
 522     * The stale VB state stays in place, but they don't do anything unless
 523     * a VE loads from them.
 524     */
 525    if (brw->vb.nr_enabled == 0) {
 526       BEGIN_BATCH(3);
 527       OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
 528       if (intel->gen >= 6) {
 529          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 530                    GEN6_VE0_VALID |
 531                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 532                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 533       } else {
 534          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 535                    BRW_VE0_VALID |
 536                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 537                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 538       }
 539       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 540                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 541                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 542                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 543       CACHED_BATCH();
 544       return;
 545    }
 546
 547    /* Now emit VB and VEP state packets.
 548     */
 549
 550    if (brw->vb.nr_buffers) {
 551       BEGIN_BATCH(1 + 4*brw->vb.nr_buffers);
 552       OUT_BATCH((CMD_VERTEX_BUFFER << 16) | (4*brw->vb.nr_buffers - 1));
 553       for (i = 0; i < brw->vb.nr_buffers; i++) {
 554          struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 555          uint32_t dw0;
 556
 557          if (intel->gen >= 6) {
 558             dw0 = GEN6_VB0_ACCESS_VERTEXDATA | (i << GEN6_VB0_INDEX_SHIFT);
 559          } else {
 560             dw0 = BRW_VB0_ACCESS_VERTEXDATA | (i << BRW_VB0_INDEX_SHIFT);
 561          }
 562
 563          OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT));
 564          OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset);
 565          if (intel->gen >= 5) {
 566             OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1);
 567          } else
 568             OUT_BATCH(buffer->bo->size / buffer->stride);
 569          OUT_BATCH(0); /* Instance data step rate */
 570
 571          brw->vb.current_buffers[i].handle = buffer->bo->handle;
 572          brw->vb.current_buffers[i].offset = buffer->offset;
 573          brw->vb.current_buffers[i].stride = buffer->stride;
 574       }
 575       brw->vb.nr_current_buffers = i;
 576       ADVANCE_BATCH();
 577    }
 578
 579    BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
 580    OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | (2*brw->vb.nr_enabled - 1));
 581    for (i = 0; i < brw->vb.nr_enabled; i++) {
 582       struct brw_vertex_element *input = brw->vb.enabled[i];
 583       uint32_t format = get_surface_type(input->glarray->Type,
 584                                          input->glarray->Size,
 585                                          input->glarray->Format,
 586                                          input->glarray->Normalized);
 587       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 588       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 589       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 590       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 591
 592       switch (input->glarray->Size) {
 593       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 594       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 595       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 596       case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
 597          break;
 598       }
 599
 600       if (intel->gen >= 6) {
 601          OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
 602                    GEN6_VE0_VALID |
 603                    (format << BRW_VE0_FORMAT_SHIFT) |
 604                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 605       } else {
 606          OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) |
 607                    BRW_VE0_VALID |
 608                    (format << BRW_VE0_FORMAT_SHIFT) |
 609                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 610       }
 611
 612       if (intel->gen >= 5)
 613           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 614                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 615                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 616                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 617       else
 618           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 619                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 620                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 621                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 622                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 623    }
 624    CACHED_BATCH();
 625 }
 626
 627 const struct brw_tracked_state brw_vertices = {
 628    .dirty = {
 629       .mesa = 0,
 630       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 631       .cache = 0,
 632    },
 633    .prepare = brw_prepare_vertices,
 634    .emit = brw_emit_vertices,
 635 };
 636
 637 static void brw_prepare_indices(struct brw_context *brw)
 638 {
 639    struct gl_context *ctx = &brw->intel.ctx;
 640    struct intel_context *intel = &brw->intel;
 641    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 642    GLuint ib_size;
 643    drm_intel_bo *bo = NULL;
 644    struct gl_buffer_object *bufferobj;
 645    GLuint offset;
 646    GLuint ib_type_size;
 647
 648    if (index_buffer == NULL)
 649       return;
 650
 651    ib_type_size = get_size(index_buffer->type);
 652    ib_size = ib_type_size * index_buffer->count;
 653    bufferobj = index_buffer->obj;
 654
 655    /* Turn into a proper VBO:
 656     */
 657    if (!_mesa_is_bufferobj(bufferobj)) {
 658
 659       /* Get new bufferobj, offset:
 660        */
 661       intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size,
 662                         &bo, &offset);
 663       brw->ib.start_vertex_offset = offset / ib_type_size;
 664       offset = 0;
 665    } else {
 666       offset = (GLuint) (unsigned long) index_buffer->ptr;
 667
 668       /* If the index buffer isn't aligned to its element size, we have to
 669        * rebase it into a temporary.
 670        */
 671        if ((get_size(index_buffer->type) - 1) & offset) {
 672            GLubyte *map = ctx->Driver.MapBuffer(ctx,
 673                                                 GL_ELEMENT_ARRAY_BUFFER_ARB,
 674                                                 GL_DYNAMIC_DRAW_ARB,
 675                                                 bufferobj);
 676            map += offset;
 677
 678            intel_upload_data(&brw->intel, map, ib_size, ib_type_size,
 679                              &bo, &offset);
 680            brw->ib.start_vertex_offset = offset / ib_type_size;
 681            offset = 0;
 682
 683            ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
 684        } else {
 685           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 686            * the index buffer state when we're just moving the start index
 687            * of our drawing.
 688            */
 689           brw->ib.start_vertex_offset = offset / ib_type_size;
 690
 691           bo = intel_bufferobj_source(intel, intel_buffer_object(bufferobj),
 692                                       &offset);
 693           drm_intel_bo_reference(bo);
 694        }
 695    }
 696
 697    if (brw->ib.bo != bo || brw->ib.offset != offset) {
 698       drm_intel_bo_unreference(brw->ib.bo);
 699       brw->ib.bo = bo;
 700       brw->ib.offset = offset;
 701
 702       brw_add_validated_bo(brw, brw->ib.bo);
 703       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 704    } else {
 705       drm_intel_bo_unreference(bo);
 706    }
 707 }
 708
 709 const struct brw_tracked_state brw_indices = {
 710    .dirty = {
 711       .mesa = 0,
 712       .brw = BRW_NEW_INDICES,
 713       .cache = 0,
 714    },
 715    .prepare = brw_prepare_indices,
 716 };
 717
 718 static void brw_emit_index_buffer(struct brw_context *brw)
 719 {
 720    struct intel_context *intel = &brw->intel;
 721    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 722
 723    if (index_buffer == NULL)
 724       return;
 725
 726    BEGIN_BATCH(3);
 727    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 728              /* cut index enable << 10 */
 729              get_index_type(index_buffer->type) << 8 |
 730              1);
 731    OUT_RELOC(brw->ib.bo,
 732              I915_GEM_DOMAIN_VERTEX, 0,
 733              brw->ib.offset);
 734    OUT_RELOC(brw->ib.bo,
 735              I915_GEM_DOMAIN_VERTEX, 0,
 736              brw->ib.bo->size - 1);
 737    ADVANCE_BATCH();
 738 }
 739
 740 const struct brw_tracked_state brw_index_buffer = {
 741    .dirty = {
 742       .mesa = 0,
 743       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 744       .cache = 0,
 745    },
 746    .emit = brw_emit_index_buffer,
 747 };