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: assert(0); 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 min, int max,
 246                         struct brw_vertex_buffer *buffer,
 247                         GLuint dst_stride)
 248 {
 249    int src_stride = element->glarray->StrideB;
 250    const unsigned char *src = element->glarray->Ptr + min * src_stride;
 251    int count = max - min + 1;
 252    GLuint size = count * dst_stride;
 253
 254    if (dst_stride == src_stride) {
 255       intel_upload_data(&brw->intel, src, size, dst_stride,
 256                         &buffer->bo, &buffer->offset);
 257    } else {
 258       char * const map = intel_upload_map(&brw->intel, size, dst_stride);
 259       char *dst = map;
 260
 261       while (count--) {
 262          memcpy(dst, src, dst_stride);
 263          src += src_stride;
 264          dst += dst_stride;
 265       }
 266       intel_upload_unmap(&brw->intel, map, size, dst_stride,
 267                          &buffer->bo, &buffer->offset);
 268    }
 269    buffer->stride = dst_stride;
 270 }
 271
 272 static void brw_prepare_vertices(struct brw_context *brw)
 273 {
 274    struct gl_context *ctx = &brw->intel.ctx;
 275    struct intel_context *intel = intel_context(ctx);
 276    GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
 277    const unsigned char *ptr = NULL;
 278    GLuint interleaved = 0, total_size = 0;
 279    unsigned int min_index = brw->vb.min_index;
 280    unsigned int max_index = brw->vb.max_index;
 281    int delta, i, j;
 282
 283    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 284    GLuint nr_uploads = 0;
 285
 286    /* First build an array of pointers to ve's in vb.inputs_read
 287     */
 288    if (0)
 289       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 290
 291    /* Accumulate the list of enabled arrays. */
 292    brw->vb.nr_enabled = 0;
 293    while (vs_inputs) {
 294       GLuint i = ffs(vs_inputs) - 1;
 295       struct brw_vertex_element *input = &brw->vb.inputs[i];
 296
 297       vs_inputs &= ~(1 << i);
 298       if (input->glarray->Size && get_size(input->glarray->Type))
 299          brw->vb.enabled[brw->vb.nr_enabled++] = input;
 300    }
 301
 302    if (brw->vb.nr_enabled == 0)
 303       return;
 304
 305    if (brw->vb.nr_buffers)
 306       goto validate;
 307
 308    /* XXX: In the rare cases where this happens we fallback all
 309     * the way to software rasterization, although a tnl fallback
 310     * would be sufficient.  I don't know of *any* real world
 311     * cases with > 17 vertex attributes enabled, so it probably
 312     * isn't an issue at this point.
 313     */
 314    if (brw->vb.nr_enabled >= BRW_VEP_MAX) {
 315       intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 316       return;
 317    }
 318
 319    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 320       struct brw_vertex_element *input = brw->vb.enabled[i];
 321       const struct gl_client_array *glarray = input->glarray;
 322       int type_size = get_size(glarray->Type);
 323
 324       input->element_size = type_size * glarray->Size;
 325
 326       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 327          struct intel_buffer_object *intel_buffer =
 328             intel_buffer_object(glarray->BufferObj);
 329          int k;
 330
 331          for (k = 0; k < i; k++) {
 332             const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
 333             if (glarray->BufferObj == other->BufferObj &&
 334                 glarray->StrideB == other->StrideB &&
 335                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 336             {
 337                input->buffer = brw->vb.enabled[k]->buffer;
 338                input->offset = glarray->Ptr - other->Ptr;
 339                break;
 340             }
 341          }
 342          if (k == i) {
 343             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 344
 345             /* Named buffer object: Just reference its contents directly. */
 346             buffer->bo = intel_bufferobj_source(intel,
 347                                                 intel_buffer, type_size,
 348                                                 &buffer->offset);
 349             drm_intel_bo_reference(buffer->bo);
 350             buffer->offset += (uintptr_t)glarray->Ptr;
 351             buffer->stride = glarray->StrideB;
 352
 353             input->buffer = j++;
 354             input->offset = 0;
 355          }
 356
 357          /* This is a common place to reach if the user mistakenly supplies
 358           * a pointer in place of a VBO offset.  If we just let it go through,
 359           * we may end up dereferencing a pointer beyond the bounds of the
 360           * GTT.  We would hope that the VBO's max_index would save us, but
 361           * Mesa appears to hand us min/max values not clipped to the
 362           * array object's _MaxElement, and _MaxElement frequently appears
 363           * to be wrong anyway.
 364           *
 365           * The VBO spec allows application termination in this case, and it's
 366           * probably a service to the poor programmer to do so rather than
 367           * trying to just not render.
 368           */
 369          assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
 370       } else {
 371          /* Queue the buffer object up to be uploaded in the next pass,
 372           * when we've decided if we're doing interleaved or not.
 373           */
 374          if (nr_uploads == 0) {
 375             /* Position array not properly enabled:
 376              */
 377             if (input->attrib == VERT_ATTRIB_POS && glarray->StrideB == 0) {
 378                intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 379                return;
 380             }
 381
 382             interleaved = glarray->StrideB;
 383             ptr = glarray->Ptr;
 384          }
 385          else if (interleaved != glarray->StrideB ||
 386                   (uintptr_t)(glarray->Ptr - ptr) > interleaved)
 387          {
 388             interleaved = 0;
 389          }
 390          else if ((uintptr_t)(glarray->Ptr - ptr) & (type_size -1))
 391          {
 392             /* enforce natural alignment (for doubles) */
 393             interleaved = 0;
 394          }
 395
 396          upload[nr_uploads++] = input;
 397          total_size = ALIGN(total_size, type_size);
 398          total_size += input->element_size;
 399       }
 400    }
 401
 402    /* If we need to upload all the arrays, then we can trim those arrays to
 403     * only the used elements [min_index, max_index] so long as we adjust all
 404     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 405     */
 406    brw->vb.start_vertex_bias = 0;
 407    delta = min_index;
 408    if (nr_uploads == brw->vb.nr_enabled) {
 409       brw->vb.start_vertex_bias = -delta;
 410       delta = 0;
 411    }
 412    if (delta && !brw->intel.intelScreen->relaxed_relocations)
 413       min_index = delta = 0;
 414
 415    /* Handle any arrays to be uploaded. */
 416    if (nr_uploads > 1) {
 417       if (interleaved && interleaved <= 2*total_size) {
 418          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 419          /* All uploads are interleaved, so upload the arrays together as
 420           * interleaved.  First, upload the contents and set up upload[0].
 421           */
 422          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 423                                  buffer, interleaved);
 424          buffer->offset -= delta * interleaved;
 425
 426          for (i = 0; i < nr_uploads; i++) {
 427             /* Then, just point upload[i] at upload[0]'s buffer. */
 428             upload[i]->offset =
 429                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 430             upload[i]->buffer = j;
 431          }
 432          j++;
 433
 434          nr_uploads = 0;
 435       }
 436       else if (total_size < 2048) {
 437          /* Upload non-interleaved arrays into a single interleaved array */
 438          struct brw_vertex_buffer *buffer;
 439          int count = max_index - min_index + 1;
 440          int offset;
 441          char *map;
 442
 443          map = intel_upload_map(&brw->intel, total_size * count, total_size);
 444          for (i = offset = 0; i < nr_uploads; i++) {
 445             const unsigned char *src = upload[i]->glarray->Ptr;
 446             int size = upload[i]->element_size;
 447             int stride = upload[i]->glarray->StrideB;
 448             char *dst;
 449             int n;
 450
 451             offset = ALIGN(offset, get_size(upload[i]->glarray->Type));
 452             dst = map + offset;
 453             src += min_index * stride;
 454
 455             for (n = 0; n < count; n++) {
 456                memcpy(dst, src, size);
 457                src += stride;
 458                dst += total_size;
 459             }
 460
 461             upload[i]->offset = offset;
 462             upload[i]->buffer = j;
 463
 464             offset += size;
 465          }
 466          assert(offset == total_size);
 467          buffer = &brw->vb.buffers[j++];
 468          intel_upload_unmap(&brw->intel, map, offset * count, offset,
 469                             &buffer->bo, &buffer->offset);
 470          buffer->stride = offset;
 471          buffer->offset -= delta * offset;
 472
 473          nr_uploads = 0;
 474       }
 475    }
 476    /* Upload non-interleaved arrays */
 477    for (i = 0; i < nr_uploads; i++) {
 478       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 479       copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 480                               buffer, upload[i]->element_size);
 481       buffer->offset -= delta * buffer->stride;
 482       upload[i]->buffer = j++;
 483       upload[i]->offset = 0;
 484    }
 485
 486    /* can we simply extend the current vb? */
 487    if (j == brw->vb.nr_current_buffers) {
 488       int delta = 0;
 489       for (i = 0; i < j; i++) {
 490          int d;
 491
 492          if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle ||
 493              brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride)
 494             break;
 495
 496          d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset;
 497          if (i == 0)
 498             delta = d / brw->vb.current_buffers[i].stride;
 499          if (delta * brw->vb.current_buffers[i].stride != d)
 500             break;
 501       }
 502
 503       if (i == j) {
 504          brw->vb.start_vertex_bias += delta;
 505          while (--j >= 0)
 506             drm_intel_bo_unreference(brw->vb.buffers[j].bo);
 507          j = 0;
 508       }
 509    }
 510
 511    brw->vb.nr_buffers = j;
 512
 513 validate:
 514    brw_prepare_query_begin(brw);
 515    for (i = 0; i < brw->vb.nr_buffers; i++) {
 516       brw_add_validated_bo(brw, brw->vb.buffers[i].bo);
 517    }
 518 }
 519
 520 static void brw_emit_vertices(struct brw_context *brw)
 521 {
 522    struct gl_context *ctx = &brw->intel.ctx;
 523    struct intel_context *intel = intel_context(ctx);
 524    GLuint i;
 525
 526    brw_emit_query_begin(brw);
 527
 528    /* If the VS doesn't read any inputs (calculating vertex position from
 529     * a state variable for some reason, for example), emit a single pad
 530     * VERTEX_ELEMENT struct and bail.
 531     *
 532     * The stale VB state stays in place, but they don't do anything unless
 533     * a VE loads from them.
 534     */
 535    if (brw->vb.nr_enabled == 0) {
 536       BEGIN_BATCH(3);
 537       OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
 538       if (intel->gen >= 6) {
 539          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 540                    GEN6_VE0_VALID |
 541                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 542                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 543       } else {
 544          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 545                    BRW_VE0_VALID |
 546                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 547                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 548       }
 549       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 550                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 551                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 552                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 553       CACHED_BATCH();
 554       return;
 555    }
 556
 557    /* Now emit VB and VEP state packets.
 558     */
 559
 560    if (brw->vb.nr_buffers) {
 561       BEGIN_BATCH(1 + 4*brw->vb.nr_buffers);
 562       OUT_BATCH((CMD_VERTEX_BUFFER << 16) | (4*brw->vb.nr_buffers - 1));
 563       for (i = 0; i < brw->vb.nr_buffers; i++) {
 564          struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 565          uint32_t dw0;
 566
 567          if (intel->gen >= 6) {
 568             dw0 = GEN6_VB0_ACCESS_VERTEXDATA | (i << GEN6_VB0_INDEX_SHIFT);
 569          } else {
 570             dw0 = BRW_VB0_ACCESS_VERTEXDATA | (i << BRW_VB0_INDEX_SHIFT);
 571          }
 572
 573          OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT));
 574          OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset);
 575          if (intel->gen >= 5) {
 576             OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1);
 577          } else
 578             OUT_BATCH(0);
 579          OUT_BATCH(0); /* Instance data step rate */
 580
 581          brw->vb.current_buffers[i].handle = buffer->bo->handle;
 582          brw->vb.current_buffers[i].offset = buffer->offset;
 583          brw->vb.current_buffers[i].stride = buffer->stride;
 584       }
 585       brw->vb.nr_current_buffers = i;
 586       ADVANCE_BATCH();
 587    }
 588
 589    BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
 590    OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | (2*brw->vb.nr_enabled - 1));
 591    for (i = 0; i < brw->vb.nr_enabled; i++) {
 592       struct brw_vertex_element *input = brw->vb.enabled[i];
 593       uint32_t format = get_surface_type(input->glarray->Type,
 594                                          input->glarray->Size,
 595                                          input->glarray->Format,
 596                                          input->glarray->Normalized);
 597       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 598       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 599       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 600       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 601
 602       switch (input->glarray->Size) {
 603       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 604       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 605       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 606       case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
 607          break;
 608       }
 609
 610       if (intel->gen >= 6) {
 611          OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
 612                    GEN6_VE0_VALID |
 613                    (format << BRW_VE0_FORMAT_SHIFT) |
 614                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 615       } else {
 616          OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) |
 617                    BRW_VE0_VALID |
 618                    (format << BRW_VE0_FORMAT_SHIFT) |
 619                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 620       }
 621
 622       if (intel->gen >= 5)
 623           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 624                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 625                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 626                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 627       else
 628           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 629                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 630                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 631                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 632                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 633    }
 634    CACHED_BATCH();
 635 }
 636
 637 const struct brw_tracked_state brw_vertices = {
 638    .dirty = {
 639       .mesa = 0,
 640       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 641       .cache = 0,
 642    },
 643    .prepare = brw_prepare_vertices,
 644    .emit = brw_emit_vertices,
 645 };
 646
 647 static void brw_prepare_indices(struct brw_context *brw)
 648 {
 649    struct gl_context *ctx = &brw->intel.ctx;
 650    struct intel_context *intel = &brw->intel;
 651    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 652    GLuint ib_size;
 653    drm_intel_bo *bo = NULL;
 654    struct gl_buffer_object *bufferobj;
 655    GLuint offset;
 656    GLuint ib_type_size;
 657
 658    if (index_buffer == NULL)
 659       return;
 660
 661    ib_type_size = get_size(index_buffer->type);
 662    ib_size = ib_type_size * index_buffer->count;
 663    bufferobj = index_buffer->obj;
 664
 665    /* Turn into a proper VBO:
 666     */
 667    if (!_mesa_is_bufferobj(bufferobj)) {
 668
 669       /* Get new bufferobj, offset:
 670        */
 671       intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size,
 672                         &bo, &offset);
 673       brw->ib.start_vertex_offset = offset / ib_type_size;
 674    } else {
 675       offset = (GLuint) (unsigned long) index_buffer->ptr;
 676
 677       /* If the index buffer isn't aligned to its element size, we have to
 678        * rebase it into a temporary.
 679        */
 680        if ((get_size(index_buffer->type) - 1) & offset) {
 681            GLubyte *map = ctx->Driver.MapBuffer(ctx,
 682                                                 GL_ELEMENT_ARRAY_BUFFER_ARB,
 683                                                 GL_DYNAMIC_DRAW_ARB,
 684                                                 bufferobj);
 685            map += offset;
 686
 687            intel_upload_data(&brw->intel, map, ib_size, ib_type_size,
 688                              &bo, &offset);
 689            brw->ib.start_vertex_offset = offset / ib_type_size;
 690
 691            ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
 692        } else {
 693           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 694            * the index buffer state when we're just moving the start index
 695            * of our drawing.
 696            */
 697           brw->ib.start_vertex_offset = offset / ib_type_size;
 698
 699           bo = intel_bufferobj_source(intel,
 700                                       intel_buffer_object(bufferobj),
 701                                       ib_type_size,
 702                                       &offset);
 703           drm_intel_bo_reference(bo);
 704
 705           brw->ib.start_vertex_offset += offset / ib_type_size;
 706        }
 707    }
 708
 709    if (brw->ib.bo != bo) {
 710       drm_intel_bo_unreference(brw->ib.bo);
 711       brw->ib.bo = bo;
 712
 713       brw_add_validated_bo(brw, brw->ib.bo);
 714       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 715    } else {
 716       drm_intel_bo_unreference(bo);
 717    }
 718
 719    if (index_buffer->type != brw->ib.type) {
 720       brw->ib.type = index_buffer->type;
 721       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 722    }
 723 }
 724
 725 const struct brw_tracked_state brw_indices = {
 726    .dirty = {
 727       .mesa = 0,
 728       .brw = BRW_NEW_INDICES,
 729       .cache = 0,
 730    },
 731    .prepare = brw_prepare_indices,
 732 };
 733
 734 static void brw_emit_index_buffer(struct brw_context *brw)
 735 {
 736    struct intel_context *intel = &brw->intel;
 737    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 738
 739    if (index_buffer == NULL)
 740       return;
 741
 742    BEGIN_BATCH(3);
 743    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 744              /* cut index enable << 10 */
 745              get_index_type(index_buffer->type) << 8 |
 746              1);
 747    OUT_RELOC(brw->ib.bo,
 748              I915_GEM_DOMAIN_VERTEX, 0,
 749              0);
 750    OUT_RELOC(brw->ib.bo,
 751              I915_GEM_DOMAIN_VERTEX, 0,
 752              brw->ib.bo->size - 1);
 753    ADVANCE_BATCH();
 754 }
 755
 756 const struct brw_tracked_state brw_index_buffer = {
 757    .dirty = {
 758       .mesa = 0,
 759       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 760       .cache = 0,
 761    },
 762    .emit = brw_emit_index_buffer,
 763 };