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