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       /* This produces GL_FIXED inputs as values between INT32_MIN and
 211        * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 212        */
 213       case GL_FIXED: return int_types_scale[size];
 214       default: assert(0); return 0;
 215       }
 216    }
 217 }
 218
 219
 220 static GLuint get_size( GLenum type )
 221 {
 222    switch (type) {
 223    case GL_DOUBLE: return sizeof(GLdouble);
 224    case GL_FLOAT: return sizeof(GLfloat);
 225    case GL_HALF_FLOAT: return sizeof(GLhalfARB);
 226    case GL_INT: return sizeof(GLint);
 227    case GL_SHORT: return sizeof(GLshort);
 228    case GL_BYTE: return sizeof(GLbyte);
 229    case GL_UNSIGNED_INT: return sizeof(GLuint);
 230    case GL_UNSIGNED_SHORT: return sizeof(GLushort);
 231    case GL_UNSIGNED_BYTE: return sizeof(GLubyte);
 232    case GL_FIXED: return sizeof(GLuint);
 233    default: assert(0); return 0;
 234    }
 235 }
 236
 237 static GLuint get_index_type(GLenum type)
 238 {
 239    switch (type) {
 240    case GL_UNSIGNED_BYTE:  return BRW_INDEX_BYTE;
 241    case GL_UNSIGNED_SHORT: return BRW_INDEX_WORD;
 242    case GL_UNSIGNED_INT:   return BRW_INDEX_DWORD;
 243    default: assert(0); return 0;
 244    }
 245 }
 246
 247 static void
 248 copy_array_to_vbo_array(struct brw_context *brw,
 249                         struct brw_vertex_element *element,
 250                         int min, int max,
 251                         struct brw_vertex_buffer *buffer,
 252                         GLuint dst_stride)
 253 {
 254    int src_stride = element->glarray->StrideB;
 255    const unsigned char *src = element->glarray->Ptr + min * src_stride;
 256    int count = max - min + 1;
 257    GLuint size = count * dst_stride;
 258
 259    if (dst_stride == src_stride) {
 260       intel_upload_data(&brw->intel, src, size, dst_stride,
 261                         &buffer->bo, &buffer->offset);
 262    } else {
 263       char * const map = intel_upload_map(&brw->intel, size, dst_stride);
 264       char *dst = map;
 265
 266       while (count--) {
 267          memcpy(dst, src, dst_stride);
 268          src += src_stride;
 269          dst += dst_stride;
 270       }
 271       intel_upload_unmap(&brw->intel, map, size, dst_stride,
 272                          &buffer->bo, &buffer->offset);
 273    }
 274    buffer->stride = dst_stride;
 275 }
 276
 277 static void brw_prepare_vertices(struct brw_context *brw)
 278 {
 279    struct gl_context *ctx = &brw->intel.ctx;
 280    struct intel_context *intel = intel_context(ctx);
 281    /* CACHE_NEW_VS_PROG */
 282    GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
 283    const unsigned char *ptr = NULL;
 284    GLuint interleaved = 0, total_size = 0;
 285    unsigned int min_index = brw->vb.min_index;
 286    unsigned int max_index = brw->vb.max_index;
 287    int delta, i, j;
 288
 289    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 290    GLuint nr_uploads = 0;
 291
 292    /* First build an array of pointers to ve's in vb.inputs_read
 293     */
 294    if (0)
 295       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 296
 297    /* Accumulate the list of enabled arrays. */
 298    brw->vb.nr_enabled = 0;
 299    while (vs_inputs) {
 300       GLuint i = ffs(vs_inputs) - 1;
 301       struct brw_vertex_element *input = &brw->vb.inputs[i];
 302
 303       vs_inputs &= ~(1 << i);
 304       if (input->glarray->Size && get_size(input->glarray->Type))
 305          brw->vb.enabled[brw->vb.nr_enabled++] = input;
 306    }
 307
 308    if (brw->vb.nr_enabled == 0)
 309       return;
 310
 311    if (brw->vb.nr_buffers)
 312       goto validate;
 313
 314    /* XXX: In the rare cases where this happens we fallback all
 315     * the way to software rasterization, although a tnl fallback
 316     * would be sufficient.  I don't know of *any* real world
 317     * cases with > 17 vertex attributes enabled, so it probably
 318     * isn't an issue at this point.
 319     */
 320    if (brw->vb.nr_enabled >= BRW_VEP_MAX) {
 321       intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 322       return;
 323    }
 324
 325    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 326       struct brw_vertex_element *input = brw->vb.enabled[i];
 327       const struct gl_client_array *glarray = input->glarray;
 328       int type_size = get_size(glarray->Type);
 329
 330       input->element_size = type_size * glarray->Size;
 331
 332       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 333          struct intel_buffer_object *intel_buffer =
 334             intel_buffer_object(glarray->BufferObj);
 335          int k;
 336
 337          for (k = 0; k < i; k++) {
 338             const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
 339             if (glarray->BufferObj == other->BufferObj &&
 340                 glarray->StrideB == other->StrideB &&
 341                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 342             {
 343                input->buffer = brw->vb.enabled[k]->buffer;
 344                input->offset = glarray->Ptr - other->Ptr;
 345                break;
 346             }
 347          }
 348          if (k == i) {
 349             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 350
 351             /* Named buffer object: Just reference its contents directly. */
 352             buffer->bo = intel_bufferobj_source(intel,
 353                                                 intel_buffer, type_size,
 354                                                 &buffer->offset);
 355             drm_intel_bo_reference(buffer->bo);
 356             buffer->offset += (uintptr_t)glarray->Ptr;
 357             buffer->stride = glarray->StrideB;
 358
 359             input->buffer = j++;
 360             input->offset = 0;
 361          }
 362
 363          /* This is a common place to reach if the user mistakenly supplies
 364           * a pointer in place of a VBO offset.  If we just let it go through,
 365           * we may end up dereferencing a pointer beyond the bounds of the
 366           * GTT.  We would hope that the VBO's max_index would save us, but
 367           * Mesa appears to hand us min/max values not clipped to the
 368           * array object's _MaxElement, and _MaxElement frequently appears
 369           * to be wrong anyway.
 370           *
 371           * The VBO spec allows application termination in this case, and it's
 372           * probably a service to the poor programmer to do so rather than
 373           * trying to just not render.
 374           */
 375          assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
 376       } else {
 377          /* Queue the buffer object up to be uploaded in the next pass,
 378           * when we've decided if we're doing interleaved or not.
 379           */
 380          if (nr_uploads == 0) {
 381             /* Position array not properly enabled:
 382              */
 383             if (input->attrib == VERT_ATTRIB_POS && glarray->StrideB == 0) {
 384                intel->Fallback = GL_TRUE; /* boolean, not bitfield */
 385                return;
 386             }
 387
 388             interleaved = glarray->StrideB;
 389             ptr = glarray->Ptr;
 390          }
 391          else if (interleaved != glarray->StrideB ||
 392                   (uintptr_t)(glarray->Ptr - ptr) > interleaved)
 393          {
 394             interleaved = 0;
 395          }
 396          else if ((uintptr_t)(glarray->Ptr - ptr) & (type_size -1))
 397          {
 398             /* enforce natural alignment (for doubles) */
 399             interleaved = 0;
 400          }
 401
 402          upload[nr_uploads++] = input;
 403          total_size = ALIGN(total_size, type_size);
 404          total_size += input->element_size;
 405       }
 406    }
 407
 408    /* If we need to upload all the arrays, then we can trim those arrays to
 409     * only the used elements [min_index, max_index] so long as we adjust all
 410     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 411     */
 412    brw->vb.start_vertex_bias = 0;
 413    delta = min_index;
 414    if (nr_uploads == brw->vb.nr_enabled) {
 415       brw->vb.start_vertex_bias = -delta;
 416       delta = 0;
 417    }
 418    if (delta && !brw->intel.intelScreen->relaxed_relocations)
 419       min_index = delta = 0;
 420
 421    /* Handle any arrays to be uploaded. */
 422    if (nr_uploads > 1) {
 423       if (interleaved && interleaved <= 2*total_size) {
 424          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 425          /* All uploads are interleaved, so upload the arrays together as
 426           * interleaved.  First, upload the contents and set up upload[0].
 427           */
 428          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 429                                  buffer, interleaved);
 430          buffer->offset -= delta * interleaved;
 431
 432          for (i = 0; i < nr_uploads; i++) {
 433             /* Then, just point upload[i] at upload[0]'s buffer. */
 434             upload[i]->offset =
 435                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 436             upload[i]->buffer = j;
 437          }
 438          j++;
 439
 440          nr_uploads = 0;
 441       }
 442       else if (total_size < 2048) {
 443          /* Upload non-interleaved arrays into a single interleaved array */
 444          struct brw_vertex_buffer *buffer;
 445          int count = max_index - min_index + 1;
 446          int offset;
 447          char *map;
 448
 449          map = intel_upload_map(&brw->intel, total_size * count, total_size);
 450          for (i = offset = 0; i < nr_uploads; i++) {
 451             const unsigned char *src = upload[i]->glarray->Ptr;
 452             int size = upload[i]->element_size;
 453             int stride = upload[i]->glarray->StrideB;
 454             char *dst;
 455             int n;
 456
 457             offset = ALIGN(offset, get_size(upload[i]->glarray->Type));
 458             dst = map + offset;
 459             src += min_index * stride;
 460
 461             for (n = 0; n < count; n++) {
 462                memcpy(dst, src, size);
 463                src += stride;
 464                dst += total_size;
 465             }
 466
 467             upload[i]->offset = offset;
 468             upload[i]->buffer = j;
 469
 470             offset += size;
 471          }
 472          assert(offset == total_size);
 473          buffer = &brw->vb.buffers[j++];
 474          intel_upload_unmap(&brw->intel, map, offset * count, offset,
 475                             &buffer->bo, &buffer->offset);
 476          buffer->stride = offset;
 477          buffer->offset -= delta * offset;
 478
 479          nr_uploads = 0;
 480       }
 481    }
 482    /* Upload non-interleaved arrays */
 483    for (i = 0; i < nr_uploads; i++) {
 484       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 485       copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 486                               buffer, upload[i]->element_size);
 487       buffer->offset -= delta * buffer->stride;
 488       upload[i]->buffer = j++;
 489       upload[i]->offset = 0;
 490    }
 491
 492    /* can we simply extend the current vb? */
 493    if (j == brw->vb.nr_current_buffers) {
 494       int delta = 0;
 495       for (i = 0; i < j; i++) {
 496          int d;
 497
 498          if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle ||
 499              brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride)
 500             break;
 501
 502          d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset;
 503          if (d < 0)
 504             break;
 505          if (i == 0)
 506             delta = d / brw->vb.current_buffers[i].stride;
 507          if (delta * brw->vb.current_buffers[i].stride != d)
 508             break;
 509       }
 510
 511       if (i == j) {
 512          brw->vb.start_vertex_bias += delta;
 513          while (--j >= 0)
 514             drm_intel_bo_unreference(brw->vb.buffers[j].bo);
 515          j = 0;
 516       }
 517    }
 518
 519    brw->vb.nr_buffers = j;
 520
 521 validate:
 522    brw_prepare_query_begin(brw);
 523    for (i = 0; i < brw->vb.nr_buffers; i++) {
 524       brw_add_validated_bo(brw, brw->vb.buffers[i].bo);
 525    }
 526 }
 527
 528 static void brw_emit_vertices(struct brw_context *brw)
 529 {
 530    struct gl_context *ctx = &brw->intel.ctx;
 531    struct intel_context *intel = intel_context(ctx);
 532    GLuint i;
 533
 534    brw_emit_query_begin(brw);
 535
 536    /* If the VS doesn't read any inputs (calculating vertex position from
 537     * a state variable for some reason, for example), emit a single pad
 538     * VERTEX_ELEMENT struct and bail.
 539     *
 540     * The stale VB state stays in place, but they don't do anything unless
 541     * a VE loads from them.
 542     */
 543    if (brw->vb.nr_enabled == 0) {
 544       BEGIN_BATCH(3);
 545       OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
 546       if (intel->gen >= 6) {
 547          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 548                    GEN6_VE0_VALID |
 549                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 550                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 551       } else {
 552          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 553                    BRW_VE0_VALID |
 554                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 555                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 556       }
 557       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 558                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 559                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 560                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 561       CACHED_BATCH();
 562       return;
 563    }
 564
 565    /* Now emit VB and VEP state packets.
 566     */
 567
 568    if (brw->vb.nr_buffers) {
 569       BEGIN_BATCH(1 + 4*brw->vb.nr_buffers);
 570       OUT_BATCH((CMD_VERTEX_BUFFER << 16) | (4*brw->vb.nr_buffers - 1));
 571       for (i = 0; i < brw->vb.nr_buffers; i++) {
 572          struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 573          uint32_t dw0;
 574
 575          if (intel->gen >= 6) {
 576             dw0 = GEN6_VB0_ACCESS_VERTEXDATA | (i << GEN6_VB0_INDEX_SHIFT);
 577          } else {
 578             dw0 = BRW_VB0_ACCESS_VERTEXDATA | (i << BRW_VB0_INDEX_SHIFT);
 579          }
 580
 581          if (intel->gen >= 7)
 582             dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
 583
 584          OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT));
 585          OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset);
 586          if (intel->gen >= 5) {
 587             OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1);
 588          } else
 589             OUT_BATCH(0);
 590          OUT_BATCH(0); /* Instance data step rate */
 591
 592          brw->vb.current_buffers[i].handle = buffer->bo->handle;
 593          brw->vb.current_buffers[i].offset = buffer->offset;
 594          brw->vb.current_buffers[i].stride = buffer->stride;
 595       }
 596       brw->vb.nr_current_buffers = i;
 597       ADVANCE_BATCH();
 598    }
 599
 600    BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
 601    OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | (2*brw->vb.nr_enabled - 1));
 602    for (i = 0; i < brw->vb.nr_enabled; i++) {
 603       struct brw_vertex_element *input = brw->vb.enabled[i];
 604       uint32_t format = get_surface_type(input->glarray->Type,
 605                                          input->glarray->Size,
 606                                          input->glarray->Format,
 607                                          input->glarray->Normalized);
 608       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 609       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 610       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 611       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 612
 613       switch (input->glarray->Size) {
 614       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 615       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 616       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 617       case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
 618          break;
 619       }
 620
 621       if (intel->gen >= 6) {
 622          OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
 623                    GEN6_VE0_VALID |
 624                    (format << BRW_VE0_FORMAT_SHIFT) |
 625                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 626       } else {
 627          OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) |
 628                    BRW_VE0_VALID |
 629                    (format << BRW_VE0_FORMAT_SHIFT) |
 630                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 631       }
 632
 633       if (intel->gen >= 5)
 634           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 635                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 636                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 637                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 638       else
 639           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 640                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 641                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 642                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 643                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 644    }
 645    CACHED_BATCH();
 646 }
 647
 648 const struct brw_tracked_state brw_vertices = {
 649    .dirty = {
 650       .mesa = 0,
 651       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 652       .cache = CACHE_NEW_VS_PROG,
 653    },
 654    .prepare = brw_prepare_vertices,
 655    .emit = brw_emit_vertices,
 656 };
 657
 658 static void brw_prepare_indices(struct brw_context *brw)
 659 {
 660    struct gl_context *ctx = &brw->intel.ctx;
 661    struct intel_context *intel = &brw->intel;
 662    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 663    GLuint ib_size;
 664    drm_intel_bo *bo = NULL;
 665    struct gl_buffer_object *bufferobj;
 666    GLuint offset;
 667    GLuint ib_type_size;
 668
 669    if (index_buffer == NULL)
 670       return;
 671
 672    ib_type_size = get_size(index_buffer->type);
 673    ib_size = ib_type_size * index_buffer->count;
 674    bufferobj = index_buffer->obj;
 675
 676    /* Turn into a proper VBO:
 677     */
 678    if (!_mesa_is_bufferobj(bufferobj)) {
 679
 680       /* Get new bufferobj, offset:
 681        */
 682       intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size,
 683                         &bo, &offset);
 684       brw->ib.start_vertex_offset = offset / ib_type_size;
 685    } else {
 686       offset = (GLuint) (unsigned long) index_buffer->ptr;
 687
 688       /* If the index buffer isn't aligned to its element size, we have to
 689        * rebase it into a temporary.
 690        */
 691        if ((get_size(index_buffer->type) - 1) & offset) {
 692            GLubyte *map = ctx->Driver.MapBuffer(ctx,
 693                                                 GL_ELEMENT_ARRAY_BUFFER_ARB,
 694                                                 GL_DYNAMIC_DRAW_ARB,
 695                                                 bufferobj);
 696            map += offset;
 697
 698            intel_upload_data(&brw->intel, map, ib_size, ib_type_size,
 699                              &bo, &offset);
 700            brw->ib.start_vertex_offset = offset / ib_type_size;
 701
 702            ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
 703        } else {
 704           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 705            * the index buffer state when we're just moving the start index
 706            * of our drawing.
 707            */
 708           brw->ib.start_vertex_offset = offset / ib_type_size;
 709
 710           bo = intel_bufferobj_source(intel,
 711                                       intel_buffer_object(bufferobj),
 712                                       ib_type_size,
 713                                       &offset);
 714           drm_intel_bo_reference(bo);
 715
 716           brw->ib.start_vertex_offset += offset / ib_type_size;
 717        }
 718    }
 719
 720    if (brw->ib.bo != bo) {
 721       drm_intel_bo_unreference(brw->ib.bo);
 722       brw->ib.bo = bo;
 723
 724       brw_add_validated_bo(brw, brw->ib.bo);
 725       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 726    } else {
 727       drm_intel_bo_unreference(bo);
 728    }
 729
 730    if (index_buffer->type != brw->ib.type) {
 731       brw->ib.type = index_buffer->type;
 732       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 733    }
 734 }
 735
 736 const struct brw_tracked_state brw_indices = {
 737    .dirty = {
 738       .mesa = 0,
 739       .brw = BRW_NEW_INDICES,
 740       .cache = 0,
 741    },
 742    .prepare = brw_prepare_indices,
 743 };
 744
 745 static void brw_emit_index_buffer(struct brw_context *brw)
 746 {
 747    struct intel_context *intel = &brw->intel;
 748    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 749
 750    if (index_buffer == NULL)
 751       return;
 752
 753    BEGIN_BATCH(3);
 754    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 755              /* cut index enable << 10 */
 756              get_index_type(index_buffer->type) << 8 |
 757              1);
 758    OUT_RELOC(brw->ib.bo,
 759              I915_GEM_DOMAIN_VERTEX, 0,
 760              0);
 761    OUT_RELOC(brw->ib.bo,
 762              I915_GEM_DOMAIN_VERTEX, 0,
 763              brw->ib.bo->size - 1);
 764    ADVANCE_BATCH();
 765 }
 766
 767 const struct brw_tracked_state brw_index_buffer = {
 768    .dirty = {
 769       .mesa = 0,
 770       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 771       .cache = 0,
 772    },
 773    .emit = brw_emit_index_buffer,
 774 };