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 #include "main/macros.h"
  35
  36 #include "brw_draw.h"
  37 #include "brw_defines.h"
  38 #include "brw_context.h"
  39 #include "brw_state.h"
  40
  41 #include "intel_batchbuffer.h"
  42 #include "intel_buffer_objects.h"
  43
  44 static GLuint double_types[5] = {
  45    0,
  46    BRW_SURFACEFORMAT_R64_FLOAT,
  47    BRW_SURFACEFORMAT_R64G64_FLOAT,
  48    BRW_SURFACEFORMAT_R64G64B64_FLOAT,
  49    BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
  50 };
  51
  52 static GLuint float_types[5] = {
  53    0,
  54    BRW_SURFACEFORMAT_R32_FLOAT,
  55    BRW_SURFACEFORMAT_R32G32_FLOAT,
  56    BRW_SURFACEFORMAT_R32G32B32_FLOAT,
  57    BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
  58 };
  59
  60 static GLuint half_float_types[5] = {
  61    0,
  62    BRW_SURFACEFORMAT_R16_FLOAT,
  63    BRW_SURFACEFORMAT_R16G16_FLOAT,
  64    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT,
  65    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
  66 };
  67
  68 static GLuint uint_types_norm[5] = {
  69    0,
  70    BRW_SURFACEFORMAT_R32_UNORM,
  71    BRW_SURFACEFORMAT_R32G32_UNORM,
  72    BRW_SURFACEFORMAT_R32G32B32_UNORM,
  73    BRW_SURFACEFORMAT_R32G32B32A32_UNORM
  74 };
  75
  76 static GLuint uint_types_scale[5] = {
  77    0,
  78    BRW_SURFACEFORMAT_R32_USCALED,
  79    BRW_SURFACEFORMAT_R32G32_USCALED,
  80    BRW_SURFACEFORMAT_R32G32B32_USCALED,
  81    BRW_SURFACEFORMAT_R32G32B32A32_USCALED
  82 };
  83
  84 static GLuint int_types_norm[5] = {
  85    0,
  86    BRW_SURFACEFORMAT_R32_SNORM,
  87    BRW_SURFACEFORMAT_R32G32_SNORM,
  88    BRW_SURFACEFORMAT_R32G32B32_SNORM,
  89    BRW_SURFACEFORMAT_R32G32B32A32_SNORM
  90 };
  91
  92 static GLuint int_types_scale[5] = {
  93    0,
  94    BRW_SURFACEFORMAT_R32_SSCALED,
  95    BRW_SURFACEFORMAT_R32G32_SSCALED,
  96    BRW_SURFACEFORMAT_R32G32B32_SSCALED,
  97    BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
  98 };
  99
 100 static GLuint ushort_types_norm[5] = {
 101    0,
 102    BRW_SURFACEFORMAT_R16_UNORM,
 103    BRW_SURFACEFORMAT_R16G16_UNORM,
 104    BRW_SURFACEFORMAT_R16G16B16_UNORM,
 105    BRW_SURFACEFORMAT_R16G16B16A16_UNORM
 106 };
 107
 108 static GLuint ushort_types_scale[5] = {
 109    0,
 110    BRW_SURFACEFORMAT_R16_USCALED,
 111    BRW_SURFACEFORMAT_R16G16_USCALED,
 112    BRW_SURFACEFORMAT_R16G16B16_USCALED,
 113    BRW_SURFACEFORMAT_R16G16B16A16_USCALED
 114 };
 115
 116 static GLuint short_types_norm[5] = {
 117    0,
 118    BRW_SURFACEFORMAT_R16_SNORM,
 119    BRW_SURFACEFORMAT_R16G16_SNORM,
 120    BRW_SURFACEFORMAT_R16G16B16_SNORM,
 121    BRW_SURFACEFORMAT_R16G16B16A16_SNORM
 122 };
 123
 124 static GLuint short_types_scale[5] = {
 125    0,
 126    BRW_SURFACEFORMAT_R16_SSCALED,
 127    BRW_SURFACEFORMAT_R16G16_SSCALED,
 128    BRW_SURFACEFORMAT_R16G16B16_SSCALED,
 129    BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
 130 };
 131
 132 static GLuint ubyte_types_norm[5] = {
 133    0,
 134    BRW_SURFACEFORMAT_R8_UNORM,
 135    BRW_SURFACEFORMAT_R8G8_UNORM,
 136    BRW_SURFACEFORMAT_R8G8B8_UNORM,
 137    BRW_SURFACEFORMAT_R8G8B8A8_UNORM
 138 };
 139
 140 static GLuint ubyte_types_scale[5] = {
 141    0,
 142    BRW_SURFACEFORMAT_R8_USCALED,
 143    BRW_SURFACEFORMAT_R8G8_USCALED,
 144    BRW_SURFACEFORMAT_R8G8B8_USCALED,
 145    BRW_SURFACEFORMAT_R8G8B8A8_USCALED
 146 };
 147
 148 static GLuint byte_types_norm[5] = {
 149    0,
 150    BRW_SURFACEFORMAT_R8_SNORM,
 151    BRW_SURFACEFORMAT_R8G8_SNORM,
 152    BRW_SURFACEFORMAT_R8G8B8_SNORM,
 153    BRW_SURFACEFORMAT_R8G8B8A8_SNORM
 154 };
 155
 156 static GLuint byte_types_scale[5] = {
 157    0,
 158    BRW_SURFACEFORMAT_R8_SSCALED,
 159    BRW_SURFACEFORMAT_R8G8_SSCALED,
 160    BRW_SURFACEFORMAT_R8G8B8_SSCALED,
 161    BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
 162 };
 163
 164
 165 /**
 166  * Given vertex array type/size/format/normalized info, return
 167  * the appopriate hardware surface type.
 168  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 169  */
 170 static GLuint get_surface_type( GLenum type, GLuint size,
 171                                 GLenum format, bool normalized )
 172 {
 173    if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
 174       printf("type %s size %d normalized %d\n",
 175                    _mesa_lookup_enum_by_nr(type), size, normalized);
 176
 177    if (normalized) {
 178       switch (type) {
 179       case GL_DOUBLE: return double_types[size];
 180       case GL_FLOAT: return float_types[size];
 181       case GL_HALF_FLOAT: return half_float_types[size];
 182       case GL_INT: return int_types_norm[size];
 183       case GL_SHORT: return short_types_norm[size];
 184       case GL_BYTE: return byte_types_norm[size];
 185       case GL_UNSIGNED_INT: return uint_types_norm[size];
 186       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 187       case GL_UNSIGNED_BYTE:
 188          if (format == GL_BGRA) {
 189             /* See GL_EXT_vertex_array_bgra */
 190             assert(size == 4);
 191             return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
 192          }
 193          else {
 194             return ubyte_types_norm[size];
 195          }
 196       default: assert(0); return 0;
 197       }
 198    }
 199    else {
 200       assert(format == GL_RGBA); /* sanity check */
 201       switch (type) {
 202       case GL_DOUBLE: return double_types[size];
 203       case GL_FLOAT: return float_types[size];
 204       case GL_HALF_FLOAT: return half_float_types[size];
 205       case GL_INT: return int_types_scale[size];
 206       case GL_SHORT: return short_types_scale[size];
 207       case GL_BYTE: return byte_types_scale[size];
 208       case GL_UNSIGNED_INT: return uint_types_scale[size];
 209       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 210       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 211       /* This produces GL_FIXED inputs as values between INT32_MIN and
 212        * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 213        */
 214       case GL_FIXED: return int_types_scale[size];
 215       default: assert(0); return 0;
 216       }
 217    }
 218 }
 219
 220
 221 static GLuint get_size( GLenum type )
 222 {
 223    switch (type) {
 224    case GL_DOUBLE: return sizeof(GLdouble);
 225    case GL_FLOAT: return sizeof(GLfloat);
 226    case GL_HALF_FLOAT: return sizeof(GLhalfARB);
 227    case GL_INT: return sizeof(GLint);
 228    case GL_SHORT: return sizeof(GLshort);
 229    case GL_BYTE: return sizeof(GLbyte);
 230    case GL_UNSIGNED_INT: return sizeof(GLuint);
 231    case GL_UNSIGNED_SHORT: return sizeof(GLushort);
 232    case GL_UNSIGNED_BYTE: return sizeof(GLubyte);
 233    case GL_FIXED: return sizeof(GLuint);
 234    default: assert(0); return 0;
 235    }
 236 }
 237
 238 static GLuint get_index_type(GLenum type)
 239 {
 240    switch (type) {
 241    case GL_UNSIGNED_BYTE:  return BRW_INDEX_BYTE;
 242    case GL_UNSIGNED_SHORT: return BRW_INDEX_WORD;
 243    case GL_UNSIGNED_INT:   return BRW_INDEX_DWORD;
 244    default: assert(0); return 0;
 245    }
 246 }
 247
 248 static void
 249 copy_array_to_vbo_array(struct brw_context *brw,
 250                         struct brw_vertex_element *element,
 251                         int min, int max,
 252                         struct brw_vertex_buffer *buffer,
 253                         GLuint dst_stride)
 254 {
 255    if (min == -1) {
 256       /* If we don't have computed min/max bounds, then this must be a use of
 257        * the current attribute, which has a 0 stride.  Otherwise, we wouldn't
 258        * know what data to upload.
 259        */
 260       assert(element->glarray->StrideB == 0);
 261
 262       intel_upload_data(&brw->intel, element->glarray->Ptr,
 263                         element->element_size,
 264                         element->element_size,
 265                         &buffer->bo, &buffer->offset);
 266
 267       buffer->stride = 0;
 268       return;
 269    }
 270
 271    int src_stride = element->glarray->StrideB;
 272    const unsigned char *src = element->glarray->Ptr + min * src_stride;
 273    int count = max - min + 1;
 274    GLuint size = count * dst_stride;
 275
 276    if (dst_stride == src_stride) {
 277       intel_upload_data(&brw->intel, src, size, dst_stride,
 278                         &buffer->bo, &buffer->offset);
 279    } else {
 280       char * const map = intel_upload_map(&brw->intel, size, dst_stride);
 281       char *dst = map;
 282
 283       while (count--) {
 284          memcpy(dst, src, dst_stride);
 285          src += src_stride;
 286          dst += dst_stride;
 287       }
 288       intel_upload_unmap(&brw->intel, map, size, dst_stride,
 289                          &buffer->bo, &buffer->offset);
 290    }
 291    buffer->stride = dst_stride;
 292 }
 293
 294 static void brw_prepare_vertices(struct brw_context *brw)
 295 {
 296    struct gl_context *ctx = &brw->intel.ctx;
 297    struct intel_context *intel = intel_context(ctx);
 298    /* CACHE_NEW_VS_PROG */
 299    GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
 300    const unsigned char *ptr = NULL;
 301    GLuint interleaved = 0, total_size = 0;
 302    unsigned int min_index = brw->vb.min_index;
 303    unsigned int max_index = brw->vb.max_index;
 304    int delta, i, j;
 305
 306    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 307    GLuint nr_uploads = 0;
 308
 309    /* First build an array of pointers to ve's in vb.inputs_read
 310     */
 311    if (0)
 312       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 313
 314    /* Accumulate the list of enabled arrays. */
 315    brw->vb.nr_enabled = 0;
 316    while (vs_inputs) {
 317       GLuint i = ffs(vs_inputs) - 1;
 318       struct brw_vertex_element *input = &brw->vb.inputs[i];
 319
 320       vs_inputs &= ~(1 << i);
 321       if (input->glarray->Size && get_size(input->glarray->Type))
 322          brw->vb.enabled[brw->vb.nr_enabled++] = input;
 323    }
 324
 325    if (brw->vb.nr_enabled == 0)
 326       return;
 327
 328    if (brw->vb.nr_buffers)
 329       goto validate;
 330
 331    /* XXX: In the rare cases where this happens we fallback all
 332     * the way to software rasterization, although a tnl fallback
 333     * would be sufficient.  I don't know of *any* real world
 334     * cases with > 17 vertex attributes enabled, so it probably
 335     * isn't an issue at this point.
 336     */
 337    if (brw->vb.nr_enabled >= BRW_VEP_MAX) {
 338       intel->Fallback = true; /* boolean, not bitfield */
 339       return;
 340    }
 341
 342    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 343       struct brw_vertex_element *input = brw->vb.enabled[i];
 344       const struct gl_client_array *glarray = input->glarray;
 345       int type_size = get_size(glarray->Type);
 346
 347       input->element_size = type_size * glarray->Size;
 348
 349       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 350          struct intel_buffer_object *intel_buffer =
 351             intel_buffer_object(glarray->BufferObj);
 352          int k;
 353
 354          for (k = 0; k < i; k++) {
 355             const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
 356             if (glarray->BufferObj == other->BufferObj &&
 357                 glarray->StrideB == other->StrideB &&
 358                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 359             {
 360                input->buffer = brw->vb.enabled[k]->buffer;
 361                input->offset = glarray->Ptr - other->Ptr;
 362                break;
 363             }
 364          }
 365          if (k == i) {
 366             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 367
 368             /* Named buffer object: Just reference its contents directly. */
 369             buffer->bo = intel_bufferobj_source(intel,
 370                                                 intel_buffer, type_size,
 371                                                 &buffer->offset);
 372             drm_intel_bo_reference(buffer->bo);
 373             buffer->offset += (uintptr_t)glarray->Ptr;
 374             buffer->stride = glarray->StrideB;
 375
 376             input->buffer = j++;
 377             input->offset = 0;
 378          }
 379
 380          /* This is a common place to reach if the user mistakenly supplies
 381           * a pointer in place of a VBO offset.  If we just let it go through,
 382           * we may end up dereferencing a pointer beyond the bounds of the
 383           * GTT.  We would hope that the VBO's max_index would save us, but
 384           * Mesa appears to hand us min/max values not clipped to the
 385           * array object's _MaxElement, and _MaxElement frequently appears
 386           * to be wrong anyway.
 387           *
 388           * The VBO spec allows application termination in this case, and it's
 389           * probably a service to the poor programmer to do so rather than
 390           * trying to just not render.
 391           */
 392          assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
 393       } else {
 394          /* Queue the buffer object up to be uploaded in the next pass,
 395           * when we've decided if we're doing interleaved or not.
 396           */
 397          if (nr_uploads == 0) {
 398             /* Position array not properly enabled:
 399              */
 400             if (input->attrib == VERT_ATTRIB_POS && glarray->StrideB == 0) {
 401                intel->Fallback = true; /* boolean, not bitfield */
 402                return;
 403             }
 404
 405             interleaved = glarray->StrideB;
 406             ptr = glarray->Ptr;
 407          }
 408          else if (interleaved != glarray->StrideB ||
 409                   (uintptr_t)(glarray->Ptr - ptr) > interleaved)
 410          {
 411             interleaved = 0;
 412          }
 413          else if ((uintptr_t)(glarray->Ptr - ptr) & (type_size -1))
 414          {
 415             /* enforce natural alignment (for doubles) */
 416             interleaved = 0;
 417          }
 418
 419          upload[nr_uploads++] = input;
 420          total_size = ALIGN(total_size, type_size);
 421          total_size += input->element_size;
 422       }
 423    }
 424
 425    /* If we need to upload all the arrays, then we can trim those arrays to
 426     * only the used elements [min_index, max_index] so long as we adjust all
 427     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 428     */
 429    brw->vb.start_vertex_bias = 0;
 430    delta = min_index;
 431    if (nr_uploads == brw->vb.nr_enabled) {
 432       brw->vb.start_vertex_bias = -delta;
 433       delta = 0;
 434    }
 435    if (delta && !brw->intel.intelScreen->relaxed_relocations)
 436       min_index = delta = 0;
 437
 438    /* Handle any arrays to be uploaded. */
 439    if (nr_uploads > 1) {
 440       if (interleaved && interleaved <= 2*total_size) {
 441          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 442          /* All uploads are interleaved, so upload the arrays together as
 443           * interleaved.  First, upload the contents and set up upload[0].
 444           */
 445          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 446                                  buffer, interleaved);
 447          buffer->offset -= delta * interleaved;
 448
 449          for (i = 0; i < nr_uploads; i++) {
 450             /* Then, just point upload[i] at upload[0]'s buffer. */
 451             upload[i]->offset =
 452                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 453             upload[i]->buffer = j;
 454          }
 455          j++;
 456
 457          nr_uploads = 0;
 458       }
 459       else if (total_size < 2048) {
 460          /* Upload non-interleaved arrays into a single interleaved array */
 461          struct brw_vertex_buffer *buffer;
 462          int count = MAX2(max_index - min_index + 1, 1);
 463          int offset;
 464          char *map;
 465
 466          map = intel_upload_map(&brw->intel, total_size * count, total_size);
 467          for (i = offset = 0; i < nr_uploads; i++) {
 468             const unsigned char *src = upload[i]->glarray->Ptr;
 469             int size = upload[i]->element_size;
 470             int stride = upload[i]->glarray->StrideB;
 471             char *dst;
 472             int n;
 473
 474             offset = ALIGN(offset, get_size(upload[i]->glarray->Type));
 475             dst = map + offset;
 476             src += min_index * stride;
 477
 478             for (n = 0; n < count; n++) {
 479                memcpy(dst, src, size);
 480                src += stride;
 481                dst += total_size;
 482             }
 483
 484             upload[i]->offset = offset;
 485             upload[i]->buffer = j;
 486
 487             offset += size;
 488          }
 489          assert(offset == total_size);
 490          buffer = &brw->vb.buffers[j++];
 491          intel_upload_unmap(&brw->intel, map, offset * count, offset,
 492                             &buffer->bo, &buffer->offset);
 493          buffer->stride = offset;
 494          buffer->offset -= delta * offset;
 495
 496          nr_uploads = 0;
 497       }
 498    }
 499    /* Upload non-interleaved arrays */
 500    for (i = 0; i < nr_uploads; i++) {
 501       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 502       copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 503                               buffer, upload[i]->element_size);
 504       buffer->offset -= delta * buffer->stride;
 505       upload[i]->buffer = j++;
 506       upload[i]->offset = 0;
 507    }
 508
 509    /* can we simply extend the current vb? */
 510    if (j == brw->vb.nr_current_buffers) {
 511       int delta = 0;
 512       for (i = 0; i < j; i++) {
 513          int d;
 514
 515          if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle ||
 516              brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride)
 517             break;
 518
 519          d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset;
 520          if (d < 0)
 521             break;
 522          if (i == 0)
 523             delta = d / brw->vb.current_buffers[i].stride;
 524          if (delta * brw->vb.current_buffers[i].stride != d)
 525             break;
 526       }
 527
 528       if (i == j) {
 529          brw->vb.start_vertex_bias += delta;
 530          while (--j >= 0)
 531             drm_intel_bo_unreference(brw->vb.buffers[j].bo);
 532          j = 0;
 533       }
 534    }
 535
 536    brw->vb.nr_buffers = j;
 537
 538 validate:
 539    brw_prepare_query_begin(brw);
 540    for (i = 0; i < brw->vb.nr_buffers; i++) {
 541       brw_add_validated_bo(brw, brw->vb.buffers[i].bo);
 542    }
 543 }
 544
 545 static void brw_emit_vertices(struct brw_context *brw)
 546 {
 547    struct gl_context *ctx = &brw->intel.ctx;
 548    struct intel_context *intel = intel_context(ctx);
 549    GLuint i;
 550
 551    brw_emit_query_begin(brw);
 552
 553    /* If the VS doesn't read any inputs (calculating vertex position from
 554     * a state variable for some reason, for example), emit a single pad
 555     * VERTEX_ELEMENT struct and bail.
 556     *
 557     * The stale VB state stays in place, but they don't do anything unless
 558     * a VE loads from them.
 559     */
 560    if (brw->vb.nr_enabled == 0) {
 561       BEGIN_BATCH(3);
 562       OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | 1);
 563       if (intel->gen >= 6) {
 564          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 565                    GEN6_VE0_VALID |
 566                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 567                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 568       } else {
 569          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 570                    BRW_VE0_VALID |
 571                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 572                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 573       }
 574       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 575                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 576                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 577                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 578       CACHED_BATCH();
 579       return;
 580    }
 581
 582    /* Now emit VB and VEP state packets.
 583     */
 584
 585    if (brw->vb.nr_buffers) {
 586       BEGIN_BATCH(1 + 4*brw->vb.nr_buffers);
 587       OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (4*brw->vb.nr_buffers - 1));
 588       for (i = 0; i < brw->vb.nr_buffers; i++) {
 589          struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 590          uint32_t dw0;
 591
 592          if (intel->gen >= 6) {
 593             dw0 = GEN6_VB0_ACCESS_VERTEXDATA | (i << GEN6_VB0_INDEX_SHIFT);
 594          } else {
 595             dw0 = BRW_VB0_ACCESS_VERTEXDATA | (i << BRW_VB0_INDEX_SHIFT);
 596          }
 597
 598          if (intel->gen >= 7)
 599             dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
 600
 601          OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT));
 602          OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset);
 603          if (intel->gen >= 5) {
 604             OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1);
 605          } else
 606             OUT_BATCH(0);
 607          OUT_BATCH(0); /* Instance data step rate */
 608
 609          brw->vb.current_buffers[i].handle = buffer->bo->handle;
 610          brw->vb.current_buffers[i].offset = buffer->offset;
 611          brw->vb.current_buffers[i].stride = buffer->stride;
 612       }
 613       brw->vb.nr_current_buffers = i;
 614       ADVANCE_BATCH();
 615    }
 616
 617    BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
 618    OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (2*brw->vb.nr_enabled - 1));
 619    for (i = 0; i < brw->vb.nr_enabled; i++) {
 620       struct brw_vertex_element *input = brw->vb.enabled[i];
 621       uint32_t format = get_surface_type(input->glarray->Type,
 622                                          input->glarray->Size,
 623                                          input->glarray->Format,
 624                                          input->glarray->Normalized);
 625       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 626       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 627       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 628       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 629
 630       switch (input->glarray->Size) {
 631       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 632       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 633       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 634       case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
 635          break;
 636       }
 637
 638       if (intel->gen >= 6) {
 639          OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
 640                    GEN6_VE0_VALID |
 641                    (format << BRW_VE0_FORMAT_SHIFT) |
 642                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 643       } else {
 644          OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) |
 645                    BRW_VE0_VALID |
 646                    (format << BRW_VE0_FORMAT_SHIFT) |
 647                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 648       }
 649
 650       if (intel->gen >= 5)
 651           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 652                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 653                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 654                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 655       else
 656           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 657                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 658                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 659                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 660                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 661    }
 662    CACHED_BATCH();
 663 }
 664
 665 const struct brw_tracked_state brw_vertices = {
 666    .dirty = {
 667       .mesa = 0,
 668       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 669       .cache = CACHE_NEW_VS_PROG,
 670    },
 671    .prepare = brw_prepare_vertices,
 672    .emit = brw_emit_vertices,
 673 };
 674
 675 static void brw_prepare_indices(struct brw_context *brw)
 676 {
 677    struct gl_context *ctx = &brw->intel.ctx;
 678    struct intel_context *intel = &brw->intel;
 679    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 680    GLuint ib_size;
 681    drm_intel_bo *bo = NULL;
 682    struct gl_buffer_object *bufferobj;
 683    GLuint offset;
 684    GLuint ib_type_size;
 685
 686    if (index_buffer == NULL)
 687       return;
 688
 689    ib_type_size = get_size(index_buffer->type);
 690    ib_size = ib_type_size * index_buffer->count;
 691    bufferobj = index_buffer->obj;
 692
 693    /* Turn into a proper VBO:
 694     */
 695    if (!_mesa_is_bufferobj(bufferobj)) {
 696
 697       /* Get new bufferobj, offset:
 698        */
 699       intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size,
 700                         &bo, &offset);
 701       brw->ib.start_vertex_offset = offset / ib_type_size;
 702    } else {
 703       offset = (GLuint) (unsigned long) index_buffer->ptr;
 704
 705       /* If the index buffer isn't aligned to its element size, we have to
 706        * rebase it into a temporary.
 707        */
 708        if ((get_size(index_buffer->type) - 1) & offset) {
 709            GLubyte *map = ctx->Driver.MapBufferRange(ctx,
 710                                                      offset,
 711                                                      ib_size,
 712                                                      GL_MAP_WRITE_BIT,
 713                                                      bufferobj);
 714
 715            intel_upload_data(&brw->intel, map, ib_size, ib_type_size,
 716                              &bo, &offset);
 717            brw->ib.start_vertex_offset = offset / ib_type_size;
 718
 719            ctx->Driver.UnmapBuffer(ctx, bufferobj);
 720        } else {
 721           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 722            * the index buffer state when we're just moving the start index
 723            * of our drawing.
 724            */
 725           brw->ib.start_vertex_offset = offset / ib_type_size;
 726
 727           bo = intel_bufferobj_source(intel,
 728                                       intel_buffer_object(bufferobj),
 729                                       ib_type_size,
 730                                       &offset);
 731           drm_intel_bo_reference(bo);
 732
 733           brw->ib.start_vertex_offset += offset / ib_type_size;
 734        }
 735    }
 736
 737    if (brw->ib.bo != bo) {
 738       drm_intel_bo_unreference(brw->ib.bo);
 739       brw->ib.bo = bo;
 740
 741       brw_add_validated_bo(brw, brw->ib.bo);
 742       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 743    } else {
 744       drm_intel_bo_unreference(bo);
 745    }
 746
 747    if (index_buffer->type != brw->ib.type) {
 748       brw->ib.type = index_buffer->type;
 749       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 750    }
 751 }
 752
 753 const struct brw_tracked_state brw_indices = {
 754    .dirty = {
 755       .mesa = 0,
 756       .brw = BRW_NEW_INDICES,
 757       .cache = 0,
 758    },
 759    .prepare = brw_prepare_indices,
 760 };
 761
 762 static void brw_emit_index_buffer(struct brw_context *brw)
 763 {
 764    struct intel_context *intel = &brw->intel;
 765    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 766
 767    if (index_buffer == NULL)
 768       return;
 769
 770    BEGIN_BATCH(3);
 771    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 772              /* cut index enable << 10 */
 773              get_index_type(index_buffer->type) << 8 |
 774              1);
 775    OUT_RELOC(brw->ib.bo,
 776              I915_GEM_DOMAIN_VERTEX, 0,
 777              0);
 778    OUT_RELOC(brw->ib.bo,
 779              I915_GEM_DOMAIN_VERTEX, 0,
 780              brw->ib.bo->size - 1);
 781    ADVANCE_BATCH();
 782 }
 783
 784 const struct brw_tracked_state brw_index_buffer = {
 785    .dirty = {
 786       .mesa = 0,
 787       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 788       .cache = 0,
 789    },
 790    .emit = brw_emit_index_buffer,
 791 };