src/mesa/drivers/dri/i965/brw_draw_upload.c

   1 /*
   2  * Copyright 2003 VMware, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sublicense, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the
  14  * next paragraph) shall be included in all copies or substantial portions
  15  * of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  20  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  21  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  22  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  23  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  */
  25
  26 #include "main/bufferobj.h"
  27 #include "main/context.h"
  28 #include "main/enums.h"
  29 #include "main/macros.h"
  30 #include "main/glformats.h"
  31
  32 #include "brw_draw.h"
  33 #include "brw_defines.h"
  34 #include "brw_context.h"
  35 #include "brw_state.h"
  36
  37 #include "intel_batchbuffer.h"
  38 #include "intel_buffer_objects.h"
  39
  40 static const GLuint double_types_float[5] = {
  41    0,
  42    ISL_FORMAT_R64_FLOAT,
  43    ISL_FORMAT_R64G64_FLOAT,
  44    ISL_FORMAT_R64G64B64_FLOAT,
  45    ISL_FORMAT_R64G64B64A64_FLOAT
  46 };
  47
  48 static const GLuint double_types_passthru[5] = {
  49    0,
  50    ISL_FORMAT_R64_PASSTHRU,
  51    ISL_FORMAT_R64G64_PASSTHRU,
  52    ISL_FORMAT_R64G64B64_PASSTHRU,
  53    ISL_FORMAT_R64G64B64A64_PASSTHRU
  54 };
  55
  56 static const GLuint float_types[5] = {
  57    0,
  58    ISL_FORMAT_R32_FLOAT,
  59    ISL_FORMAT_R32G32_FLOAT,
  60    ISL_FORMAT_R32G32B32_FLOAT,
  61    ISL_FORMAT_R32G32B32A32_FLOAT
  62 };
  63
  64 static const GLuint half_float_types[5] = {
  65    0,
  66    ISL_FORMAT_R16_FLOAT,
  67    ISL_FORMAT_R16G16_FLOAT,
  68    ISL_FORMAT_R16G16B16_FLOAT,
  69    ISL_FORMAT_R16G16B16A16_FLOAT
  70 };
  71
  72 static const GLuint fixed_point_types[5] = {
  73    0,
  74    ISL_FORMAT_R32_SFIXED,
  75    ISL_FORMAT_R32G32_SFIXED,
  76    ISL_FORMAT_R32G32B32_SFIXED,
  77    ISL_FORMAT_R32G32B32A32_SFIXED,
  78 };
  79
  80 static const GLuint uint_types_direct[5] = {
  81    0,
  82    ISL_FORMAT_R32_UINT,
  83    ISL_FORMAT_R32G32_UINT,
  84    ISL_FORMAT_R32G32B32_UINT,
  85    ISL_FORMAT_R32G32B32A32_UINT
  86 };
  87
  88 static const GLuint uint_types_norm[5] = {
  89    0,
  90    ISL_FORMAT_R32_UNORM,
  91    ISL_FORMAT_R32G32_UNORM,
  92    ISL_FORMAT_R32G32B32_UNORM,
  93    ISL_FORMAT_R32G32B32A32_UNORM
  94 };
  95
  96 static const GLuint uint_types_scale[5] = {
  97    0,
  98    ISL_FORMAT_R32_USCALED,
  99    ISL_FORMAT_R32G32_USCALED,
 100    ISL_FORMAT_R32G32B32_USCALED,
 101    ISL_FORMAT_R32G32B32A32_USCALED
 102 };
 103
 104 static const GLuint int_types_direct[5] = {
 105    0,
 106    ISL_FORMAT_R32_SINT,
 107    ISL_FORMAT_R32G32_SINT,
 108    ISL_FORMAT_R32G32B32_SINT,
 109    ISL_FORMAT_R32G32B32A32_SINT
 110 };
 111
 112 static const GLuint int_types_norm[5] = {
 113    0,
 114    ISL_FORMAT_R32_SNORM,
 115    ISL_FORMAT_R32G32_SNORM,
 116    ISL_FORMAT_R32G32B32_SNORM,
 117    ISL_FORMAT_R32G32B32A32_SNORM
 118 };
 119
 120 static const GLuint int_types_scale[5] = {
 121    0,
 122    ISL_FORMAT_R32_SSCALED,
 123    ISL_FORMAT_R32G32_SSCALED,
 124    ISL_FORMAT_R32G32B32_SSCALED,
 125    ISL_FORMAT_R32G32B32A32_SSCALED
 126 };
 127
 128 static const GLuint ushort_types_direct[5] = {
 129    0,
 130    ISL_FORMAT_R16_UINT,
 131    ISL_FORMAT_R16G16_UINT,
 132    ISL_FORMAT_R16G16B16_UINT,
 133    ISL_FORMAT_R16G16B16A16_UINT
 134 };
 135
 136 static const GLuint ushort_types_norm[5] = {
 137    0,
 138    ISL_FORMAT_R16_UNORM,
 139    ISL_FORMAT_R16G16_UNORM,
 140    ISL_FORMAT_R16G16B16_UNORM,
 141    ISL_FORMAT_R16G16B16A16_UNORM
 142 };
 143
 144 static const GLuint ushort_types_scale[5] = {
 145    0,
 146    ISL_FORMAT_R16_USCALED,
 147    ISL_FORMAT_R16G16_USCALED,
 148    ISL_FORMAT_R16G16B16_USCALED,
 149    ISL_FORMAT_R16G16B16A16_USCALED
 150 };
 151
 152 static const GLuint short_types_direct[5] = {
 153    0,
 154    ISL_FORMAT_R16_SINT,
 155    ISL_FORMAT_R16G16_SINT,
 156    ISL_FORMAT_R16G16B16_SINT,
 157    ISL_FORMAT_R16G16B16A16_SINT
 158 };
 159
 160 static const GLuint short_types_norm[5] = {
 161    0,
 162    ISL_FORMAT_R16_SNORM,
 163    ISL_FORMAT_R16G16_SNORM,
 164    ISL_FORMAT_R16G16B16_SNORM,
 165    ISL_FORMAT_R16G16B16A16_SNORM
 166 };
 167
 168 static const GLuint short_types_scale[5] = {
 169    0,
 170    ISL_FORMAT_R16_SSCALED,
 171    ISL_FORMAT_R16G16_SSCALED,
 172    ISL_FORMAT_R16G16B16_SSCALED,
 173    ISL_FORMAT_R16G16B16A16_SSCALED
 174 };
 175
 176 static const GLuint ubyte_types_direct[5] = {
 177    0,
 178    ISL_FORMAT_R8_UINT,
 179    ISL_FORMAT_R8G8_UINT,
 180    ISL_FORMAT_R8G8B8_UINT,
 181    ISL_FORMAT_R8G8B8A8_UINT
 182 };
 183
 184 static const GLuint ubyte_types_norm[5] = {
 185    0,
 186    ISL_FORMAT_R8_UNORM,
 187    ISL_FORMAT_R8G8_UNORM,
 188    ISL_FORMAT_R8G8B8_UNORM,
 189    ISL_FORMAT_R8G8B8A8_UNORM
 190 };
 191
 192 static const GLuint ubyte_types_scale[5] = {
 193    0,
 194    ISL_FORMAT_R8_USCALED,
 195    ISL_FORMAT_R8G8_USCALED,
 196    ISL_FORMAT_R8G8B8_USCALED,
 197    ISL_FORMAT_R8G8B8A8_USCALED
 198 };
 199
 200 static const GLuint byte_types_direct[5] = {
 201    0,
 202    ISL_FORMAT_R8_SINT,
 203    ISL_FORMAT_R8G8_SINT,
 204    ISL_FORMAT_R8G8B8_SINT,
 205    ISL_FORMAT_R8G8B8A8_SINT
 206 };
 207
 208 static const GLuint byte_types_norm[5] = {
 209    0,
 210    ISL_FORMAT_R8_SNORM,
 211    ISL_FORMAT_R8G8_SNORM,
 212    ISL_FORMAT_R8G8B8_SNORM,
 213    ISL_FORMAT_R8G8B8A8_SNORM
 214 };
 215
 216 static const GLuint byte_types_scale[5] = {
 217    0,
 218    ISL_FORMAT_R8_SSCALED,
 219    ISL_FORMAT_R8G8_SSCALED,
 220    ISL_FORMAT_R8G8B8_SSCALED,
 221    ISL_FORMAT_R8G8B8A8_SSCALED
 222 };
 223
 224 static GLuint
 225 double_types(struct brw_context *brw,
 226              int size,
 227              GLboolean doubles)
 228 {
 229    /* From the BDW PRM, Volume 2d, page 588 (VERTEX_ELEMENT_STATE):
 230     * "When SourceElementFormat is set to one of the *64*_PASSTHRU formats,
 231     * 64-bit components are stored in the URB without any conversion."
 232     * Also included on BDW PRM, Volume 7, page 470, table "Source Element
 233     * Formats Supported in VF Unit"
 234     *
 235     * Previous PRMs don't include those references, so for gen7 we can't use
 236     * PASSTHRU formats directly. But in any case, we prefer to return passthru
 237     * even in that case, because that reflects what we want to achieve, even
 238     * if we would need to workaround on gen < 8.
 239     */
 240    return (doubles
 241            ? double_types_passthru[size]
 242            : double_types_float[size]);
 243 }
 244
 245 /**
 246  * Given vertex array type/size/format/normalized info, return
 247  * the appopriate hardware surface type.
 248  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 249  */
 250 unsigned
 251 brw_get_vertex_surface_type(struct brw_context *brw,
 252                             const struct gl_vertex_array *glarray)
 253 {
 254    int size = glarray->Size;
 255    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 256    const bool is_ivybridge_or_older =
 257       devinfo->gen <= 7 && !devinfo->is_baytrail && !devinfo->is_haswell;
 258
 259    if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
 260       fprintf(stderr, "type %s size %d normalized %d\n",
 261               _mesa_enum_to_string(glarray->Type),
 262               glarray->Size, glarray->Normalized);
 263
 264    if (glarray->Integer) {
 265       assert(glarray->Format == GL_RGBA); /* sanity check */
 266       switch (glarray->Type) {
 267       case GL_INT: return int_types_direct[size];
 268       case GL_SHORT:
 269          if (is_ivybridge_or_older && size == 3)
 270             return short_types_direct[4];
 271          else
 272             return short_types_direct[size];
 273       case GL_BYTE:
 274          if (is_ivybridge_or_older && size == 3)
 275             return byte_types_direct[4];
 276          else
 277             return byte_types_direct[size];
 278       case GL_UNSIGNED_INT: return uint_types_direct[size];
 279       case GL_UNSIGNED_SHORT:
 280          if (is_ivybridge_or_older && size == 3)
 281             return ushort_types_direct[4];
 282          else
 283             return ushort_types_direct[size];
 284       case GL_UNSIGNED_BYTE:
 285          if (is_ivybridge_or_older && size == 3)
 286             return ubyte_types_direct[4];
 287          else
 288             return ubyte_types_direct[size];
 289       default: unreachable("not reached");
 290       }
 291    } else if (glarray->Type == GL_UNSIGNED_INT_10F_11F_11F_REV) {
 292       return ISL_FORMAT_R11G11B10_FLOAT;
 293    } else if (glarray->Normalized) {
 294       switch (glarray->Type) {
 295       case GL_DOUBLE: return double_types(brw, size, glarray->Doubles);
 296       case GL_FLOAT: return float_types[size];
 297       case GL_HALF_FLOAT:
 298       case GL_HALF_FLOAT_OES:
 299          if (devinfo->gen < 6 && size == 3)
 300             return half_float_types[4];
 301          else
 302             return half_float_types[size];
 303       case GL_INT: return int_types_norm[size];
 304       case GL_SHORT: return short_types_norm[size];
 305       case GL_BYTE: return byte_types_norm[size];
 306       case GL_UNSIGNED_INT: return uint_types_norm[size];
 307       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 308       case GL_UNSIGNED_BYTE:
 309          if (glarray->Format == GL_BGRA) {
 310             /* See GL_EXT_vertex_array_bgra */
 311             assert(size == 4);
 312             return ISL_FORMAT_B8G8R8A8_UNORM;
 313          }
 314          else {
 315             return ubyte_types_norm[size];
 316          }
 317       case GL_FIXED:
 318          if (devinfo->gen >= 8 || devinfo->is_haswell)
 319             return fixed_point_types[size];
 320
 321          /* This produces GL_FIXED inputs as values between INT32_MIN and
 322           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 323           */
 324          return int_types_scale[size];
 325       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 326        * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
 327        * like to use here, so upload everything as UINT and fix
 328        * it in the shader
 329        */
 330       case GL_INT_2_10_10_10_REV:
 331          assert(size == 4);
 332          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 333             return glarray->Format == GL_BGRA
 334                ? ISL_FORMAT_B10G10R10A2_SNORM
 335                : ISL_FORMAT_R10G10B10A2_SNORM;
 336          }
 337          return ISL_FORMAT_R10G10B10A2_UINT;
 338       case GL_UNSIGNED_INT_2_10_10_10_REV:
 339          assert(size == 4);
 340          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 341             return glarray->Format == GL_BGRA
 342                ? ISL_FORMAT_B10G10R10A2_UNORM
 343                : ISL_FORMAT_R10G10B10A2_UNORM;
 344          }
 345          return ISL_FORMAT_R10G10B10A2_UINT;
 346       default: unreachable("not reached");
 347       }
 348    }
 349    else {
 350       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 351        * W/A: the hardware doesn't really support the formats we'd
 352        * like to use here, so upload everything as UINT and fix
 353        * it in the shader
 354        */
 355       if (glarray->Type == GL_INT_2_10_10_10_REV) {
 356          assert(size == 4);
 357          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 358             return glarray->Format == GL_BGRA
 359                ? ISL_FORMAT_B10G10R10A2_SSCALED
 360                : ISL_FORMAT_R10G10B10A2_SSCALED;
 361          }
 362          return ISL_FORMAT_R10G10B10A2_UINT;
 363       } else if (glarray->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 364          assert(size == 4);
 365          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 366             return glarray->Format == GL_BGRA
 367                ? ISL_FORMAT_B10G10R10A2_USCALED
 368                : ISL_FORMAT_R10G10B10A2_USCALED;
 369          }
 370          return ISL_FORMAT_R10G10B10A2_UINT;
 371       }
 372       assert(glarray->Format == GL_RGBA); /* sanity check */
 373       switch (glarray->Type) {
 374       case GL_DOUBLE: return double_types(brw, size, glarray->Doubles);
 375       case GL_FLOAT: return float_types[size];
 376       case GL_HALF_FLOAT:
 377       case GL_HALF_FLOAT_OES:
 378          if (devinfo->gen < 6 && size == 3)
 379             return half_float_types[4];
 380          else
 381             return half_float_types[size];
 382       case GL_INT: return int_types_scale[size];
 383       case GL_SHORT: return short_types_scale[size];
 384       case GL_BYTE: return byte_types_scale[size];
 385       case GL_UNSIGNED_INT: return uint_types_scale[size];
 386       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 387       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 388       case GL_FIXED:
 389          if (devinfo->gen >= 8 || devinfo->is_haswell)
 390             return fixed_point_types[size];
 391
 392          /* This produces GL_FIXED inputs as values between INT32_MIN and
 393           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 394           */
 395          return int_types_scale[size];
 396       default: unreachable("not reached");
 397       }
 398    }
 399 }
 400
 401 static void
 402 copy_array_to_vbo_array(struct brw_context *brw,
 403                         struct brw_vertex_element *element,
 404                         int min, int max,
 405                         struct brw_vertex_buffer *buffer,
 406                         GLuint dst_stride)
 407 {
 408    const int src_stride = element->glarray->StrideB;
 409
 410    /* If the source stride is zero, we just want to upload the current
 411     * attribute once and set the buffer's stride to 0.  There's no need
 412     * to replicate it out.
 413     */
 414    if (src_stride == 0) {
 415       intel_upload_data(brw, element->glarray->Ptr,
 416                         element->glarray->_ElementSize,
 417                         element->glarray->_ElementSize,
 418                         &buffer->bo, &buffer->offset);
 419
 420       buffer->stride = 0;
 421       buffer->size = element->glarray->_ElementSize;
 422       return;
 423    }
 424
 425    const unsigned char *src = element->glarray->Ptr + min * src_stride;
 426    int count = max - min + 1;
 427    GLuint size = count * dst_stride;
 428    uint8_t *dst = intel_upload_space(brw, size, dst_stride,
 429                                      &buffer->bo, &buffer->offset);
 430
 431    /* The GL 4.5 spec says:
 432     *      "If any enabled array’s buffer binding is zero when DrawArrays or
 433     *      one of the other drawing commands defined in section 10.4 is called,
 434     *      the result is undefined."
 435     *
 436     * In this case, let's the dst with undefined values
 437     */
 438    if (src != NULL) {
 439       if (dst_stride == src_stride) {
 440          memcpy(dst, src, size);
 441       } else {
 442          while (count--) {
 443             memcpy(dst, src, dst_stride);
 444             src += src_stride;
 445             dst += dst_stride;
 446          }
 447       }
 448    }
 449    buffer->stride = dst_stride;
 450    buffer->size = size;
 451 }
 452
 453 void
 454 brw_prepare_vertices(struct brw_context *brw)
 455 {
 456    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 457    struct gl_context *ctx = &brw->ctx;
 458    /* BRW_NEW_VS_PROG_DATA */
 459    const struct brw_vs_prog_data *vs_prog_data =
 460       brw_vs_prog_data(brw->vs.base.prog_data);
 461    GLbitfield64 vs_inputs = vs_prog_data->inputs_read;
 462    const unsigned char *ptr = NULL;
 463    GLuint interleaved = 0;
 464    unsigned int min_index = brw->vb.min_index + brw->basevertex;
 465    unsigned int max_index = brw->vb.max_index + brw->basevertex;
 466    unsigned i;
 467    int delta, j;
 468
 469    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 470    GLuint nr_uploads = 0;
 471
 472    /* _NEW_POLYGON
 473     *
 474     * On gen6+, edge flags don't end up in the VUE (either in or out of the
 475     * VS).  Instead, they're uploaded as the last vertex element, and the data
 476     * is passed sideband through the fixed function units.  So, we need to
 477     * prepare the vertex buffer for it, but it's not present in inputs_read.
 478     */
 479    if (devinfo->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
 480                            ctx->Polygon.BackMode != GL_FILL)) {
 481       vs_inputs |= VERT_BIT_EDGEFLAG;
 482    }
 483
 484    if (0)
 485       fprintf(stderr, "%s %d..%d\n", __func__, min_index, max_index);
 486
 487    /* Accumulate the list of enabled arrays. */
 488    brw->vb.nr_enabled = 0;
 489    while (vs_inputs) {
 490       GLuint first = ffsll(vs_inputs) - 1;
 491       assert (first < 64);
 492       GLuint index =
 493          first - DIV_ROUND_UP(_mesa_bitcount_64(vs_prog_data->double_inputs_read &
 494                                                 BITFIELD64_MASK(first)), 2);
 495       struct brw_vertex_element *input = &brw->vb.inputs[index];
 496       input->is_dual_slot = (vs_prog_data->double_inputs_read & BITFIELD64_BIT(first)) != 0;
 497       vs_inputs &= ~BITFIELD64_BIT(first);
 498       if (input->is_dual_slot)
 499          vs_inputs &= ~BITFIELD64_BIT(first + 1);
 500       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 501    }
 502
 503    if (brw->vb.nr_enabled == 0)
 504       return;
 505
 506    if (brw->vb.nr_buffers)
 507       return;
 508
 509    /* The range of data in a given buffer represented as [min, max) */
 510    struct intel_buffer_object *enabled_buffer[VERT_ATTRIB_MAX];
 511    uint32_t buffer_range_start[VERT_ATTRIB_MAX];
 512    uint32_t buffer_range_end[VERT_ATTRIB_MAX];
 513
 514    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 515       struct brw_vertex_element *input = brw->vb.enabled[i];
 516       const struct gl_vertex_array *glarray = input->glarray;
 517
 518       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 519          struct intel_buffer_object *intel_buffer =
 520             intel_buffer_object(glarray->BufferObj);
 521
 522          const uint32_t offset = (uintptr_t)glarray->Ptr;
 523
 524          /* Start with the worst case */
 525          uint32_t start = 0;
 526          uint32_t range = intel_buffer->Base.Size;
 527          if (glarray->InstanceDivisor) {
 528             if (brw->num_instances) {
 529                start = offset + glarray->StrideB * brw->baseinstance;
 530                range = (glarray->StrideB * ((brw->num_instances - 1) /
 531                                             glarray->InstanceDivisor) +
 532                         glarray->_ElementSize);
 533             }
 534          } else {
 535             if (brw->vb.index_bounds_valid) {
 536                start = offset + min_index * glarray->StrideB;
 537                range = (glarray->StrideB * (max_index - min_index) +
 538                         glarray->_ElementSize);
 539             }
 540          }
 541
 542          /* If we have a VB set to be uploaded for this buffer object
 543           * already, reuse that VB state so that we emit fewer
 544           * relocations.
 545           */
 546          unsigned k;
 547          for (k = 0; k < i; k++) {
 548             const struct gl_vertex_array *other = brw->vb.enabled[k]->glarray;
 549             if (glarray->BufferObj == other->BufferObj &&
 550                 glarray->StrideB == other->StrideB &&
 551                 glarray->InstanceDivisor == other->InstanceDivisor &&
 552                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 553             {
 554                input->buffer = brw->vb.enabled[k]->buffer;
 555                input->offset = glarray->Ptr - other->Ptr;
 556
 557                buffer_range_start[input->buffer] =
 558                   MIN2(buffer_range_start[input->buffer], start);
 559                buffer_range_end[input->buffer] =
 560                   MAX2(buffer_range_end[input->buffer], start + range);
 561                break;
 562             }
 563          }
 564          if (k == i) {
 565             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 566
 567             /* Named buffer object: Just reference its contents directly. */
 568             buffer->offset = offset;
 569             buffer->stride = glarray->StrideB;
 570             buffer->step_rate = glarray->InstanceDivisor;
 571             buffer->size = glarray->BufferObj->Size - offset;
 572
 573             enabled_buffer[j] = intel_buffer;
 574             buffer_range_start[j] = start;
 575             buffer_range_end[j] = start + range;
 576
 577             input->buffer = j++;
 578             input->offset = 0;
 579          }
 580       } else {
 581          /* Queue the buffer object up to be uploaded in the next pass,
 582           * when we've decided if we're doing interleaved or not.
 583           */
 584          if (nr_uploads == 0) {
 585             interleaved = glarray->StrideB;
 586             ptr = glarray->Ptr;
 587          }
 588          else if (interleaved != glarray->StrideB ||
 589                   glarray->InstanceDivisor != 0 ||
 590                   glarray->Ptr < ptr ||
 591                   (uintptr_t)(glarray->Ptr - ptr) + glarray->_ElementSize > interleaved)
 592          {
 593             /* If our stride is different from the first attribute's stride,
 594              * or if we are using an instance divisor or if the first
 595              * attribute's stride didn't cover our element, disable the
 596              * interleaved upload optimization.  The second case can most
 597              * commonly occur in cases where there is a single vertex and, for
 598              * example, the data is stored on the application's stack.
 599              *
 600              * NOTE: This will also disable the optimization in cases where
 601              * the data is in a different order than the array indices.
 602              * Something like:
 603              *
 604              *     float data[...];
 605              *     glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
 606              *     glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
 607              */
 608             interleaved = 0;
 609          }
 610
 611          upload[nr_uploads++] = input;
 612       }
 613    }
 614
 615    /* Now that we've set up all of the buffers, we walk through and reference
 616     * each of them.  We do this late so that we get the right size in each
 617     * buffer and don't reference too little data.
 618     */
 619    for (i = 0; i < j; i++) {
 620       struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 621       if (buffer->bo)
 622          continue;
 623
 624       const uint32_t start = buffer_range_start[i];
 625       const uint32_t range = buffer_range_end[i] - buffer_range_start[i];
 626
 627       buffer->bo = intel_bufferobj_buffer(brw, enabled_buffer[i], start,
 628                                           range, false);
 629       brw_bo_reference(buffer->bo);
 630    }
 631
 632    /* If we need to upload all the arrays, then we can trim those arrays to
 633     * only the used elements [min_index, max_index] so long as we adjust all
 634     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 635     */
 636    brw->vb.start_vertex_bias = 0;
 637    delta = min_index;
 638    if (nr_uploads == brw->vb.nr_enabled) {
 639       brw->vb.start_vertex_bias = -delta;
 640       delta = 0;
 641    }
 642
 643    /* Handle any arrays to be uploaded. */
 644    if (nr_uploads > 1) {
 645       if (interleaved) {
 646          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 647          /* All uploads are interleaved, so upload the arrays together as
 648           * interleaved.  First, upload the contents and set up upload[0].
 649           */
 650          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 651                                  buffer, interleaved);
 652          buffer->offset -= delta * interleaved;
 653          buffer->size += delta * interleaved;
 654          buffer->step_rate = 0;
 655
 656          for (i = 0; i < nr_uploads; i++) {
 657             /* Then, just point upload[i] at upload[0]'s buffer. */
 658             upload[i]->offset =
 659                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 660             upload[i]->buffer = j;
 661          }
 662          j++;
 663
 664          nr_uploads = 0;
 665       }
 666    }
 667    /* Upload non-interleaved arrays */
 668    for (i = 0; i < nr_uploads; i++) {
 669       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 670       if (upload[i]->glarray->InstanceDivisor == 0) {
 671          copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 672                                  buffer, upload[i]->glarray->_ElementSize);
 673       } else {
 674          /* This is an instanced attribute, since its InstanceDivisor
 675           * is not zero. Therefore, its data will be stepped after the
 676           * instanced draw has been run InstanceDivisor times.
 677           */
 678          uint32_t instanced_attr_max_index =
 679             (brw->num_instances - 1) / upload[i]->glarray->InstanceDivisor;
 680          copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
 681                                  buffer, upload[i]->glarray->_ElementSize);
 682       }
 683       buffer->offset -= delta * buffer->stride;
 684       buffer->size += delta * buffer->stride;
 685       buffer->step_rate = upload[i]->glarray->InstanceDivisor;
 686       upload[i]->buffer = j++;
 687       upload[i]->offset = 0;
 688    }
 689
 690    brw->vb.nr_buffers = j;
 691 }
 692
 693 void
 694 brw_prepare_shader_draw_parameters(struct brw_context *brw)
 695 {
 696    const struct brw_vs_prog_data *vs_prog_data =
 697       brw_vs_prog_data(brw->vs.base.prog_data);
 698
 699    /* For non-indirect draws, upload gl_BaseVertex. */
 700    if ((vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance) &&
 701        brw->draw.draw_params_bo == NULL) {
 702       intel_upload_data(brw, &brw->draw.params, sizeof(brw->draw.params), 4,
 703                         &brw->draw.draw_params_bo,
 704                         &brw->draw.draw_params_offset);
 705    }
 706
 707    if (vs_prog_data->uses_drawid) {
 708       intel_upload_data(brw, &brw->draw.gl_drawid, sizeof(brw->draw.gl_drawid), 4,
 709                         &brw->draw.draw_id_bo,
 710                         &brw->draw.draw_id_offset);
 711    }
 712 }
 713
 714 static void
 715 brw_upload_indices(struct brw_context *brw)
 716 {
 717    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 718    GLuint ib_size;
 719    struct brw_bo *old_bo = brw->ib.bo;
 720    struct gl_buffer_object *bufferobj;
 721    GLuint offset;
 722    GLuint ib_type_size;
 723
 724    if (index_buffer == NULL)
 725       return;
 726
 727    ib_type_size = index_buffer->index_size;
 728    ib_size = index_buffer->count ? ib_type_size * index_buffer->count :
 729                                    index_buffer->obj->Size;
 730    bufferobj = index_buffer->obj;
 731
 732    /* Turn into a proper VBO:
 733     */
 734    if (!_mesa_is_bufferobj(bufferobj)) {
 735       /* Get new bufferobj, offset:
 736        */
 737       intel_upload_data(brw, index_buffer->ptr, ib_size, ib_type_size,
 738                         &brw->ib.bo, &offset);
 739       brw->ib.size = brw->ib.bo->size;
 740    } else {
 741       offset = (GLuint) (unsigned long) index_buffer->ptr;
 742
 743       struct brw_bo *bo =
 744          intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 745                                 offset, ib_size, false);
 746       if (bo != brw->ib.bo) {
 747          brw_bo_unreference(brw->ib.bo);
 748          brw->ib.bo = bo;
 749          brw->ib.size = bufferobj->Size;
 750          brw_bo_reference(bo);
 751       }
 752    }
 753
 754    /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
 755     * the index buffer state when we're just moving the start index
 756     * of our drawing.
 757     */
 758    brw->ib.start_vertex_offset = offset / ib_type_size;
 759
 760    if (brw->ib.bo != old_bo)
 761       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 762
 763    if (index_buffer->index_size != brw->ib.index_size) {
 764       brw->ib.index_size = index_buffer->index_size;
 765       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 766    }
 767 }
 768
 769 const struct brw_tracked_state brw_indices = {
 770    .dirty = {
 771       .mesa = 0,
 772       .brw = BRW_NEW_BLORP |
 773              BRW_NEW_INDICES,
 774    },
 775    .emit = brw_upload_indices,
 776 };