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 bool is_ivybridge_or_older =
 256       brw->gen <= 7 && !brw->is_baytrail && !brw->is_haswell;
 257
 258    if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
 259       fprintf(stderr, "type %s size %d normalized %d\n",
 260               _mesa_enum_to_string(glarray->Type),
 261               glarray->Size, glarray->Normalized);
 262
 263    if (glarray->Integer) {
 264       assert(glarray->Format == GL_RGBA); /* sanity check */
 265       switch (glarray->Type) {
 266       case GL_INT: return int_types_direct[size];
 267       case GL_SHORT:
 268          if (is_ivybridge_or_older && size == 3)
 269             return short_types_direct[4];
 270          else
 271             return short_types_direct[size];
 272       case GL_BYTE:
 273          if (is_ivybridge_or_older && size == 3)
 274             return byte_types_direct[4];
 275          else
 276             return byte_types_direct[size];
 277       case GL_UNSIGNED_INT: return uint_types_direct[size];
 278       case GL_UNSIGNED_SHORT:
 279          if (is_ivybridge_or_older && size == 3)
 280             return ushort_types_direct[4];
 281          else
 282             return ushort_types_direct[size];
 283       case GL_UNSIGNED_BYTE:
 284          if (is_ivybridge_or_older && size == 3)
 285             return ubyte_types_direct[4];
 286          else
 287             return ubyte_types_direct[size];
 288       default: unreachable("not reached");
 289       }
 290    } else if (glarray->Type == GL_UNSIGNED_INT_10F_11F_11F_REV) {
 291       return ISL_FORMAT_R11G11B10_FLOAT;
 292    } else if (glarray->Normalized) {
 293       switch (glarray->Type) {
 294       case GL_DOUBLE: return double_types(brw, size, glarray->Doubles);
 295       case GL_FLOAT: return float_types[size];
 296       case GL_HALF_FLOAT:
 297       case GL_HALF_FLOAT_OES:
 298          if (brw->gen < 6 && size == 3)
 299             return half_float_types[4];
 300          else
 301             return half_float_types[size];
 302       case GL_INT: return int_types_norm[size];
 303       case GL_SHORT: return short_types_norm[size];
 304       case GL_BYTE: return byte_types_norm[size];
 305       case GL_UNSIGNED_INT: return uint_types_norm[size];
 306       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 307       case GL_UNSIGNED_BYTE:
 308          if (glarray->Format == GL_BGRA) {
 309             /* See GL_EXT_vertex_array_bgra */
 310             assert(size == 4);
 311             return ISL_FORMAT_B8G8R8A8_UNORM;
 312          }
 313          else {
 314             return ubyte_types_norm[size];
 315          }
 316       case GL_FIXED:
 317          if (brw->gen >= 8 || brw->is_haswell)
 318             return fixed_point_types[size];
 319
 320          /* This produces GL_FIXED inputs as values between INT32_MIN and
 321           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 322           */
 323          return int_types_scale[size];
 324       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 325        * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
 326        * like to use here, so upload everything as UINT and fix
 327        * it in the shader
 328        */
 329       case GL_INT_2_10_10_10_REV:
 330          assert(size == 4);
 331          if (brw->gen >= 8 || brw->is_haswell) {
 332             return glarray->Format == GL_BGRA
 333                ? ISL_FORMAT_B10G10R10A2_SNORM
 334                : ISL_FORMAT_R10G10B10A2_SNORM;
 335          }
 336          return ISL_FORMAT_R10G10B10A2_UINT;
 337       case GL_UNSIGNED_INT_2_10_10_10_REV:
 338          assert(size == 4);
 339          if (brw->gen >= 8 || brw->is_haswell) {
 340             return glarray->Format == GL_BGRA
 341                ? ISL_FORMAT_B10G10R10A2_UNORM
 342                : ISL_FORMAT_R10G10B10A2_UNORM;
 343          }
 344          return ISL_FORMAT_R10G10B10A2_UINT;
 345       default: unreachable("not reached");
 346       }
 347    }
 348    else {
 349       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 350        * W/A: the hardware doesn't really support the formats we'd
 351        * like to use here, so upload everything as UINT and fix
 352        * it in the shader
 353        */
 354       if (glarray->Type == GL_INT_2_10_10_10_REV) {
 355          assert(size == 4);
 356          if (brw->gen >= 8 || brw->is_haswell) {
 357             return glarray->Format == GL_BGRA
 358                ? ISL_FORMAT_B10G10R10A2_SSCALED
 359                : ISL_FORMAT_R10G10B10A2_SSCALED;
 360          }
 361          return ISL_FORMAT_R10G10B10A2_UINT;
 362       } else if (glarray->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 363          assert(size == 4);
 364          if (brw->gen >= 8 || brw->is_haswell) {
 365             return glarray->Format == GL_BGRA
 366                ? ISL_FORMAT_B10G10R10A2_USCALED
 367                : ISL_FORMAT_R10G10B10A2_USCALED;
 368          }
 369          return ISL_FORMAT_R10G10B10A2_UINT;
 370       }
 371       assert(glarray->Format == GL_RGBA); /* sanity check */
 372       switch (glarray->Type) {
 373       case GL_DOUBLE: return double_types(brw, size, glarray->Doubles);
 374       case GL_FLOAT: return float_types[size];
 375       case GL_HALF_FLOAT:
 376       case GL_HALF_FLOAT_OES:
 377          if (brw->gen < 6 && size == 3)
 378             return half_float_types[4];
 379          else
 380             return half_float_types[size];
 381       case GL_INT: return int_types_scale[size];
 382       case GL_SHORT: return short_types_scale[size];
 383       case GL_BYTE: return byte_types_scale[size];
 384       case GL_UNSIGNED_INT: return uint_types_scale[size];
 385       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 386       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 387       case GL_FIXED:
 388          if (brw->gen >= 8 || brw->is_haswell)
 389             return fixed_point_types[size];
 390
 391          /* This produces GL_FIXED inputs as values between INT32_MIN and
 392           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 393           */
 394          return int_types_scale[size];
 395       default: unreachable("not reached");
 396       }
 397    }
 398 }
 399
 400 static void
 401 copy_array_to_vbo_array(struct brw_context *brw,
 402                         struct brw_vertex_element *element,
 403                         int min, int max,
 404                         struct brw_vertex_buffer *buffer,
 405                         GLuint dst_stride)
 406 {
 407    const int src_stride = element->glarray->StrideB;
 408
 409    /* If the source stride is zero, we just want to upload the current
 410     * attribute once and set the buffer's stride to 0.  There's no need
 411     * to replicate it out.
 412     */
 413    if (src_stride == 0) {
 414       intel_upload_data(brw, element->glarray->Ptr,
 415                         element->glarray->_ElementSize,
 416                         element->glarray->_ElementSize,
 417                         &buffer->bo, &buffer->offset);
 418
 419       buffer->stride = 0;
 420       buffer->size = element->glarray->_ElementSize;
 421       return;
 422    }
 423
 424    const unsigned char *src = element->glarray->Ptr + min * src_stride;
 425    int count = max - min + 1;
 426    GLuint size = count * dst_stride;
 427    uint8_t *dst = intel_upload_space(brw, size, dst_stride,
 428                                      &buffer->bo, &buffer->offset);
 429
 430    /* The GL 4.5 spec says:
 431     *      "If any enabled array’s buffer binding is zero when DrawArrays or
 432     *      one of the other drawing commands defined in section 10.4 is called,
 433     *      the result is undefined."
 434     *
 435     * In this case, let's the dst with undefined values
 436     */
 437    if (src != NULL) {
 438       if (dst_stride == src_stride) {
 439          memcpy(dst, src, size);
 440       } else {
 441          while (count--) {
 442             memcpy(dst, src, dst_stride);
 443             src += src_stride;
 444             dst += dst_stride;
 445          }
 446       }
 447    }
 448    buffer->stride = dst_stride;
 449    buffer->size = size;
 450 }
 451
 452 void
 453 brw_prepare_vertices(struct brw_context *brw)
 454 {
 455    struct gl_context *ctx = &brw->ctx;
 456    /* BRW_NEW_VS_PROG_DATA */
 457    const struct brw_vs_prog_data *vs_prog_data =
 458       brw_vs_prog_data(brw->vs.base.prog_data);
 459    GLbitfield64 vs_inputs = vs_prog_data->inputs_read;
 460    const unsigned char *ptr = NULL;
 461    GLuint interleaved = 0;
 462    unsigned int min_index = brw->vb.min_index + brw->basevertex;
 463    unsigned int max_index = brw->vb.max_index + brw->basevertex;
 464    unsigned i;
 465    int delta, j;
 466
 467    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 468    GLuint nr_uploads = 0;
 469
 470    /* _NEW_POLYGON
 471     *
 472     * On gen6+, edge flags don't end up in the VUE (either in or out of the
 473     * VS).  Instead, they're uploaded as the last vertex element, and the data
 474     * is passed sideband through the fixed function units.  So, we need to
 475     * prepare the vertex buffer for it, but it's not present in inputs_read.
 476     */
 477    if (brw->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
 478                            ctx->Polygon.BackMode != GL_FILL)) {
 479       vs_inputs |= VERT_BIT_EDGEFLAG;
 480    }
 481
 482    if (0)
 483       fprintf(stderr, "%s %d..%d\n", __func__, min_index, max_index);
 484
 485    /* Accumulate the list of enabled arrays. */
 486    brw->vb.nr_enabled = 0;
 487    while (vs_inputs) {
 488       GLuint first = ffsll(vs_inputs) - 1;
 489       assert (first < 64);
 490       GLuint index =
 491          first - DIV_ROUND_UP(_mesa_bitcount_64(vs_prog_data->double_inputs_read &
 492                                                 BITFIELD64_MASK(first)), 2);
 493       struct brw_vertex_element *input = &brw->vb.inputs[index];
 494       input->is_dual_slot = (vs_prog_data->double_inputs_read & BITFIELD64_BIT(first)) != 0;
 495       vs_inputs &= ~BITFIELD64_BIT(first);
 496       if (input->is_dual_slot)
 497          vs_inputs &= ~BITFIELD64_BIT(first + 1);
 498       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 499    }
 500
 501    if (brw->vb.nr_enabled == 0)
 502       return;
 503
 504    if (brw->vb.nr_buffers)
 505       return;
 506
 507    /* The range of data in a given buffer represented as [min, max) */
 508    struct intel_buffer_object *enabled_buffer[VERT_ATTRIB_MAX];
 509    uint32_t buffer_range_start[VERT_ATTRIB_MAX];
 510    uint32_t buffer_range_end[VERT_ATTRIB_MAX];
 511
 512    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 513       struct brw_vertex_element *input = brw->vb.enabled[i];
 514       const struct gl_vertex_array *glarray = input->glarray;
 515
 516       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 517          struct intel_buffer_object *intel_buffer =
 518             intel_buffer_object(glarray->BufferObj);
 519
 520          const uint32_t offset = (uintptr_t)glarray->Ptr;
 521
 522          /* Start with the worst case */
 523          uint32_t start = 0;
 524          uint32_t range = intel_buffer->Base.Size;
 525          if (glarray->InstanceDivisor) {
 526             if (brw->num_instances) {
 527                start = offset + glarray->StrideB * brw->baseinstance;
 528                range = (glarray->StrideB * ((brw->num_instances - 1) /
 529                                             glarray->InstanceDivisor) +
 530                         glarray->_ElementSize);
 531             }
 532          } else {
 533             if (brw->vb.index_bounds_valid) {
 534                start = offset + min_index * glarray->StrideB;
 535                range = (glarray->StrideB * (max_index - min_index) +
 536                         glarray->_ElementSize);
 537             }
 538          }
 539
 540          /* If we have a VB set to be uploaded for this buffer object
 541           * already, reuse that VB state so that we emit fewer
 542           * relocations.
 543           */
 544          unsigned k;
 545          for (k = 0; k < i; k++) {
 546             const struct gl_vertex_array *other = brw->vb.enabled[k]->glarray;
 547             if (glarray->BufferObj == other->BufferObj &&
 548                 glarray->StrideB == other->StrideB &&
 549                 glarray->InstanceDivisor == other->InstanceDivisor &&
 550                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 551             {
 552                input->buffer = brw->vb.enabled[k]->buffer;
 553                input->offset = glarray->Ptr - other->Ptr;
 554
 555                buffer_range_start[input->buffer] =
 556                   MIN2(buffer_range_start[input->buffer], start);
 557                buffer_range_end[input->buffer] =
 558                   MAX2(buffer_range_end[input->buffer], start + range);
 559                break;
 560             }
 561          }
 562          if (k == i) {
 563             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 564
 565             /* Named buffer object: Just reference its contents directly. */
 566             buffer->offset = offset;
 567             buffer->stride = glarray->StrideB;
 568             buffer->step_rate = glarray->InstanceDivisor;
 569             buffer->size = glarray->BufferObj->Size - offset;
 570
 571             enabled_buffer[j] = intel_buffer;
 572             buffer_range_start[j] = start;
 573             buffer_range_end[j] = start + range;
 574
 575             input->buffer = j++;
 576             input->offset = 0;
 577          }
 578       } else {
 579          /* Queue the buffer object up to be uploaded in the next pass,
 580           * when we've decided if we're doing interleaved or not.
 581           */
 582          if (nr_uploads == 0) {
 583             interleaved = glarray->StrideB;
 584             ptr = glarray->Ptr;
 585          }
 586          else if (interleaved != glarray->StrideB ||
 587                   glarray->InstanceDivisor != 0 ||
 588                   glarray->Ptr < ptr ||
 589                   (uintptr_t)(glarray->Ptr - ptr) + glarray->_ElementSize > interleaved)
 590          {
 591             /* If our stride is different from the first attribute's stride,
 592              * or if we are using an instance divisor or if the first
 593              * attribute's stride didn't cover our element, disable the
 594              * interleaved upload optimization.  The second case can most
 595              * commonly occur in cases where there is a single vertex and, for
 596              * example, the data is stored on the application's stack.
 597              *
 598              * NOTE: This will also disable the optimization in cases where
 599              * the data is in a different order than the array indices.
 600              * Something like:
 601              *
 602              *     float data[...];
 603              *     glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
 604              *     glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
 605              */
 606             interleaved = 0;
 607          }
 608
 609          upload[nr_uploads++] = input;
 610       }
 611    }
 612
 613    /* Now that we've set up all of the buffers, we walk through and reference
 614     * each of them.  We do this late so that we get the right size in each
 615     * buffer and don't reference too little data.
 616     */
 617    for (i = 0; i < j; i++) {
 618       struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 619       if (buffer->bo)
 620          continue;
 621
 622       const uint32_t start = buffer_range_start[i];
 623       const uint32_t range = buffer_range_end[i] - buffer_range_start[i];
 624
 625       buffer->bo = intel_bufferobj_buffer(brw, enabled_buffer[i], start, range);
 626       brw_bo_reference(buffer->bo);
 627    }
 628
 629    /* If we need to upload all the arrays, then we can trim those arrays to
 630     * only the used elements [min_index, max_index] so long as we adjust all
 631     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 632     */
 633    brw->vb.start_vertex_bias = 0;
 634    delta = min_index;
 635    if (nr_uploads == brw->vb.nr_enabled) {
 636       brw->vb.start_vertex_bias = -delta;
 637       delta = 0;
 638    }
 639
 640    /* Handle any arrays to be uploaded. */
 641    if (nr_uploads > 1) {
 642       if (interleaved) {
 643          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 644          /* All uploads are interleaved, so upload the arrays together as
 645           * interleaved.  First, upload the contents and set up upload[0].
 646           */
 647          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 648                                  buffer, interleaved);
 649          buffer->offset -= delta * interleaved;
 650          buffer->size += delta * interleaved;
 651          buffer->step_rate = 0;
 652
 653          for (i = 0; i < nr_uploads; i++) {
 654             /* Then, just point upload[i] at upload[0]'s buffer. */
 655             upload[i]->offset =
 656                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 657             upload[i]->buffer = j;
 658          }
 659          j++;
 660
 661          nr_uploads = 0;
 662       }
 663    }
 664    /* Upload non-interleaved arrays */
 665    for (i = 0; i < nr_uploads; i++) {
 666       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 667       if (upload[i]->glarray->InstanceDivisor == 0) {
 668          copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 669                                  buffer, upload[i]->glarray->_ElementSize);
 670       } else {
 671          /* This is an instanced attribute, since its InstanceDivisor
 672           * is not zero. Therefore, its data will be stepped after the
 673           * instanced draw has been run InstanceDivisor times.
 674           */
 675          uint32_t instanced_attr_max_index =
 676             (brw->num_instances - 1) / upload[i]->glarray->InstanceDivisor;
 677          copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
 678                                  buffer, upload[i]->glarray->_ElementSize);
 679       }
 680       buffer->offset -= delta * buffer->stride;
 681       buffer->size += delta * buffer->stride;
 682       buffer->step_rate = upload[i]->glarray->InstanceDivisor;
 683       upload[i]->buffer = j++;
 684       upload[i]->offset = 0;
 685    }
 686
 687    brw->vb.nr_buffers = j;
 688 }
 689
 690 void
 691 brw_prepare_shader_draw_parameters(struct brw_context *brw)
 692 {
 693    const struct brw_vs_prog_data *vs_prog_data =
 694       brw_vs_prog_data(brw->vs.base.prog_data);
 695
 696    /* For non-indirect draws, upload gl_BaseVertex. */
 697    if ((vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance) &&
 698        brw->draw.draw_params_bo == NULL) {
 699       intel_upload_data(brw, &brw->draw.params, sizeof(brw->draw.params), 4,
 700                         &brw->draw.draw_params_bo,
 701                         &brw->draw.draw_params_offset);
 702    }
 703
 704    if (vs_prog_data->uses_drawid) {
 705       intel_upload_data(brw, &brw->draw.gl_drawid, sizeof(brw->draw.gl_drawid), 4,
 706                         &brw->draw.draw_id_bo,
 707                         &brw->draw.draw_id_offset);
 708    }
 709 }
 710
 711 static void
 712 brw_upload_indices(struct brw_context *brw)
 713 {
 714    struct gl_context *ctx = &brw->ctx;
 715    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 716    GLuint ib_size;
 717    struct brw_bo *old_bo = brw->ib.bo;
 718    struct gl_buffer_object *bufferobj;
 719    GLuint offset;
 720    GLuint ib_type_size;
 721
 722    if (index_buffer == NULL)
 723       return;
 724
 725    ib_type_size = index_buffer->index_size;
 726    ib_size = index_buffer->count ? ib_type_size * index_buffer->count :
 727                                    index_buffer->obj->Size;
 728    bufferobj = index_buffer->obj;
 729
 730    /* Turn into a proper VBO:
 731     */
 732    if (!_mesa_is_bufferobj(bufferobj)) {
 733       /* Get new bufferobj, offset:
 734        */
 735       intel_upload_data(brw, index_buffer->ptr, ib_size, ib_type_size,
 736                         &brw->ib.bo, &offset);
 737       brw->ib.size = brw->ib.bo->size;
 738    } else {
 739       offset = (GLuint) (unsigned long) index_buffer->ptr;
 740
 741       /* If the index buffer isn't aligned to its element size, we have to
 742        * rebase it into a temporary.
 743        */
 744       if ((ib_type_size - 1) & offset) {
 745          perf_debug("copying index buffer to a temporary to work around "
 746                     "misaligned offset %d\n", offset);
 747
 748          GLubyte *map = ctx->Driver.MapBufferRange(ctx,
 749                                                    offset,
 750                                                    ib_size,
 751                                                    GL_MAP_READ_BIT,
 752                                                    bufferobj,
 753                                                    MAP_INTERNAL);
 754
 755          intel_upload_data(brw, map, ib_size, ib_type_size,
 756                            &brw->ib.bo, &offset);
 757          brw->ib.size = brw->ib.bo->size;
 758
 759          ctx->Driver.UnmapBuffer(ctx, bufferobj, MAP_INTERNAL);
 760       } else {
 761          struct brw_bo *bo =
 762             intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 763                                    offset, ib_size);
 764          if (bo != brw->ib.bo) {
 765             brw_bo_unreference(brw->ib.bo);
 766             brw->ib.bo = bo;
 767             brw->ib.size = bufferobj->Size;
 768             brw_bo_reference(bo);
 769          }
 770       }
 771    }
 772
 773    /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
 774     * the index buffer state when we're just moving the start index
 775     * of our drawing.
 776     */
 777    brw->ib.start_vertex_offset = offset / ib_type_size;
 778
 779    if (brw->ib.bo != old_bo)
 780       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 781
 782    if (index_buffer->index_size != brw->ib.index_size) {
 783       brw->ib.index_size = index_buffer->index_size;
 784       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 785    }
 786 }
 787
 788 const struct brw_tracked_state brw_indices = {
 789    .dirty = {
 790       .mesa = 0,
 791       .brw = BRW_NEW_BLORP |
 792              BRW_NEW_INDICES,
 793    },
 794    .emit = brw_upload_indices,
 795 };