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/arrayobj.h"
  27 #include "main/bufferobj.h"
  28 #include "main/context.h"
  29 #include "main/enums.h"
  30 #include "main/macros.h"
  31 #include "main/glformats.h"
  32
  33 #include "brw_draw.h"
  34 #include "brw_defines.h"
  35 #include "brw_context.h"
  36 #include "brw_state.h"
  37
  38 #include "intel_batchbuffer.h"
  39 #include "intel_buffer_objects.h"
  40
  41 static const GLuint double_types_float[5] = {
  42    0,
  43    ISL_FORMAT_R64_FLOAT,
  44    ISL_FORMAT_R64G64_FLOAT,
  45    ISL_FORMAT_R64G64B64_FLOAT,
  46    ISL_FORMAT_R64G64B64A64_FLOAT
  47 };
  48
  49 static const GLuint double_types_passthru[5] = {
  50    0,
  51    ISL_FORMAT_R64_PASSTHRU,
  52    ISL_FORMAT_R64G64_PASSTHRU,
  53    ISL_FORMAT_R64G64B64_PASSTHRU,
  54    ISL_FORMAT_R64G64B64A64_PASSTHRU
  55 };
  56
  57 static const GLuint float_types[5] = {
  58    0,
  59    ISL_FORMAT_R32_FLOAT,
  60    ISL_FORMAT_R32G32_FLOAT,
  61    ISL_FORMAT_R32G32B32_FLOAT,
  62    ISL_FORMAT_R32G32B32A32_FLOAT
  63 };
  64
  65 static const GLuint half_float_types[5] = {
  66    0,
  67    ISL_FORMAT_R16_FLOAT,
  68    ISL_FORMAT_R16G16_FLOAT,
  69    ISL_FORMAT_R16G16B16_FLOAT,
  70    ISL_FORMAT_R16G16B16A16_FLOAT
  71 };
  72
  73 static const GLuint fixed_point_types[5] = {
  74    0,
  75    ISL_FORMAT_R32_SFIXED,
  76    ISL_FORMAT_R32G32_SFIXED,
  77    ISL_FORMAT_R32G32B32_SFIXED,
  78    ISL_FORMAT_R32G32B32A32_SFIXED,
  79 };
  80
  81 static const GLuint uint_types_direct[5] = {
  82    0,
  83    ISL_FORMAT_R32_UINT,
  84    ISL_FORMAT_R32G32_UINT,
  85    ISL_FORMAT_R32G32B32_UINT,
  86    ISL_FORMAT_R32G32B32A32_UINT
  87 };
  88
  89 static const GLuint uint_types_norm[5] = {
  90    0,
  91    ISL_FORMAT_R32_UNORM,
  92    ISL_FORMAT_R32G32_UNORM,
  93    ISL_FORMAT_R32G32B32_UNORM,
  94    ISL_FORMAT_R32G32B32A32_UNORM
  95 };
  96
  97 static const GLuint uint_types_scale[5] = {
  98    0,
  99    ISL_FORMAT_R32_USCALED,
 100    ISL_FORMAT_R32G32_USCALED,
 101    ISL_FORMAT_R32G32B32_USCALED,
 102    ISL_FORMAT_R32G32B32A32_USCALED
 103 };
 104
 105 static const GLuint int_types_direct[5] = {
 106    0,
 107    ISL_FORMAT_R32_SINT,
 108    ISL_FORMAT_R32G32_SINT,
 109    ISL_FORMAT_R32G32B32_SINT,
 110    ISL_FORMAT_R32G32B32A32_SINT
 111 };
 112
 113 static const GLuint int_types_norm[5] = {
 114    0,
 115    ISL_FORMAT_R32_SNORM,
 116    ISL_FORMAT_R32G32_SNORM,
 117    ISL_FORMAT_R32G32B32_SNORM,
 118    ISL_FORMAT_R32G32B32A32_SNORM
 119 };
 120
 121 static const GLuint int_types_scale[5] = {
 122    0,
 123    ISL_FORMAT_R32_SSCALED,
 124    ISL_FORMAT_R32G32_SSCALED,
 125    ISL_FORMAT_R32G32B32_SSCALED,
 126    ISL_FORMAT_R32G32B32A32_SSCALED
 127 };
 128
 129 static const GLuint ushort_types_direct[5] = {
 130    0,
 131    ISL_FORMAT_R16_UINT,
 132    ISL_FORMAT_R16G16_UINT,
 133    ISL_FORMAT_R16G16B16_UINT,
 134    ISL_FORMAT_R16G16B16A16_UINT
 135 };
 136
 137 static const GLuint ushort_types_norm[5] = {
 138    0,
 139    ISL_FORMAT_R16_UNORM,
 140    ISL_FORMAT_R16G16_UNORM,
 141    ISL_FORMAT_R16G16B16_UNORM,
 142    ISL_FORMAT_R16G16B16A16_UNORM
 143 };
 144
 145 static const GLuint ushort_types_scale[5] = {
 146    0,
 147    ISL_FORMAT_R16_USCALED,
 148    ISL_FORMAT_R16G16_USCALED,
 149    ISL_FORMAT_R16G16B16_USCALED,
 150    ISL_FORMAT_R16G16B16A16_USCALED
 151 };
 152
 153 static const GLuint short_types_direct[5] = {
 154    0,
 155    ISL_FORMAT_R16_SINT,
 156    ISL_FORMAT_R16G16_SINT,
 157    ISL_FORMAT_R16G16B16_SINT,
 158    ISL_FORMAT_R16G16B16A16_SINT
 159 };
 160
 161 static const GLuint short_types_norm[5] = {
 162    0,
 163    ISL_FORMAT_R16_SNORM,
 164    ISL_FORMAT_R16G16_SNORM,
 165    ISL_FORMAT_R16G16B16_SNORM,
 166    ISL_FORMAT_R16G16B16A16_SNORM
 167 };
 168
 169 static const GLuint short_types_scale[5] = {
 170    0,
 171    ISL_FORMAT_R16_SSCALED,
 172    ISL_FORMAT_R16G16_SSCALED,
 173    ISL_FORMAT_R16G16B16_SSCALED,
 174    ISL_FORMAT_R16G16B16A16_SSCALED
 175 };
 176
 177 static const GLuint ubyte_types_direct[5] = {
 178    0,
 179    ISL_FORMAT_R8_UINT,
 180    ISL_FORMAT_R8G8_UINT,
 181    ISL_FORMAT_R8G8B8_UINT,
 182    ISL_FORMAT_R8G8B8A8_UINT
 183 };
 184
 185 static const GLuint ubyte_types_norm[5] = {
 186    0,
 187    ISL_FORMAT_R8_UNORM,
 188    ISL_FORMAT_R8G8_UNORM,
 189    ISL_FORMAT_R8G8B8_UNORM,
 190    ISL_FORMAT_R8G8B8A8_UNORM
 191 };
 192
 193 static const GLuint ubyte_types_scale[5] = {
 194    0,
 195    ISL_FORMAT_R8_USCALED,
 196    ISL_FORMAT_R8G8_USCALED,
 197    ISL_FORMAT_R8G8B8_USCALED,
 198    ISL_FORMAT_R8G8B8A8_USCALED
 199 };
 200
 201 static const GLuint byte_types_direct[5] = {
 202    0,
 203    ISL_FORMAT_R8_SINT,
 204    ISL_FORMAT_R8G8_SINT,
 205    ISL_FORMAT_R8G8B8_SINT,
 206    ISL_FORMAT_R8G8B8A8_SINT
 207 };
 208
 209 static const GLuint byte_types_norm[5] = {
 210    0,
 211    ISL_FORMAT_R8_SNORM,
 212    ISL_FORMAT_R8G8_SNORM,
 213    ISL_FORMAT_R8G8B8_SNORM,
 214    ISL_FORMAT_R8G8B8A8_SNORM
 215 };
 216
 217 static const GLuint byte_types_scale[5] = {
 218    0,
 219    ISL_FORMAT_R8_SSCALED,
 220    ISL_FORMAT_R8G8_SSCALED,
 221    ISL_FORMAT_R8G8B8_SSCALED,
 222    ISL_FORMAT_R8G8B8A8_SSCALED
 223 };
 224
 225 static GLuint
 226 double_types(int size, GLboolean doubles)
 227 {
 228    /* From the BDW PRM, Volume 2d, page 588 (VERTEX_ELEMENT_STATE):
 229     * "When SourceElementFormat is set to one of the *64*_PASSTHRU formats,
 230     * 64-bit components are stored in the URB without any conversion."
 231     * Also included on BDW PRM, Volume 7, page 470, table "Source Element
 232     * Formats Supported in VF Unit"
 233     *
 234     * Previous PRMs don't include those references, so for gen7 we can't use
 235     * PASSTHRU formats directly. But in any case, we prefer to return passthru
 236     * even in that case, because that reflects what we want to achieve, even
 237     * if we would need to workaround on gen < 8.
 238     */
 239    return (doubles
 240            ? double_types_passthru[size]
 241            : double_types_float[size]);
 242 }
 243
 244 /**
 245  * Given vertex array type/size/format/normalized info, return
 246  * the appopriate hardware surface type.
 247  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 248  */
 249 unsigned
 250 brw_get_vertex_surface_type(struct brw_context *brw,
 251                             const struct gl_array_attributes *glattrib)
 252 {
 253    int size = glattrib->Size;
 254    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 255    const bool is_ivybridge_or_older =
 256       devinfo->gen <= 7 && !devinfo->is_baytrail && !devinfo->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(glattrib->Type),
 261               glattrib->Size, glattrib->Normalized);
 262
 263    if (glattrib->Integer) {
 264       assert(glattrib->Format == GL_RGBA); /* sanity check */
 265       switch (glattrib->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 (glattrib->Type == GL_UNSIGNED_INT_10F_11F_11F_REV) {
 291       return ISL_FORMAT_R11G11B10_FLOAT;
 292    } else if (glattrib->Normalized) {
 293       switch (glattrib->Type) {
 294       case GL_DOUBLE: return double_types(size, glattrib->Doubles);
 295       case GL_FLOAT: return float_types[size];
 296       case GL_HALF_FLOAT:
 297       case GL_HALF_FLOAT_OES:
 298          if (devinfo->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 (glattrib->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 (devinfo->gen >= 8 || devinfo->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 (devinfo->gen >= 8 || devinfo->is_haswell) {
 332             return glattrib->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 (devinfo->gen >= 8 || devinfo->is_haswell) {
 340             return glattrib->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 (glattrib->Type == GL_INT_2_10_10_10_REV) {
 355          assert(size == 4);
 356          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 357             return glattrib->Format == GL_BGRA
 358                ? ISL_FORMAT_B10G10R10A2_SSCALED
 359                : ISL_FORMAT_R10G10B10A2_SSCALED;
 360          }
 361          return ISL_FORMAT_R10G10B10A2_UINT;
 362       } else if (glattrib->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 363          assert(size == 4);
 364          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 365             return glattrib->Format == GL_BGRA
 366                ? ISL_FORMAT_B10G10R10A2_USCALED
 367                : ISL_FORMAT_R10G10B10A2_USCALED;
 368          }
 369          return ISL_FORMAT_R10G10B10A2_UINT;
 370       }
 371       assert(glattrib->Format == GL_RGBA); /* sanity check */
 372       switch (glattrib->Type) {
 373       case GL_DOUBLE: return double_types(size, glattrib->Doubles);
 374       case GL_FLOAT: return float_types[size];
 375       case GL_HALF_FLOAT:
 376       case GL_HALF_FLOAT_OES:
 377          if (devinfo->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 (devinfo->gen >= 8 || devinfo->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 struct gl_vertex_buffer_binding *glbinding = element->glbinding;
 408    const struct gl_array_attributes *glattrib = element->glattrib;
 409    const int src_stride = glbinding->Stride;
 410
 411    /* If the source stride is zero, we just want to upload the current
 412     * attribute once and set the buffer's stride to 0.  There's no need
 413     * to replicate it out.
 414     */
 415    if (src_stride == 0) {
 416       brw_upload_data(&brw->upload, glattrib->Ptr, glattrib->_ElementSize,
 417                       glattrib->_ElementSize, &buffer->bo, &buffer->offset);
 418
 419       buffer->stride = 0;
 420       buffer->size = glattrib->_ElementSize;
 421       return;
 422    }
 423
 424    const unsigned char *src = glattrib->Ptr + min * src_stride;
 425    int count = max - min + 1;
 426    GLuint size = count * dst_stride;
 427    uint8_t *dst = brw_upload_space(&brw->upload, 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    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 456    struct gl_context *ctx = &brw->ctx;
 457    /* BRW_NEW_VS_PROG_DATA */
 458    const struct brw_vs_prog_data *vs_prog_data =
 459       brw_vs_prog_data(brw->vs.base.prog_data);
 460    GLbitfield64 vs_inputs = vs_prog_data->inputs_read;
 461    const unsigned char *ptr = NULL;
 462    GLuint interleaved = 0;
 463    unsigned int min_index = brw->vb.min_index + brw->basevertex;
 464    unsigned int max_index = brw->vb.max_index + brw->basevertex;
 465    unsigned i;
 466    int delta, j;
 467
 468    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 469    GLuint nr_uploads = 0;
 470
 471    /* _NEW_POLYGON
 472     *
 473     * On gen6+, edge flags don't end up in the VUE (either in or out of the
 474     * VS).  Instead, they're uploaded as the last vertex element, and the data
 475     * is passed sideband through the fixed function units.  So, we need to
 476     * prepare the vertex buffer for it, but it's not present in inputs_read.
 477     */
 478    if (devinfo->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
 479                            ctx->Polygon.BackMode != GL_FILL)) {
 480       vs_inputs |= VERT_BIT_EDGEFLAG;
 481    }
 482
 483    if (0)
 484       fprintf(stderr, "%s %d..%d\n", __func__, min_index, max_index);
 485
 486    /* Accumulate the list of enabled arrays. */
 487    brw->vb.nr_enabled = 0;
 488    while (vs_inputs) {
 489       GLuint first = ffsll(vs_inputs) - 1;
 490       assert (first < 64);
 491       GLuint index =
 492          first - DIV_ROUND_UP(_mesa_bitcount_64(vs_prog_data->double_inputs_read &
 493                                                 BITFIELD64_MASK(first)), 2);
 494       struct brw_vertex_element *input = &brw->vb.inputs[index];
 495       input->is_dual_slot = (vs_prog_data->double_inputs_read & BITFIELD64_BIT(first)) != 0;
 496       vs_inputs &= ~BITFIELD64_BIT(first);
 497       if (input->is_dual_slot)
 498          vs_inputs &= ~BITFIELD64_BIT(first + 1);
 499       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 500    }
 501
 502    if (brw->vb.nr_enabled == 0)
 503       return;
 504
 505    if (brw->vb.nr_buffers)
 506       return;
 507
 508    /* The range of data in a given buffer represented as [min, max) */
 509    struct intel_buffer_object *enabled_buffer[VERT_ATTRIB_MAX];
 510    uint32_t buffer_range_start[VERT_ATTRIB_MAX];
 511    uint32_t buffer_range_end[VERT_ATTRIB_MAX];
 512
 513    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 514       struct brw_vertex_element *input = brw->vb.enabled[i];
 515       const struct gl_vertex_buffer_binding *glbinding = input->glbinding;
 516       const struct gl_array_attributes *glattrib = input->glattrib;
 517
 518       if (_mesa_is_bufferobj(glbinding->BufferObj)) {
 519          struct intel_buffer_object *intel_buffer =
 520             intel_buffer_object(glbinding->BufferObj);
 521
 522          const uint32_t offset = _mesa_draw_binding_offset(glbinding) +
 523             _mesa_draw_attributes_relative_offset(glattrib);
 524
 525          /* Start with the worst case */
 526          uint32_t start = 0;
 527          uint32_t range = intel_buffer->Base.Size;
 528          if (glbinding->InstanceDivisor) {
 529             if (brw->num_instances) {
 530                start = offset + glbinding->Stride * brw->baseinstance;
 531                range = (glbinding->Stride * ((brw->num_instances - 1) /
 532                                             glbinding->InstanceDivisor) +
 533                         glattrib->_ElementSize);
 534             }
 535          } else {
 536             if (brw->vb.index_bounds_valid) {
 537                start = offset + min_index * glbinding->Stride;
 538                range = (glbinding->Stride * (max_index - min_index) +
 539                         glattrib->_ElementSize);
 540             }
 541          }
 542
 543          /* If we have a VB set to be uploaded for this buffer object
 544           * already, reuse that VB state so that we emit fewer
 545           * relocations.
 546           */
 547          unsigned k;
 548          for (k = 0; k < i; k++) {
 549             struct brw_vertex_element *other = brw->vb.enabled[k];
 550             const struct gl_vertex_buffer_binding *obind = other->glbinding;
 551             const struct gl_array_attributes *oattrib = other->glattrib;
 552             const uint32_t ooffset = _mesa_draw_binding_offset(obind) +
 553                _mesa_draw_attributes_relative_offset(oattrib);
 554             if (glbinding->BufferObj == obind->BufferObj &&
 555                 glbinding->Stride == obind->Stride &&
 556                 glbinding->InstanceDivisor == obind->InstanceDivisor &&
 557                 (offset - ooffset) < glbinding->Stride)
 558             {
 559                input->buffer = brw->vb.enabled[k]->buffer;
 560                input->offset = offset - ooffset;
 561
 562                buffer_range_start[input->buffer] =
 563                   MIN2(buffer_range_start[input->buffer], start);
 564                buffer_range_end[input->buffer] =
 565                   MAX2(buffer_range_end[input->buffer], start + range);
 566                break;
 567             }
 568          }
 569          if (k == i) {
 570             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 571
 572             /* Named buffer object: Just reference its contents directly. */
 573             buffer->offset = offset;
 574             buffer->stride = glbinding->Stride;
 575             buffer->step_rate = glbinding->InstanceDivisor;
 576             buffer->size = glbinding->BufferObj->Size - offset;
 577
 578             enabled_buffer[j] = intel_buffer;
 579             buffer_range_start[j] = start;
 580             buffer_range_end[j] = start + range;
 581
 582             input->buffer = j++;
 583             input->offset = 0;
 584          }
 585       } else {
 586          /* Queue the buffer object up to be uploaded in the next pass,
 587           * when we've decided if we're doing interleaved or not.
 588           */
 589          if (nr_uploads == 0) {
 590             interleaved = glbinding->Stride;
 591             ptr = glattrib->Ptr;
 592          }
 593          else if (interleaved != glbinding->Stride ||
 594                   glbinding->InstanceDivisor != 0 ||
 595                   glattrib->Ptr < ptr ||
 596                   (uintptr_t)(glattrib->Ptr - ptr) + glattrib->_ElementSize > interleaved)
 597          {
 598             /* If our stride is different from the first attribute's stride,
 599              * or if we are using an instance divisor or if the first
 600              * attribute's stride didn't cover our element, disable the
 601              * interleaved upload optimization.  The second case can most
 602              * commonly occur in cases where there is a single vertex and, for
 603              * example, the data is stored on the application's stack.
 604              *
 605              * NOTE: This will also disable the optimization in cases where
 606              * the data is in a different order than the array indices.
 607              * Something like:
 608              *
 609              *     float data[...];
 610              *     glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
 611              *     glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
 612              */
 613             interleaved = 0;
 614          }
 615
 616          upload[nr_uploads++] = input;
 617       }
 618    }
 619
 620    /* Now that we've set up all of the buffers, we walk through and reference
 621     * each of them.  We do this late so that we get the right size in each
 622     * buffer and don't reference too little data.
 623     */
 624    for (i = 0; i < j; i++) {
 625       struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 626       if (buffer->bo)
 627          continue;
 628
 629       const uint32_t start = buffer_range_start[i];
 630       const uint32_t range = buffer_range_end[i] - buffer_range_start[i];
 631
 632       buffer->bo = intel_bufferobj_buffer(brw, enabled_buffer[i], start,
 633                                           range, false);
 634       brw_bo_reference(buffer->bo);
 635    }
 636
 637    /* If we need to upload all the arrays, then we can trim those arrays to
 638     * only the used elements [min_index, max_index] so long as we adjust all
 639     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 640     */
 641    brw->vb.start_vertex_bias = 0;
 642    delta = min_index;
 643    if (nr_uploads == brw->vb.nr_enabled) {
 644       brw->vb.start_vertex_bias = -delta;
 645       delta = 0;
 646    }
 647
 648    /* Handle any arrays to be uploaded. */
 649    if (nr_uploads > 1) {
 650       if (interleaved) {
 651          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 652          /* All uploads are interleaved, so upload the arrays together as
 653           * interleaved.  First, upload the contents and set up upload[0].
 654           */
 655          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 656                                  buffer, interleaved);
 657          buffer->offset -= delta * interleaved;
 658          buffer->size += delta * interleaved;
 659          buffer->step_rate = 0;
 660
 661          for (i = 0; i < nr_uploads; i++) {
 662             const struct gl_array_attributes *glattrib = upload[i]->glattrib;
 663             /* Then, just point upload[i] at upload[0]'s buffer. */
 664             upload[i]->offset = ((const unsigned char *)glattrib->Ptr - ptr);
 665             upload[i]->buffer = j;
 666          }
 667          j++;
 668
 669          nr_uploads = 0;
 670       }
 671    }
 672    /* Upload non-interleaved arrays */
 673    for (i = 0; i < nr_uploads; i++) {
 674       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 675       const struct gl_vertex_buffer_binding *glbinding = upload[i]->glbinding;
 676       const struct gl_array_attributes *glattrib = upload[i]->glattrib;
 677       if (glbinding->InstanceDivisor == 0) {
 678          copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 679                                  buffer, glattrib->_ElementSize);
 680       } else {
 681          /* This is an instanced attribute, since its InstanceDivisor
 682           * is not zero. Therefore, its data will be stepped after the
 683           * instanced draw has been run InstanceDivisor times.
 684           */
 685          uint32_t instanced_attr_max_index =
 686             (brw->num_instances - 1) / glbinding->InstanceDivisor;
 687          copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
 688                                  buffer, glattrib->_ElementSize);
 689       }
 690       buffer->offset -= delta * buffer->stride;
 691       buffer->size += delta * buffer->stride;
 692       buffer->step_rate = glbinding->InstanceDivisor;
 693       upload[i]->buffer = j++;
 694       upload[i]->offset = 0;
 695    }
 696
 697    brw->vb.nr_buffers = j;
 698 }
 699
 700 void
 701 brw_prepare_shader_draw_parameters(struct brw_context *brw)
 702 {
 703    const struct brw_vs_prog_data *vs_prog_data =
 704       brw_vs_prog_data(brw->vs.base.prog_data);
 705
 706    /* For non-indirect draws, upload the shader draw parameters */
 707    if ((vs_prog_data->uses_firstvertex || vs_prog_data->uses_baseinstance) &&
 708        brw->draw.draw_params_bo == NULL) {
 709       brw_upload_data(&brw->upload,
 710                       &brw->draw.params, sizeof(brw->draw.params), 4,
 711                       &brw->draw.draw_params_bo,
 712                       &brw->draw.draw_params_offset);
 713    }
 714
 715    if (vs_prog_data->uses_drawid || vs_prog_data->uses_is_indexed_draw) {
 716       brw_upload_data(&brw->upload,
 717                       &brw->draw.derived_params, sizeof(brw->draw.derived_params), 4,
 718                       &brw->draw.derived_draw_params_bo,
 719                       &brw->draw.derived_draw_params_offset);
 720    }
 721 }
 722
 723 static void
 724 brw_upload_indices(struct brw_context *brw)
 725 {
 726    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 727    GLuint ib_size;
 728    struct brw_bo *old_bo = brw->ib.bo;
 729    struct gl_buffer_object *bufferobj;
 730    GLuint offset;
 731    GLuint ib_type_size;
 732
 733    if (index_buffer == NULL)
 734       return;
 735
 736    ib_type_size = index_buffer->index_size;
 737    ib_size = index_buffer->count ? ib_type_size * index_buffer->count :
 738                                    index_buffer->obj->Size;
 739    bufferobj = index_buffer->obj;
 740
 741    /* Turn into a proper VBO:
 742     */
 743    if (!_mesa_is_bufferobj(bufferobj)) {
 744       /* Get new bufferobj, offset:
 745        */
 746       brw_upload_data(&brw->upload, index_buffer->ptr, ib_size, ib_type_size,
 747                       &brw->ib.bo, &offset);
 748       brw->ib.size = brw->ib.bo->size;
 749    } else {
 750       offset = (GLuint) (unsigned long) index_buffer->ptr;
 751
 752       struct brw_bo *bo =
 753          intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 754                                 offset, ib_size, false);
 755       if (bo != brw->ib.bo) {
 756          brw_bo_unreference(brw->ib.bo);
 757          brw->ib.bo = bo;
 758          brw->ib.size = bufferobj->Size;
 759          brw_bo_reference(bo);
 760       }
 761    }
 762
 763    /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
 764     * the index buffer state when we're just moving the start index
 765     * of our drawing.
 766     */
 767    brw->ib.start_vertex_offset = offset / ib_type_size;
 768
 769    if (brw->ib.bo != old_bo)
 770       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 771
 772    if (index_buffer->index_size != brw->ib.index_size) {
 773       brw->ib.index_size = index_buffer->index_size;
 774       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 775    }
 776 }
 777
 778 const struct brw_tracked_state brw_indices = {
 779    .dirty = {
 780       .mesa = 0,
 781       .brw = BRW_NEW_BLORP |
 782              BRW_NEW_INDICES,
 783    },
 784    .emit = brw_upload_indices,
 785 };