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 #include "nir.h"
  33
  34 #include "brw_draw.h"
  35 #include "brw_defines.h"
  36 #include "brw_context.h"
  37 #include "brw_state.h"
  38
  39 #include "intel_batchbuffer.h"
  40 #include "intel_buffer_objects.h"
  41
  42 static const GLuint double_types_float[5] = {
  43    0,
  44    ISL_FORMAT_R64_FLOAT,
  45    ISL_FORMAT_R64G64_FLOAT,
  46    ISL_FORMAT_R64G64B64_FLOAT,
  47    ISL_FORMAT_R64G64B64A64_FLOAT
  48 };
  49
  50 static const GLuint double_types_passthru[5] = {
  51    0,
  52    ISL_FORMAT_R64_PASSTHRU,
  53    ISL_FORMAT_R64G64_PASSTHRU,
  54    ISL_FORMAT_R64G64B64_PASSTHRU,
  55    ISL_FORMAT_R64G64B64A64_PASSTHRU
  56 };
  57
  58 static const GLuint float_types[5] = {
  59    0,
  60    ISL_FORMAT_R32_FLOAT,
  61    ISL_FORMAT_R32G32_FLOAT,
  62    ISL_FORMAT_R32G32B32_FLOAT,
  63    ISL_FORMAT_R32G32B32A32_FLOAT
  64 };
  65
  66 static const GLuint half_float_types[5] = {
  67    0,
  68    ISL_FORMAT_R16_FLOAT,
  69    ISL_FORMAT_R16G16_FLOAT,
  70    ISL_FORMAT_R16G16B16_FLOAT,
  71    ISL_FORMAT_R16G16B16A16_FLOAT
  72 };
  73
  74 static const GLuint fixed_point_types[5] = {
  75    0,
  76    ISL_FORMAT_R32_SFIXED,
  77    ISL_FORMAT_R32G32_SFIXED,
  78    ISL_FORMAT_R32G32B32_SFIXED,
  79    ISL_FORMAT_R32G32B32A32_SFIXED,
  80 };
  81
  82 static const GLuint uint_types_direct[5] = {
  83    0,
  84    ISL_FORMAT_R32_UINT,
  85    ISL_FORMAT_R32G32_UINT,
  86    ISL_FORMAT_R32G32B32_UINT,
  87    ISL_FORMAT_R32G32B32A32_UINT
  88 };
  89
  90 static const GLuint uint_types_norm[5] = {
  91    0,
  92    ISL_FORMAT_R32_UNORM,
  93    ISL_FORMAT_R32G32_UNORM,
  94    ISL_FORMAT_R32G32B32_UNORM,
  95    ISL_FORMAT_R32G32B32A32_UNORM
  96 };
  97
  98 static const GLuint uint_types_scale[5] = {
  99    0,
 100    ISL_FORMAT_R32_USCALED,
 101    ISL_FORMAT_R32G32_USCALED,
 102    ISL_FORMAT_R32G32B32_USCALED,
 103    ISL_FORMAT_R32G32B32A32_USCALED
 104 };
 105
 106 static const GLuint int_types_direct[5] = {
 107    0,
 108    ISL_FORMAT_R32_SINT,
 109    ISL_FORMAT_R32G32_SINT,
 110    ISL_FORMAT_R32G32B32_SINT,
 111    ISL_FORMAT_R32G32B32A32_SINT
 112 };
 113
 114 static const GLuint int_types_norm[5] = {
 115    0,
 116    ISL_FORMAT_R32_SNORM,
 117    ISL_FORMAT_R32G32_SNORM,
 118    ISL_FORMAT_R32G32B32_SNORM,
 119    ISL_FORMAT_R32G32B32A32_SNORM
 120 };
 121
 122 static const GLuint int_types_scale[5] = {
 123    0,
 124    ISL_FORMAT_R32_SSCALED,
 125    ISL_FORMAT_R32G32_SSCALED,
 126    ISL_FORMAT_R32G32B32_SSCALED,
 127    ISL_FORMAT_R32G32B32A32_SSCALED
 128 };
 129
 130 static const GLuint ushort_types_direct[5] = {
 131    0,
 132    ISL_FORMAT_R16_UINT,
 133    ISL_FORMAT_R16G16_UINT,
 134    ISL_FORMAT_R16G16B16_UINT,
 135    ISL_FORMAT_R16G16B16A16_UINT
 136 };
 137
 138 static const GLuint ushort_types_norm[5] = {
 139    0,
 140    ISL_FORMAT_R16_UNORM,
 141    ISL_FORMAT_R16G16_UNORM,
 142    ISL_FORMAT_R16G16B16_UNORM,
 143    ISL_FORMAT_R16G16B16A16_UNORM
 144 };
 145
 146 static const GLuint ushort_types_scale[5] = {
 147    0,
 148    ISL_FORMAT_R16_USCALED,
 149    ISL_FORMAT_R16G16_USCALED,
 150    ISL_FORMAT_R16G16B16_USCALED,
 151    ISL_FORMAT_R16G16B16A16_USCALED
 152 };
 153
 154 static const GLuint short_types_direct[5] = {
 155    0,
 156    ISL_FORMAT_R16_SINT,
 157    ISL_FORMAT_R16G16_SINT,
 158    ISL_FORMAT_R16G16B16_SINT,
 159    ISL_FORMAT_R16G16B16A16_SINT
 160 };
 161
 162 static const GLuint short_types_norm[5] = {
 163    0,
 164    ISL_FORMAT_R16_SNORM,
 165    ISL_FORMAT_R16G16_SNORM,
 166    ISL_FORMAT_R16G16B16_SNORM,
 167    ISL_FORMAT_R16G16B16A16_SNORM
 168 };
 169
 170 static const GLuint short_types_scale[5] = {
 171    0,
 172    ISL_FORMAT_R16_SSCALED,
 173    ISL_FORMAT_R16G16_SSCALED,
 174    ISL_FORMAT_R16G16B16_SSCALED,
 175    ISL_FORMAT_R16G16B16A16_SSCALED
 176 };
 177
 178 static const GLuint ubyte_types_direct[5] = {
 179    0,
 180    ISL_FORMAT_R8_UINT,
 181    ISL_FORMAT_R8G8_UINT,
 182    ISL_FORMAT_R8G8B8_UINT,
 183    ISL_FORMAT_R8G8B8A8_UINT
 184 };
 185
 186 static const GLuint ubyte_types_norm[5] = {
 187    0,
 188    ISL_FORMAT_R8_UNORM,
 189    ISL_FORMAT_R8G8_UNORM,
 190    ISL_FORMAT_R8G8B8_UNORM,
 191    ISL_FORMAT_R8G8B8A8_UNORM
 192 };
 193
 194 static const GLuint ubyte_types_scale[5] = {
 195    0,
 196    ISL_FORMAT_R8_USCALED,
 197    ISL_FORMAT_R8G8_USCALED,
 198    ISL_FORMAT_R8G8B8_USCALED,
 199    ISL_FORMAT_R8G8B8A8_USCALED
 200 };
 201
 202 static const GLuint byte_types_direct[5] = {
 203    0,
 204    ISL_FORMAT_R8_SINT,
 205    ISL_FORMAT_R8G8_SINT,
 206    ISL_FORMAT_R8G8B8_SINT,
 207    ISL_FORMAT_R8G8B8A8_SINT
 208 };
 209
 210 static const GLuint byte_types_norm[5] = {
 211    0,
 212    ISL_FORMAT_R8_SNORM,
 213    ISL_FORMAT_R8G8_SNORM,
 214    ISL_FORMAT_R8G8B8_SNORM,
 215    ISL_FORMAT_R8G8B8A8_SNORM
 216 };
 217
 218 static const GLuint byte_types_scale[5] = {
 219    0,
 220    ISL_FORMAT_R8_SSCALED,
 221    ISL_FORMAT_R8G8_SSCALED,
 222    ISL_FORMAT_R8G8B8_SSCALED,
 223    ISL_FORMAT_R8G8B8A8_SSCALED
 224 };
 225
 226 static GLuint
 227 double_types(int size, 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_format *glformat)
 253 {
 254    int size = glformat->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(glformat->Type),
 262               glformat->Size, glformat->Normalized);
 263
 264    if (glformat->Integer) {
 265       assert(glformat->Format == GL_RGBA); /* sanity check */
 266       switch (glformat->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 (glformat->Type == GL_UNSIGNED_INT_10F_11F_11F_REV) {
 292       return ISL_FORMAT_R11G11B10_FLOAT;
 293    } else if (glformat->Normalized) {
 294       switch (glformat->Type) {
 295       case GL_DOUBLE: return double_types(size, glformat->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 (glformat->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 glformat->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 glformat->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 (glformat->Type == GL_INT_2_10_10_10_REV) {
 356          assert(size == 4);
 357          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 358             return glformat->Format == GL_BGRA
 359                ? ISL_FORMAT_B10G10R10A2_SSCALED
 360                : ISL_FORMAT_R10G10B10A2_SSCALED;
 361          }
 362          return ISL_FORMAT_R10G10B10A2_UINT;
 363       } else if (glformat->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 364          assert(size == 4);
 365          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 366             return glformat->Format == GL_BGRA
 367                ? ISL_FORMAT_B10G10R10A2_USCALED
 368                : ISL_FORMAT_R10G10B10A2_USCALED;
 369          }
 370          return ISL_FORMAT_R10G10B10A2_UINT;
 371       }
 372       assert(glformat->Format == GL_RGBA); /* sanity check */
 373       switch (glformat->Type) {
 374       case GL_DOUBLE: return double_types(size, glformat->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 struct gl_vertex_buffer_binding *glbinding = element->glbinding;
 409    const struct gl_array_attributes *glattrib = element->glattrib;
 410    const struct gl_vertex_format *glformat = &glattrib->Format;
 411    const int src_stride = glbinding->Stride;
 412
 413    /* If the source stride is zero, we just want to upload the current
 414     * attribute once and set the buffer's stride to 0.  There's no need
 415     * to replicate it out.
 416     */
 417    if (src_stride == 0) {
 418       brw_upload_data(&brw->upload, glattrib->Ptr, glformat->_ElementSize,
 419                       glformat->_ElementSize, &buffer->bo, &buffer->offset);
 420
 421       buffer->stride = 0;
 422       buffer->size = glformat->_ElementSize;
 423       return;
 424    }
 425
 426    const unsigned char *src = glattrib->Ptr + min * src_stride;
 427    int count = max - min + 1;
 428    GLuint size = count * dst_stride;
 429    uint8_t *dst = brw_upload_space(&brw->upload, size, dst_stride,
 430                                    &buffer->bo, &buffer->offset);
 431
 432    /* The GL 4.5 spec says:
 433     *      "If any enabled array’s buffer binding is zero when DrawArrays or
 434     *      one of the other drawing commands defined in section 10.4 is called,
 435     *      the result is undefined."
 436     *
 437     * In this case, let's the dst with undefined values
 438     */
 439    if (src != NULL) {
 440       if (dst_stride == src_stride) {
 441          memcpy(dst, src, size);
 442       } else {
 443          while (count--) {
 444             memcpy(dst, src, dst_stride);
 445             src += src_stride;
 446             dst += dst_stride;
 447          }
 448       }
 449    }
 450    buffer->stride = dst_stride;
 451    buffer->size = size;
 452 }
 453
 454 void
 455 brw_prepare_vertices(struct brw_context *brw)
 456 {
 457    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 458    struct gl_context *ctx = &brw->ctx;
 459    /* BRW_NEW_VERTEX_PROGRAM */
 460    const struct gl_program *vp = brw->programs[MESA_SHADER_VERTEX];
 461    /* BRW_NEW_VS_PROG_DATA */
 462    const struct brw_vs_prog_data *vs_prog_data =
 463       brw_vs_prog_data(brw->vs.base.prog_data);
 464    GLbitfield64 vs_inputs =
 465       nir_get_single_slot_attribs_mask(vs_prog_data->inputs_read,
 466                                        vp->DualSlotInputs);
 467    const unsigned char *ptr = NULL;
 468    GLuint interleaved = 0;
 469    unsigned int min_index = brw->vb.min_index + brw->basevertex;
 470    unsigned int max_index = brw->vb.max_index + brw->basevertex;
 471    unsigned i;
 472    int delta, j;
 473
 474    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 475    GLuint nr_uploads = 0;
 476
 477    /* _NEW_POLYGON
 478     *
 479     * On gen6+, edge flags don't end up in the VUE (either in or out of the
 480     * VS).  Instead, they're uploaded as the last vertex element, and the data
 481     * is passed sideband through the fixed function units.  So, we need to
 482     * prepare the vertex buffer for it, but it's not present in inputs_read.
 483     */
 484    if (devinfo->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
 485                            ctx->Polygon.BackMode != GL_FILL)) {
 486       vs_inputs |= VERT_BIT_EDGEFLAG;
 487    }
 488
 489    if (0)
 490       fprintf(stderr, "%s %d..%d\n", __func__, min_index, max_index);
 491
 492    /* Accumulate the list of enabled arrays. */
 493    brw->vb.nr_enabled = 0;
 494    while (vs_inputs) {
 495       const unsigned index = ffsll(vs_inputs) - 1;
 496       assert(index < 64);
 497
 498       struct brw_vertex_element *input = &brw->vb.inputs[index];
 499       input->is_dual_slot = (vp->DualSlotInputs & BITFIELD64_BIT(index)) != 0;
 500       vs_inputs &= ~BITFIELD64_BIT(index);
 501       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 502    }
 503
 504    if (brw->vb.nr_enabled == 0)
 505       return;
 506
 507    if (brw->vb.nr_buffers)
 508       return;
 509
 510    /* The range of data in a given buffer represented as [min, max) */
 511    struct intel_buffer_object *enabled_buffer[VERT_ATTRIB_MAX];
 512    uint32_t buffer_range_start[VERT_ATTRIB_MAX];
 513    uint32_t buffer_range_end[VERT_ATTRIB_MAX];
 514
 515    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 516       struct brw_vertex_element *input = brw->vb.enabled[i];
 517       const struct gl_vertex_buffer_binding *glbinding = input->glbinding;
 518       const struct gl_array_attributes *glattrib = input->glattrib;
 519
 520       if (_mesa_is_bufferobj(glbinding->BufferObj)) {
 521          struct intel_buffer_object *intel_buffer =
 522             intel_buffer_object(glbinding->BufferObj);
 523
 524          const uint32_t offset = _mesa_draw_binding_offset(glbinding) +
 525             _mesa_draw_attributes_relative_offset(glattrib);
 526
 527          /* Start with the worst case */
 528          uint32_t start = 0;
 529          uint32_t range = intel_buffer->Base.Size;
 530          if (glbinding->InstanceDivisor) {
 531             if (brw->num_instances) {
 532                start = offset + glbinding->Stride * brw->baseinstance;
 533                range = (glbinding->Stride * ((brw->num_instances - 1) /
 534                                             glbinding->InstanceDivisor) +
 535                         glattrib->Format._ElementSize);
 536             }
 537          } else {
 538             if (brw->vb.index_bounds_valid) {
 539                start = offset + min_index * glbinding->Stride;
 540                range = (glbinding->Stride * (max_index - min_index) +
 541                         glattrib->Format._ElementSize);
 542             }
 543          }
 544
 545          /* If we have a VB set to be uploaded for this buffer object
 546           * already, reuse that VB state so that we emit fewer
 547           * relocations.
 548           */
 549          unsigned k;
 550          for (k = 0; k < i; k++) {
 551             struct brw_vertex_element *other = brw->vb.enabled[k];
 552             const struct gl_vertex_buffer_binding *obind = other->glbinding;
 553             const struct gl_array_attributes *oattrib = other->glattrib;
 554             const uint32_t ooffset = _mesa_draw_binding_offset(obind) +
 555                _mesa_draw_attributes_relative_offset(oattrib);
 556             if (glbinding->BufferObj == obind->BufferObj &&
 557                 glbinding->Stride == obind->Stride &&
 558                 glbinding->InstanceDivisor == obind->InstanceDivisor &&
 559                 (offset - ooffset) < glbinding->Stride)
 560             {
 561                input->buffer = brw->vb.enabled[k]->buffer;
 562                input->offset = offset - ooffset;
 563
 564                buffer_range_start[input->buffer] =
 565                   MIN2(buffer_range_start[input->buffer], start);
 566                buffer_range_end[input->buffer] =
 567                   MAX2(buffer_range_end[input->buffer], start + range);
 568                break;
 569             }
 570          }
 571          if (k == i) {
 572             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 573
 574             /* Named buffer object: Just reference its contents directly. */
 575             buffer->offset = offset;
 576             buffer->stride = glbinding->Stride;
 577             buffer->step_rate = glbinding->InstanceDivisor;
 578             buffer->size = glbinding->BufferObj->Size - offset;
 579
 580             enabled_buffer[j] = intel_buffer;
 581             buffer_range_start[j] = start;
 582             buffer_range_end[j] = start + range;
 583
 584             input->buffer = j++;
 585             input->offset = 0;
 586          }
 587       } else {
 588          /* Queue the buffer object up to be uploaded in the next pass,
 589           * when we've decided if we're doing interleaved or not.
 590           */
 591          if (nr_uploads == 0) {
 592             interleaved = glbinding->Stride;
 593             ptr = glattrib->Ptr;
 594          }
 595          else if (interleaved != glbinding->Stride ||
 596                   glbinding->InstanceDivisor != 0 ||
 597                   glattrib->Ptr < ptr ||
 598                   (uintptr_t)(glattrib->Ptr - ptr) +
 599                   glattrib->Format._ElementSize > interleaved)
 600          {
 601             /* If our stride is different from the first attribute's stride,
 602              * or if we are using an instance divisor or if the first
 603              * attribute's stride didn't cover our element, disable the
 604              * interleaved upload optimization.  The second case can most
 605              * commonly occur in cases where there is a single vertex and, for
 606              * example, the data is stored on the application's stack.
 607              *
 608              * NOTE: This will also disable the optimization in cases where
 609              * the data is in a different order than the array indices.
 610              * Something like:
 611              *
 612              *     float data[...];
 613              *     glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
 614              *     glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
 615              */
 616             interleaved = 0;
 617          }
 618
 619          upload[nr_uploads++] = input;
 620       }
 621    }
 622
 623    /* Now that we've set up all of the buffers, we walk through and reference
 624     * each of them.  We do this late so that we get the right size in each
 625     * buffer and don't reference too little data.
 626     */
 627    for (i = 0; i < j; i++) {
 628       struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 629       if (buffer->bo)
 630          continue;
 631
 632       const uint32_t start = buffer_range_start[i];
 633       const uint32_t range = buffer_range_end[i] - buffer_range_start[i];
 634
 635       buffer->bo = intel_bufferobj_buffer(brw, enabled_buffer[i], start,
 636                                           range, false);
 637       brw_bo_reference(buffer->bo);
 638    }
 639
 640    /* If we need to upload all the arrays, then we can trim those arrays to
 641     * only the used elements [min_index, max_index] so long as we adjust all
 642     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 643     */
 644    brw->vb.start_vertex_bias = 0;
 645    delta = min_index;
 646    if (nr_uploads == brw->vb.nr_enabled) {
 647       brw->vb.start_vertex_bias = -delta;
 648       delta = 0;
 649    }
 650
 651    /* Handle any arrays to be uploaded. */
 652    if (nr_uploads > 1) {
 653       if (interleaved) {
 654          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 655          /* All uploads are interleaved, so upload the arrays together as
 656           * interleaved.  First, upload the contents and set up upload[0].
 657           */
 658          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 659                                  buffer, interleaved);
 660          buffer->offset -= delta * interleaved;
 661          buffer->size += delta * interleaved;
 662          buffer->step_rate = 0;
 663
 664          for (i = 0; i < nr_uploads; i++) {
 665             const struct gl_array_attributes *glattrib = upload[i]->glattrib;
 666             /* Then, just point upload[i] at upload[0]'s buffer. */
 667             upload[i]->offset = ((const unsigned char *)glattrib->Ptr - ptr);
 668             upload[i]->buffer = j;
 669          }
 670          j++;
 671
 672          nr_uploads = 0;
 673       }
 674    }
 675    /* Upload non-interleaved arrays */
 676    for (i = 0; i < nr_uploads; i++) {
 677       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 678       const struct gl_vertex_buffer_binding *glbinding = upload[i]->glbinding;
 679       const struct gl_array_attributes *glattrib = upload[i]->glattrib;
 680       if (glbinding->InstanceDivisor == 0) {
 681          copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 682                                  buffer, glattrib->Format._ElementSize);
 683       } else {
 684          /* This is an instanced attribute, since its InstanceDivisor
 685           * is not zero. Therefore, its data will be stepped after the
 686           * instanced draw has been run InstanceDivisor times.
 687           */
 688          uint32_t instanced_attr_max_index =
 689             (brw->num_instances - 1) / glbinding->InstanceDivisor;
 690          copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
 691                                  buffer, glattrib->Format._ElementSize);
 692       }
 693       buffer->offset -= delta * buffer->stride;
 694       buffer->size += delta * buffer->stride;
 695       buffer->step_rate = glbinding->InstanceDivisor;
 696       upload[i]->buffer = j++;
 697       upload[i]->offset = 0;
 698    }
 699
 700    brw->vb.nr_buffers = j;
 701 }
 702
 703 void
 704 brw_prepare_shader_draw_parameters(struct brw_context *brw)
 705 {
 706    const struct brw_vs_prog_data *vs_prog_data =
 707       brw_vs_prog_data(brw->vs.base.prog_data);
 708
 709    /* For non-indirect draws, upload the shader draw parameters */
 710    if ((vs_prog_data->uses_firstvertex || vs_prog_data->uses_baseinstance) &&
 711        brw->draw.draw_params_bo == NULL) {
 712       brw_upload_data(&brw->upload,
 713                       &brw->draw.params, sizeof(brw->draw.params), 4,
 714                       &brw->draw.draw_params_bo,
 715                       &brw->draw.draw_params_offset);
 716    }
 717
 718    if (vs_prog_data->uses_drawid || vs_prog_data->uses_is_indexed_draw) {
 719       brw_upload_data(&brw->upload,
 720                       &brw->draw.derived_params, sizeof(brw->draw.derived_params), 4,
 721                       &brw->draw.derived_draw_params_bo,
 722                       &brw->draw.derived_draw_params_offset);
 723    }
 724 }
 725
 726 static void
 727 brw_upload_indices(struct brw_context *brw)
 728 {
 729    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 730    GLuint ib_size;
 731    struct brw_bo *old_bo = brw->ib.bo;
 732    struct gl_buffer_object *bufferobj;
 733    GLuint offset;
 734    GLuint ib_type_size;
 735
 736    if (index_buffer == NULL)
 737       return;
 738
 739    ib_type_size = index_buffer->index_size;
 740    ib_size = index_buffer->count ? ib_type_size * index_buffer->count :
 741                                    index_buffer->obj->Size;
 742    bufferobj = index_buffer->obj;
 743
 744    /* Turn into a proper VBO:
 745     */
 746    if (!_mesa_is_bufferobj(bufferobj)) {
 747       /* Get new bufferobj, offset:
 748        */
 749       brw_upload_data(&brw->upload, index_buffer->ptr, ib_size, ib_type_size,
 750                       &brw->ib.bo, &offset);
 751       brw->ib.size = brw->ib.bo->size;
 752    } else {
 753       offset = (GLuint) (unsigned long) index_buffer->ptr;
 754
 755       struct brw_bo *bo =
 756          intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 757                                 offset, ib_size, false);
 758       if (bo != brw->ib.bo) {
 759          brw_bo_unreference(brw->ib.bo);
 760          brw->ib.bo = bo;
 761          brw->ib.size = bufferobj->Size;
 762          brw_bo_reference(bo);
 763       }
 764    }
 765
 766    /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
 767     * the index buffer state when we're just moving the start index
 768     * of our drawing.
 769     */
 770    brw->ib.start_vertex_offset = offset / ib_type_size;
 771
 772    if (brw->ib.bo != old_bo)
 773       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 774
 775    if (index_buffer->index_size != brw->ib.index_size) {
 776       brw->ib.index_size = index_buffer->index_size;
 777       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 778    }
 779
 780    /* We need to re-emit an index buffer state each time
 781     * when cut index flag is changed
 782     */
 783    if (brw->prim_restart.enable_cut_index != brw->ib.enable_cut_index) {
 784       brw->ib.enable_cut_index = brw->prim_restart.enable_cut_index;
 785       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 786    }
 787 }
 788
 789 const struct brw_tracked_state brw_indices = {
 790    .dirty = {
 791       .mesa = 0,
 792       .brw = BRW_NEW_BLORP |
 793              BRW_NEW_INDICES,
 794    },
 795    .emit = brw_upload_indices,
 796 };