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(int size, GLboolean doubles)
 226 {
 227    /* From the BDW PRM, Volume 2d, page 588 (VERTEX_ELEMENT_STATE):
 228     * "When SourceElementFormat is set to one of the *64*_PASSTHRU formats,
 229     * 64-bit components are stored in the URB without any conversion."
 230     * Also included on BDW PRM, Volume 7, page 470, table "Source Element
 231     * Formats Supported in VF Unit"
 232     *
 233     * Previous PRMs don't include those references, so for gen7 we can't use
 234     * PASSTHRU formats directly. But in any case, we prefer to return passthru
 235     * even in that case, because that reflects what we want to achieve, even
 236     * if we would need to workaround on gen < 8.
 237     */
 238    return (doubles
 239            ? double_types_passthru[size]
 240            : double_types_float[size]);
 241 }
 242
 243 /**
 244  * Given vertex array type/size/format/normalized info, return
 245  * the appopriate hardware surface type.
 246  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 247  */
 248 unsigned
 249 brw_get_vertex_surface_type(struct brw_context *brw,
 250                             const struct gl_array_attributes *glattrib)
 251 {
 252    int size = glattrib->Size;
 253    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 254    const bool is_ivybridge_or_older =
 255       devinfo->gen <= 7 && !devinfo->is_baytrail && !devinfo->is_haswell;
 256
 257    if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
 258       fprintf(stderr, "type %s size %d normalized %d\n",
 259               _mesa_enum_to_string(glattrib->Type),
 260               glattrib->Size, glattrib->Normalized);
 261
 262    if (glattrib->Integer) {
 263       assert(glattrib->Format == GL_RGBA); /* sanity check */
 264       switch (glattrib->Type) {
 265       case GL_INT: return int_types_direct[size];
 266       case GL_SHORT:
 267          if (is_ivybridge_or_older && size == 3)
 268             return short_types_direct[4];
 269          else
 270             return short_types_direct[size];
 271       case GL_BYTE:
 272          if (is_ivybridge_or_older && size == 3)
 273             return byte_types_direct[4];
 274          else
 275             return byte_types_direct[size];
 276       case GL_UNSIGNED_INT: return uint_types_direct[size];
 277       case GL_UNSIGNED_SHORT:
 278          if (is_ivybridge_or_older && size == 3)
 279             return ushort_types_direct[4];
 280          else
 281             return ushort_types_direct[size];
 282       case GL_UNSIGNED_BYTE:
 283          if (is_ivybridge_or_older && size == 3)
 284             return ubyte_types_direct[4];
 285          else
 286             return ubyte_types_direct[size];
 287       default: unreachable("not reached");
 288       }
 289    } else if (glattrib->Type == GL_UNSIGNED_INT_10F_11F_11F_REV) {
 290       return ISL_FORMAT_R11G11B10_FLOAT;
 291    } else if (glattrib->Normalized) {
 292       switch (glattrib->Type) {
 293       case GL_DOUBLE: return double_types(size, glattrib->Doubles);
 294       case GL_FLOAT: return float_types[size];
 295       case GL_HALF_FLOAT:
 296       case GL_HALF_FLOAT_OES:
 297          if (devinfo->gen < 6 && size == 3)
 298             return half_float_types[4];
 299          else
 300             return half_float_types[size];
 301       case GL_INT: return int_types_norm[size];
 302       case GL_SHORT: return short_types_norm[size];
 303       case GL_BYTE: return byte_types_norm[size];
 304       case GL_UNSIGNED_INT: return uint_types_norm[size];
 305       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 306       case GL_UNSIGNED_BYTE:
 307          if (glattrib->Format == GL_BGRA) {
 308             /* See GL_EXT_vertex_array_bgra */
 309             assert(size == 4);
 310             return ISL_FORMAT_B8G8R8A8_UNORM;
 311          }
 312          else {
 313             return ubyte_types_norm[size];
 314          }
 315       case GL_FIXED:
 316          if (devinfo->gen >= 8 || devinfo->is_haswell)
 317             return fixed_point_types[size];
 318
 319          /* This produces GL_FIXED inputs as values between INT32_MIN and
 320           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 321           */
 322          return int_types_scale[size];
 323       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 324        * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
 325        * like to use here, so upload everything as UINT and fix
 326        * it in the shader
 327        */
 328       case GL_INT_2_10_10_10_REV:
 329          assert(size == 4);
 330          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 331             return glattrib->Format == GL_BGRA
 332                ? ISL_FORMAT_B10G10R10A2_SNORM
 333                : ISL_FORMAT_R10G10B10A2_SNORM;
 334          }
 335          return ISL_FORMAT_R10G10B10A2_UINT;
 336       case GL_UNSIGNED_INT_2_10_10_10_REV:
 337          assert(size == 4);
 338          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 339             return glattrib->Format == GL_BGRA
 340                ? ISL_FORMAT_B10G10R10A2_UNORM
 341                : ISL_FORMAT_R10G10B10A2_UNORM;
 342          }
 343          return ISL_FORMAT_R10G10B10A2_UINT;
 344       default: unreachable("not reached");
 345       }
 346    }
 347    else {
 348       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 349        * W/A: the hardware doesn't really support the formats we'd
 350        * like to use here, so upload everything as UINT and fix
 351        * it in the shader
 352        */
 353       if (glattrib->Type == GL_INT_2_10_10_10_REV) {
 354          assert(size == 4);
 355          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 356             return glattrib->Format == GL_BGRA
 357                ? ISL_FORMAT_B10G10R10A2_SSCALED
 358                : ISL_FORMAT_R10G10B10A2_SSCALED;
 359          }
 360          return ISL_FORMAT_R10G10B10A2_UINT;
 361       } else if (glattrib->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 362          assert(size == 4);
 363          if (devinfo->gen >= 8 || devinfo->is_haswell) {
 364             return glattrib->Format == GL_BGRA
 365                ? ISL_FORMAT_B10G10R10A2_USCALED
 366                : ISL_FORMAT_R10G10B10A2_USCALED;
 367          }
 368          return ISL_FORMAT_R10G10B10A2_UINT;
 369       }
 370       assert(glattrib->Format == GL_RGBA); /* sanity check */
 371       switch (glattrib->Type) {
 372       case GL_DOUBLE: return double_types(size, glattrib->Doubles);
 373       case GL_FLOAT: return float_types[size];
 374       case GL_HALF_FLOAT:
 375       case GL_HALF_FLOAT_OES:
 376          if (devinfo->gen < 6 && size == 3)
 377             return half_float_types[4];
 378          else
 379             return half_float_types[size];
 380       case GL_INT: return int_types_scale[size];
 381       case GL_SHORT: return short_types_scale[size];
 382       case GL_BYTE: return byte_types_scale[size];
 383       case GL_UNSIGNED_INT: return uint_types_scale[size];
 384       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 385       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 386       case GL_FIXED:
 387          if (devinfo->gen >= 8 || devinfo->is_haswell)
 388             return fixed_point_types[size];
 389
 390          /* This produces GL_FIXED inputs as values between INT32_MIN and
 391           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 392           */
 393          return int_types_scale[size];
 394       default: unreachable("not reached");
 395       }
 396    }
 397 }
 398
 399 static void
 400 copy_array_to_vbo_array(struct brw_context *brw,
 401                         struct brw_vertex_element *element,
 402                         int min, int max,
 403                         struct brw_vertex_buffer *buffer,
 404                         GLuint dst_stride)
 405 {
 406    const struct gl_vertex_array *glarray = element->glarray;
 407    const struct gl_vertex_buffer_binding *glbinding = glarray->BufferBinding;
 408    const struct gl_array_attributes *glattrib = glarray->VertexAttrib;
 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_array *glarray = input->glarray;
 516       const struct gl_vertex_buffer_binding *glbinding = glarray->BufferBinding;
 517       const struct gl_array_attributes *glattrib = glarray->VertexAttrib;
 518
 519       if (_mesa_is_bufferobj(glbinding->BufferObj)) {
 520          struct intel_buffer_object *intel_buffer =
 521             intel_buffer_object(glbinding->BufferObj);
 522
 523          const uint32_t offset = glbinding->Offset + glattrib->RelativeOffset;
 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             const struct gl_vertex_array *other = brw->vb.enabled[k]->glarray;
 550             const struct gl_vertex_buffer_binding *obind = other->BufferBinding;
 551             const struct gl_array_attributes *oattrib = other->VertexAttrib;
 552             const uint32_t ooffset = obind->Offset + oattrib->RelativeOffset;
 553             if (glbinding->BufferObj == obind->BufferObj &&
 554                 glbinding->Stride == obind->Stride &&
 555                 glbinding->InstanceDivisor == obind->InstanceDivisor &&
 556                 (offset - ooffset) < glbinding->Stride)
 557             {
 558                input->buffer = brw->vb.enabled[k]->buffer;
 559                input->offset = offset - ooffset;
 560
 561                buffer_range_start[input->buffer] =
 562                   MIN2(buffer_range_start[input->buffer], start);
 563                buffer_range_end[input->buffer] =
 564                   MAX2(buffer_range_end[input->buffer], start + range);
 565                break;
 566             }
 567          }
 568          if (k == i) {
 569             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 570
 571             /* Named buffer object: Just reference its contents directly. */
 572             buffer->offset = offset;
 573             buffer->stride = glbinding->Stride;
 574             buffer->step_rate = glbinding->InstanceDivisor;
 575             buffer->size = glbinding->BufferObj->Size - offset;
 576
 577             enabled_buffer[j] = intel_buffer;
 578             buffer_range_start[j] = start;
 579             buffer_range_end[j] = start + range;
 580
 581             input->buffer = j++;
 582             input->offset = 0;
 583          }
 584       } else {
 585          /* Queue the buffer object up to be uploaded in the next pass,
 586           * when we've decided if we're doing interleaved or not.
 587           */
 588          if (nr_uploads == 0) {
 589             interleaved = glbinding->Stride;
 590             ptr = glattrib->Ptr;
 591          }
 592          else if (interleaved != glbinding->Stride ||
 593                   glbinding->InstanceDivisor != 0 ||
 594                   glattrib->Ptr < ptr ||
 595                   (uintptr_t)(glattrib->Ptr - ptr) + glattrib->_ElementSize > interleaved)
 596          {
 597             /* If our stride is different from the first attribute's stride,
 598              * or if we are using an instance divisor or if the first
 599              * attribute's stride didn't cover our element, disable the
 600              * interleaved upload optimization.  The second case can most
 601              * commonly occur in cases where there is a single vertex and, for
 602              * example, the data is stored on the application's stack.
 603              *
 604              * NOTE: This will also disable the optimization in cases where
 605              * the data is in a different order than the array indices.
 606              * Something like:
 607              *
 608              *     float data[...];
 609              *     glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
 610              *     glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
 611              */
 612             interleaved = 0;
 613          }
 614
 615          upload[nr_uploads++] = input;
 616       }
 617    }
 618
 619    /* Now that we've set up all of the buffers, we walk through and reference
 620     * each of them.  We do this late so that we get the right size in each
 621     * buffer and don't reference too little data.
 622     */
 623    for (i = 0; i < j; i++) {
 624       struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 625       if (buffer->bo)
 626          continue;
 627
 628       const uint32_t start = buffer_range_start[i];
 629       const uint32_t range = buffer_range_end[i] - buffer_range_start[i];
 630
 631       buffer->bo = intel_bufferobj_buffer(brw, enabled_buffer[i], start,
 632                                           range, false);
 633       brw_bo_reference(buffer->bo);
 634    }
 635
 636    /* If we need to upload all the arrays, then we can trim those arrays to
 637     * only the used elements [min_index, max_index] so long as we adjust all
 638     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 639     */
 640    brw->vb.start_vertex_bias = 0;
 641    delta = min_index;
 642    if (nr_uploads == brw->vb.nr_enabled) {
 643       brw->vb.start_vertex_bias = -delta;
 644       delta = 0;
 645    }
 646
 647    /* Handle any arrays to be uploaded. */
 648    if (nr_uploads > 1) {
 649       if (interleaved) {
 650          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 651          /* All uploads are interleaved, so upload the arrays together as
 652           * interleaved.  First, upload the contents and set up upload[0].
 653           */
 654          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 655                                  buffer, interleaved);
 656          buffer->offset -= delta * interleaved;
 657          buffer->size += delta * interleaved;
 658          buffer->step_rate = 0;
 659
 660          for (i = 0; i < nr_uploads; i++) {
 661             const struct gl_vertex_array *glarray = upload[i]->glarray;
 662             const struct gl_array_attributes *glattrib = glarray->VertexAttrib;
 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_array *glarray = upload[i]->glarray;
 676       const struct gl_vertex_buffer_binding *glbinding = glarray->BufferBinding;
 677       const struct gl_array_attributes *glattrib = glarray->VertexAttrib;
 678       if (glbinding->InstanceDivisor == 0) {
 679          copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 680                                  buffer, glattrib->_ElementSize);
 681       } else {
 682          /* This is an instanced attribute, since its InstanceDivisor
 683           * is not zero. Therefore, its data will be stepped after the
 684           * instanced draw has been run InstanceDivisor times.
 685           */
 686          uint32_t instanced_attr_max_index =
 687             (brw->num_instances - 1) / glbinding->InstanceDivisor;
 688          copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
 689                                  buffer, glattrib->_ElementSize);
 690       }
 691       buffer->offset -= delta * buffer->stride;
 692       buffer->size += delta * buffer->stride;
 693       buffer->step_rate = glbinding->InstanceDivisor;
 694       upload[i]->buffer = j++;
 695       upload[i]->offset = 0;
 696    }
 697
 698    brw->vb.nr_buffers = j;
 699 }
 700
 701 void
 702 brw_prepare_shader_draw_parameters(struct brw_context *brw)
 703 {
 704    const struct brw_vs_prog_data *vs_prog_data =
 705       brw_vs_prog_data(brw->vs.base.prog_data);
 706
 707    const bool uses_firstvertex =
 708       vs_prog_data->uses_basevertex || vs_prog_data->uses_firstvertex;
 709
 710    /* For non-indirect draws, upload the shader draw parameters */
 711    if ((uses_firstvertex || vs_prog_data->uses_baseinstance) &&
 712        brw->draw.draw_params_bo == NULL) {
 713       brw_upload_data(&brw->upload,
 714                       &brw->draw.params, sizeof(brw->draw.params), 4,
 715                       &brw->draw.draw_params_bo,
 716                       &brw->draw.draw_params_offset);
 717    }
 718
 719    if (vs_prog_data->uses_drawid || vs_prog_data->uses_is_indexed_draw) {
 720       brw_upload_data(&brw->upload,
 721                       &brw->draw.derived_params, sizeof(brw->draw.derived_params), 4,
 722                       &brw->draw.derived_draw_params_bo,
 723                       &brw->draw.derived_draw_params_offset);
 724    }
 725 }
 726
 727 static void
 728 brw_upload_indices(struct brw_context *brw)
 729 {
 730    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 731    GLuint ib_size;
 732    struct brw_bo *old_bo = brw->ib.bo;
 733    struct gl_buffer_object *bufferobj;
 734    GLuint offset;
 735    GLuint ib_type_size;
 736
 737    if (index_buffer == NULL)
 738       return;
 739
 740    ib_type_size = index_buffer->index_size;
 741    ib_size = index_buffer->count ? ib_type_size * index_buffer->count :
 742                                    index_buffer->obj->Size;
 743    bufferobj = index_buffer->obj;
 744
 745    /* Turn into a proper VBO:
 746     */
 747    if (!_mesa_is_bufferobj(bufferobj)) {
 748       /* Get new bufferobj, offset:
 749        */
 750       brw_upload_data(&brw->upload, index_buffer->ptr, ib_size, ib_type_size,
 751                       &brw->ib.bo, &offset);
 752       brw->ib.size = brw->ib.bo->size;
 753    } else {
 754       offset = (GLuint) (unsigned long) index_buffer->ptr;
 755
 756       struct brw_bo *bo =
 757          intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 758                                 offset, ib_size, false);
 759       if (bo != brw->ib.bo) {
 760          brw_bo_unreference(brw->ib.bo);
 761          brw->ib.bo = bo;
 762          brw->ib.size = bufferobj->Size;
 763          brw_bo_reference(bo);
 764       }
 765    }
 766
 767    /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
 768     * the index buffer state when we're just moving the start index
 769     * of our drawing.
 770     */
 771    brw->ib.start_vertex_offset = offset / ib_type_size;
 772
 773    if (brw->ib.bo != old_bo)
 774       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 775
 776    if (index_buffer->index_size != brw->ib.index_size) {
 777       brw->ib.index_size = index_buffer->index_size;
 778       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 779    }
 780 }
 781
 782 const struct brw_tracked_state brw_indices = {
 783    .dirty = {
 784       .mesa = 0,
 785       .brw = BRW_NEW_BLORP |
 786              BRW_NEW_INDICES,
 787    },
 788    .emit = brw_upload_indices,
 789 };