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                         const uint8_t *const ptr, const int src_stride,
 404                         int min, int max,
 405                         struct brw_vertex_buffer *buffer,
 406                         GLuint dst_stride)
 407 {
 408    const unsigned char *src = ptr + min * src_stride;
 409    int count = max - min + 1;
 410    GLuint size = count * dst_stride;
 411    uint8_t *dst = brw_upload_space(&brw->upload, size, dst_stride,
 412                                    &buffer->bo, &buffer->offset);
 413
 414    /* The GL 4.5 spec says:
 415     *      "If any enabled array’s buffer binding is zero when DrawArrays or
 416     *      one of the other drawing commands defined in section 10.4 is called,
 417     *      the result is undefined."
 418     *
 419     * In this case, let's the dst with undefined values
 420     */
 421    if (ptr != NULL) {
 422       if (dst_stride == src_stride) {
 423          memcpy(dst, src, size);
 424       } else {
 425          while (count--) {
 426             memcpy(dst, src, dst_stride);
 427             src += src_stride;
 428             dst += dst_stride;
 429          }
 430       }
 431    }
 432    buffer->stride = dst_stride;
 433    buffer->size = size;
 434 }
 435
 436 void
 437 brw_prepare_vertices(struct brw_context *brw)
 438 {
 439    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 440    struct gl_context *ctx = &brw->ctx;
 441    /* BRW_NEW_VERTEX_PROGRAM */
 442    const struct gl_program *vp = brw->programs[MESA_SHADER_VERTEX];
 443    /* BRW_NEW_VS_PROG_DATA */
 444    const struct brw_vs_prog_data *vs_prog_data =
 445       brw_vs_prog_data(brw->vs.base.prog_data);
 446    const uint64_t vs_inputs64 =
 447       nir_get_single_slot_attribs_mask(vs_prog_data->inputs_read,
 448                                        vp->DualSlotInputs);
 449    assert((vs_inputs64 & ~(uint64_t)VERT_BIT_ALL) == 0);
 450    unsigned vs_inputs = (unsigned)vs_inputs64;
 451    unsigned int min_index = brw->vb.min_index + brw->basevertex;
 452    unsigned int max_index = brw->vb.max_index + brw->basevertex;
 453    int delta, j;
 454
 455    /* _NEW_POLYGON
 456     *
 457     * On gen6+, edge flags don't end up in the VUE (either in or out of the
 458     * VS).  Instead, they're uploaded as the last vertex element, and the data
 459     * is passed sideband through the fixed function units.  So, we need to
 460     * prepare the vertex buffer for it, but it's not present in inputs_read.
 461     */
 462    if (devinfo->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
 463                            ctx->Polygon.BackMode != GL_FILL)) {
 464       vs_inputs |= VERT_BIT_EDGEFLAG;
 465    }
 466
 467    if (0)
 468       fprintf(stderr, "%s %d..%d\n", __func__, min_index, max_index);
 469
 470    /* Accumulate the list of enabled arrays. */
 471    brw->vb.nr_enabled = 0;
 472
 473    unsigned mask = vs_inputs;
 474    while (mask) {
 475       const gl_vert_attrib attr = u_bit_scan(&mask);
 476       struct brw_vertex_element *input = &brw->vb.inputs[attr];
 477       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 478    }
 479    assert(brw->vb.nr_enabled <= VERT_ATTRIB_MAX);
 480
 481    if (brw->vb.nr_enabled == 0)
 482       return;
 483
 484    if (brw->vb.nr_buffers)
 485       return;
 486
 487    j = 0;
 488    const struct gl_vertex_array_object *vao = ctx->Array._DrawVAO;
 489
 490    unsigned vbomask = vs_inputs & _mesa_draw_vbo_array_bits(ctx);
 491    while (vbomask) {
 492       const struct gl_vertex_buffer_binding *const glbinding =
 493          _mesa_draw_buffer_binding(vao, ffs(vbomask) - 1);
 494       const GLsizei stride = glbinding->Stride;
 495
 496       assert(glbinding->BufferObj);
 497
 498       /* Accumulate the range of a single vertex, start with inverted range */
 499       uint32_t vertex_range_start = ~(uint32_t)0;
 500       uint32_t vertex_range_end = 0;
 501
 502       const unsigned boundmask = _mesa_draw_bound_attrib_bits(glbinding);
 503       unsigned attrmask = vbomask & boundmask;
 504       /* Mark the those attributes as processed */
 505       vbomask ^= attrmask;
 506       /* We can assume that we have an array for the binding */
 507       assert(attrmask);
 508       /* Walk attributes belonging to the binding */
 509       while (attrmask) {
 510          const gl_vert_attrib attr = u_bit_scan(&attrmask);
 511          const struct gl_array_attributes *const glattrib =
 512             _mesa_draw_array_attrib(vao, attr);
 513          const uint32_t rel_offset =
 514             _mesa_draw_attributes_relative_offset(glattrib);
 515          const uint32_t rel_end = rel_offset + glattrib->Format._ElementSize;
 516
 517          vertex_range_start = MIN2(vertex_range_start, rel_offset);
 518          vertex_range_end = MAX2(vertex_range_end, rel_end);
 519
 520          struct brw_vertex_element *input = &brw->vb.inputs[attr];
 521          input->glformat = &glattrib->Format;
 522          input->buffer = j;
 523          input->is_dual_slot = (vp->DualSlotInputs & BITFIELD64_BIT(attr)) != 0;
 524          input->offset = rel_offset;
 525       }
 526       assert(vertex_range_start <= vertex_range_end);
 527
 528       struct intel_buffer_object *intel_buffer =
 529          intel_buffer_object(glbinding->BufferObj);
 530       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 531
 532       const uint32_t offset = _mesa_draw_binding_offset(glbinding);
 533
 534       /* If nothing else is known take the buffer size and offset as a bound */
 535       uint32_t start = vertex_range_start;
 536       uint32_t range = intel_buffer->Base.Size - offset - vertex_range_start;
 537       /* Check if we can get a more narrow range */
 538       if (glbinding->InstanceDivisor) {
 539          if (brw->num_instances) {
 540             const uint32_t vertex_size = vertex_range_end - vertex_range_start;
 541             start = vertex_range_start + stride * brw->baseinstance;
 542             range = (stride * ((brw->num_instances - 1) /
 543                                glbinding->InstanceDivisor) +
 544                      vertex_size);
 545          }
 546       } else {
 547          if (brw->vb.index_bounds_valid) {
 548             const uint32_t vertex_size = vertex_range_end - vertex_range_start;
 549             start = vertex_range_start + stride * min_index;
 550             range = (stride * (max_index - min_index) +
 551                      vertex_size);
 552
 553             /**
 554              * Unreal Engine 4 has a bug in usage of glDrawRangeElements,
 555              * causing it to be called with a number of vertices in place
 556              * of "end" parameter (which specifies the maximum array index
 557              * contained in indices).
 558              *
 559              * Since there is unknown amount of games affected and we
 560              * could not identify that a game is built with UE4 - we are
 561              * forced to make a blanket workaround, disregarding max_index
 562              * in range calculations. Fortunately all such calls look like:
 563              *   glDrawRangeElements(GL_TRIANGLES, 0, 3, 3, ...);
 564              * So we are able to narrow down this workaround.
 565              *
 566              * See: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2917
 567              */
 568             if (unlikely(max_index == 3 && min_index == 0 &&
 569                          brw->draw.derived_params.is_indexed_draw)) {
 570                   range = intel_buffer->Base.Size - offset - start;
 571             }
 572          }
 573       }
 574
 575       buffer->offset = offset;
 576       buffer->size = start + range;
 577       buffer->stride = stride;
 578       buffer->step_rate = glbinding->InstanceDivisor;
 579
 580       buffer->bo = intel_bufferobj_buffer(brw, intel_buffer, offset + start,
 581                                           range, false);
 582       brw_bo_reference(buffer->bo);
 583
 584       j++;
 585    }
 586
 587    /* If we need to upload all the arrays, then we can trim those arrays to
 588     * only the used elements [min_index, max_index] so long as we adjust all
 589     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 590     */
 591    brw->vb.start_vertex_bias = 0;
 592    delta = min_index;
 593    if ((vs_inputs & _mesa_draw_vbo_array_bits(ctx)) == 0) {
 594       brw->vb.start_vertex_bias = -delta;
 595       delta = 0;
 596    }
 597
 598    unsigned usermask = vs_inputs & _mesa_draw_user_array_bits(ctx);
 599    while (usermask) {
 600       const struct gl_vertex_buffer_binding *const glbinding =
 601          _mesa_draw_buffer_binding(vao, ffs(usermask) - 1);
 602       const GLsizei stride = glbinding->Stride;
 603
 604       assert(!glbinding->BufferObj);
 605       assert(brw->vb.index_bounds_valid);
 606
 607       /* Accumulate the range of a single vertex, start with inverted range */
 608       uint32_t vertex_range_start = ~(uint32_t)0;
 609       uint32_t vertex_range_end = 0;
 610
 611       const unsigned boundmask = _mesa_draw_bound_attrib_bits(glbinding);
 612       unsigned attrmask = usermask & boundmask;
 613       /* Mark the those attributes as processed */
 614       usermask ^= attrmask;
 615       /* We can assume that we have an array for the binding */
 616       assert(attrmask);
 617       /* Walk attributes belonging to the binding */
 618       while (attrmask) {
 619          const gl_vert_attrib attr = u_bit_scan(&attrmask);
 620          const struct gl_array_attributes *const glattrib =
 621             _mesa_draw_array_attrib(vao, attr);
 622          const uint32_t rel_offset =
 623             _mesa_draw_attributes_relative_offset(glattrib);
 624          const uint32_t rel_end = rel_offset + glattrib->Format._ElementSize;
 625
 626          vertex_range_start = MIN2(vertex_range_start, rel_offset);
 627          vertex_range_end = MAX2(vertex_range_end, rel_end);
 628
 629          struct brw_vertex_element *input = &brw->vb.inputs[attr];
 630          input->glformat = &glattrib->Format;
 631          input->buffer = j;
 632          input->is_dual_slot = (vp->DualSlotInputs & BITFIELD64_BIT(attr)) != 0;
 633          input->offset = rel_offset;
 634       }
 635       assert(vertex_range_start <= vertex_range_end);
 636
 637       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 638
 639       const uint8_t *ptr = (const uint8_t*)_mesa_draw_binding_offset(glbinding);
 640       ptr += vertex_range_start;
 641       const uint32_t vertex_size = vertex_range_end - vertex_range_start;
 642       if (glbinding->Stride == 0) {
 643          /* If the source stride is zero, we just want to upload the current
 644           * attribute once and set the buffer's stride to 0.  There's no need
 645           * to replicate it out.
 646           */
 647          copy_array_to_vbo_array(brw, ptr, 0, 0, 0, buffer, vertex_size);
 648       } else if (glbinding->InstanceDivisor == 0) {
 649          copy_array_to_vbo_array(brw, ptr, stride, min_index,
 650                                  max_index, buffer, vertex_size);
 651       } else {
 652          /* This is an instanced attribute, since its InstanceDivisor
 653           * is not zero. Therefore, its data will be stepped after the
 654           * instanced draw has been run InstanceDivisor times.
 655           */
 656          uint32_t instanced_attr_max_index =
 657             (brw->num_instances - 1) / glbinding->InstanceDivisor;
 658          copy_array_to_vbo_array(brw, ptr, stride, 0,
 659                                  instanced_attr_max_index, buffer, vertex_size);
 660       }
 661       buffer->offset -= delta * buffer->stride + vertex_range_start;
 662       buffer->size += delta * buffer->stride + vertex_range_start;
 663       buffer->step_rate = glbinding->InstanceDivisor;
 664
 665       j++;
 666    }
 667
 668    /* Upload the current values */
 669    unsigned curmask = vs_inputs & _mesa_draw_current_bits(ctx);
 670    if (curmask) {
 671       /* For each attribute, upload the maximum possible size. */
 672       uint8_t data[VERT_ATTRIB_MAX * sizeof(GLdouble) * 4];
 673       uint8_t *cursor = data;
 674
 675       do {
 676          const gl_vert_attrib attr = u_bit_scan(&curmask);
 677          const struct gl_array_attributes *const glattrib =
 678             _mesa_draw_current_attrib(ctx, attr);
 679          const unsigned size = glattrib->Format._ElementSize;
 680          const unsigned alignment = align(size, sizeof(GLdouble));
 681          memcpy(cursor, glattrib->Ptr, size);
 682          if (alignment != size)
 683             memset(cursor + size, 0, alignment - size);
 684
 685          struct brw_vertex_element *input = &brw->vb.inputs[attr];
 686          input->glformat = &glattrib->Format;
 687          input->buffer = j;
 688          input->is_dual_slot = (vp->DualSlotInputs & BITFIELD64_BIT(attr)) != 0;
 689          input->offset = cursor - data;
 690
 691          cursor += alignment;
 692       } while (curmask);
 693
 694       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 695       const unsigned size = cursor - data;
 696       brw_upload_data(&brw->upload, data, size, size,
 697                       &buffer->bo, &buffer->offset);
 698       buffer->stride = 0;
 699       buffer->size = size;
 700       buffer->step_rate = 0;
 701
 702       j++;
 703    }
 704    brw->vb.nr_buffers = j;
 705 }
 706
 707 void
 708 brw_prepare_shader_draw_parameters(struct brw_context *brw)
 709 {
 710    const struct brw_vs_prog_data *vs_prog_data =
 711       brw_vs_prog_data(brw->vs.base.prog_data);
 712
 713    /* For non-indirect draws, upload the shader draw parameters */
 714    if ((vs_prog_data->uses_firstvertex || vs_prog_data->uses_baseinstance) &&
 715        brw->draw.draw_params_bo == NULL) {
 716       brw_upload_data(&brw->upload,
 717                       &brw->draw.params, sizeof(brw->draw.params), 4,
 718                       &brw->draw.draw_params_bo,
 719                       &brw->draw.draw_params_offset);
 720    }
 721
 722    if (vs_prog_data->uses_drawid || vs_prog_data->uses_is_indexed_draw) {
 723       brw_upload_data(&brw->upload,
 724                       &brw->draw.derived_params, sizeof(brw->draw.derived_params), 4,
 725                       &brw->draw.derived_draw_params_bo,
 726                       &brw->draw.derived_draw_params_offset);
 727    }
 728 }
 729
 730 static void
 731 brw_upload_indices(struct brw_context *brw)
 732 {
 733    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 734    GLuint ib_size;
 735    struct brw_bo *old_bo = brw->ib.bo;
 736    struct gl_buffer_object *bufferobj;
 737    GLuint offset;
 738    GLuint ib_type_size;
 739
 740    if (index_buffer == NULL)
 741       return;
 742
 743    ib_type_size = 1 << index_buffer->index_size_shift;
 744    ib_size = index_buffer->count ? ib_type_size * index_buffer->count :
 745                                    index_buffer->obj->Size;
 746    bufferobj = index_buffer->obj;
 747
 748    /* Turn into a proper VBO:
 749     */
 750    if (!bufferobj) {
 751       /* Get new bufferobj, offset:
 752        */
 753       brw_upload_data(&brw->upload, index_buffer->ptr, ib_size, ib_type_size,
 754                       &brw->ib.bo, &offset);
 755       brw->ib.size = brw->ib.bo->size;
 756    } else {
 757       offset = (GLuint) (unsigned long) index_buffer->ptr;
 758
 759       struct brw_bo *bo =
 760          intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 761                                 offset, ib_size, false);
 762       if (bo != brw->ib.bo) {
 763          brw_bo_unreference(brw->ib.bo);
 764          brw->ib.bo = bo;
 765          brw->ib.size = bufferobj->Size;
 766          brw_bo_reference(bo);
 767       }
 768    }
 769
 770    /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
 771     * the index buffer state when we're just moving the start index
 772     * of our drawing.
 773     */
 774    brw->ib.start_vertex_offset = offset / ib_type_size;
 775
 776    if (brw->ib.bo != old_bo)
 777       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 778
 779    unsigned index_size = 1 << index_buffer->index_size_shift;
 780    if (index_size != brw->ib.index_size) {
 781       brw->ib.index_size = index_size;
 782       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 783    }
 784
 785    /* We need to re-emit an index buffer state each time
 786     * when cut index flag is changed
 787     */
 788    if (brw->prim_restart.enable_cut_index != brw->ib.enable_cut_index) {
 789       brw->ib.enable_cut_index = brw->prim_restart.enable_cut_index;
 790       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 791    }
 792 }
 793
 794 const struct brw_tracked_state brw_indices = {
 795    .dirty = {
 796       .mesa = 0,
 797       .brw = BRW_NEW_BLORP |
 798              BRW_NEW_INDICES,
 799    },
 800    .emit = brw_upload_indices,
 801 };