src/mesa/drivers/dri/i965/brw_draw_upload.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "main/glheader.h"
  29 #include "main/bufferobj.h"
  30 #include "main/context.h"
  31 #include "main/enums.h"
  32 #include "main/macros.h"
  33 #include "main/glformats.h"
  34
  35 #include "brw_draw.h"
  36 #include "brw_defines.h"
  37 #include "brw_context.h"
  38 #include "brw_state.h"
  39
  40 #include "intel_batchbuffer.h"
  41 #include "intel_buffer_objects.h"
  42
  43 static GLuint double_types[5] = {
  44    0,
  45    BRW_SURFACEFORMAT_R64_FLOAT,
  46    BRW_SURFACEFORMAT_R64G64_FLOAT,
  47    BRW_SURFACEFORMAT_R64G64B64_FLOAT,
  48    BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
  49 };
  50
  51 static GLuint float_types[5] = {
  52    0,
  53    BRW_SURFACEFORMAT_R32_FLOAT,
  54    BRW_SURFACEFORMAT_R32G32_FLOAT,
  55    BRW_SURFACEFORMAT_R32G32B32_FLOAT,
  56    BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
  57 };
  58
  59 static GLuint half_float_types[5] = {
  60    0,
  61    BRW_SURFACEFORMAT_R16_FLOAT,
  62    BRW_SURFACEFORMAT_R16G16_FLOAT,
  63    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT,
  64    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
  65 };
  66
  67 static GLuint fixed_point_types[5] = {
  68    0,
  69    BRW_SURFACEFORMAT_R32_SFIXED,
  70    BRW_SURFACEFORMAT_R32G32_SFIXED,
  71    BRW_SURFACEFORMAT_R32G32B32_SFIXED,
  72    BRW_SURFACEFORMAT_R32G32B32A32_SFIXED,
  73 };
  74
  75 static GLuint uint_types_direct[5] = {
  76    0,
  77    BRW_SURFACEFORMAT_R32_UINT,
  78    BRW_SURFACEFORMAT_R32G32_UINT,
  79    BRW_SURFACEFORMAT_R32G32B32_UINT,
  80    BRW_SURFACEFORMAT_R32G32B32A32_UINT
  81 };
  82
  83 static GLuint uint_types_norm[5] = {
  84    0,
  85    BRW_SURFACEFORMAT_R32_UNORM,
  86    BRW_SURFACEFORMAT_R32G32_UNORM,
  87    BRW_SURFACEFORMAT_R32G32B32_UNORM,
  88    BRW_SURFACEFORMAT_R32G32B32A32_UNORM
  89 };
  90
  91 static GLuint uint_types_scale[5] = {
  92    0,
  93    BRW_SURFACEFORMAT_R32_USCALED,
  94    BRW_SURFACEFORMAT_R32G32_USCALED,
  95    BRW_SURFACEFORMAT_R32G32B32_USCALED,
  96    BRW_SURFACEFORMAT_R32G32B32A32_USCALED
  97 };
  98
  99 static GLuint int_types_direct[5] = {
 100    0,
 101    BRW_SURFACEFORMAT_R32_SINT,
 102    BRW_SURFACEFORMAT_R32G32_SINT,
 103    BRW_SURFACEFORMAT_R32G32B32_SINT,
 104    BRW_SURFACEFORMAT_R32G32B32A32_SINT
 105 };
 106
 107 static GLuint int_types_norm[5] = {
 108    0,
 109    BRW_SURFACEFORMAT_R32_SNORM,
 110    BRW_SURFACEFORMAT_R32G32_SNORM,
 111    BRW_SURFACEFORMAT_R32G32B32_SNORM,
 112    BRW_SURFACEFORMAT_R32G32B32A32_SNORM
 113 };
 114
 115 static GLuint int_types_scale[5] = {
 116    0,
 117    BRW_SURFACEFORMAT_R32_SSCALED,
 118    BRW_SURFACEFORMAT_R32G32_SSCALED,
 119    BRW_SURFACEFORMAT_R32G32B32_SSCALED,
 120    BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
 121 };
 122
 123 static GLuint ushort_types_direct[5] = {
 124    0,
 125    BRW_SURFACEFORMAT_R16_UINT,
 126    BRW_SURFACEFORMAT_R16G16_UINT,
 127    BRW_SURFACEFORMAT_R16G16B16A16_UINT,
 128    BRW_SURFACEFORMAT_R16G16B16A16_UINT
 129 };
 130
 131 static GLuint ushort_types_norm[5] = {
 132    0,
 133    BRW_SURFACEFORMAT_R16_UNORM,
 134    BRW_SURFACEFORMAT_R16G16_UNORM,
 135    BRW_SURFACEFORMAT_R16G16B16_UNORM,
 136    BRW_SURFACEFORMAT_R16G16B16A16_UNORM
 137 };
 138
 139 static GLuint ushort_types_scale[5] = {
 140    0,
 141    BRW_SURFACEFORMAT_R16_USCALED,
 142    BRW_SURFACEFORMAT_R16G16_USCALED,
 143    BRW_SURFACEFORMAT_R16G16B16_USCALED,
 144    BRW_SURFACEFORMAT_R16G16B16A16_USCALED
 145 };
 146
 147 static GLuint short_types_direct[5] = {
 148    0,
 149    BRW_SURFACEFORMAT_R16_SINT,
 150    BRW_SURFACEFORMAT_R16G16_SINT,
 151    BRW_SURFACEFORMAT_R16G16B16A16_SINT,
 152    BRW_SURFACEFORMAT_R16G16B16A16_SINT
 153 };
 154
 155 static GLuint short_types_norm[5] = {
 156    0,
 157    BRW_SURFACEFORMAT_R16_SNORM,
 158    BRW_SURFACEFORMAT_R16G16_SNORM,
 159    BRW_SURFACEFORMAT_R16G16B16_SNORM,
 160    BRW_SURFACEFORMAT_R16G16B16A16_SNORM
 161 };
 162
 163 static GLuint short_types_scale[5] = {
 164    0,
 165    BRW_SURFACEFORMAT_R16_SSCALED,
 166    BRW_SURFACEFORMAT_R16G16_SSCALED,
 167    BRW_SURFACEFORMAT_R16G16B16_SSCALED,
 168    BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
 169 };
 170
 171 static GLuint ubyte_types_direct[5] = {
 172    0,
 173    BRW_SURFACEFORMAT_R8_UINT,
 174    BRW_SURFACEFORMAT_R8G8_UINT,
 175    BRW_SURFACEFORMAT_R8G8B8A8_UINT,
 176    BRW_SURFACEFORMAT_R8G8B8A8_UINT
 177 };
 178
 179 static GLuint ubyte_types_norm[5] = {
 180    0,
 181    BRW_SURFACEFORMAT_R8_UNORM,
 182    BRW_SURFACEFORMAT_R8G8_UNORM,
 183    BRW_SURFACEFORMAT_R8G8B8_UNORM,
 184    BRW_SURFACEFORMAT_R8G8B8A8_UNORM
 185 };
 186
 187 static GLuint ubyte_types_scale[5] = {
 188    0,
 189    BRW_SURFACEFORMAT_R8_USCALED,
 190    BRW_SURFACEFORMAT_R8G8_USCALED,
 191    BRW_SURFACEFORMAT_R8G8B8_USCALED,
 192    BRW_SURFACEFORMAT_R8G8B8A8_USCALED
 193 };
 194
 195 static GLuint byte_types_direct[5] = {
 196    0,
 197    BRW_SURFACEFORMAT_R8_SINT,
 198    BRW_SURFACEFORMAT_R8G8_SINT,
 199    BRW_SURFACEFORMAT_R8G8B8A8_SINT,
 200    BRW_SURFACEFORMAT_R8G8B8A8_SINT
 201 };
 202
 203 static GLuint byte_types_norm[5] = {
 204    0,
 205    BRW_SURFACEFORMAT_R8_SNORM,
 206    BRW_SURFACEFORMAT_R8G8_SNORM,
 207    BRW_SURFACEFORMAT_R8G8B8_SNORM,
 208    BRW_SURFACEFORMAT_R8G8B8A8_SNORM
 209 };
 210
 211 static GLuint byte_types_scale[5] = {
 212    0,
 213    BRW_SURFACEFORMAT_R8_SSCALED,
 214    BRW_SURFACEFORMAT_R8G8_SSCALED,
 215    BRW_SURFACEFORMAT_R8G8B8_SSCALED,
 216    BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
 217 };
 218
 219
 220 /**
 221  * Given vertex array type/size/format/normalized info, return
 222  * the appopriate hardware surface type.
 223  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 224  */
 225 unsigned
 226 brw_get_vertex_surface_type(struct brw_context *brw,
 227                             const struct gl_client_array *glarray)
 228 {
 229    int size = glarray->Size;
 230
 231    if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
 232       printf("type %s size %d normalized %d\n",
 233              _mesa_lookup_enum_by_nr(glarray->Type),
 234              glarray->Size, glarray->Normalized);
 235
 236    if (glarray->Integer) {
 237       assert(glarray->Format == GL_RGBA); /* sanity check */
 238       switch (glarray->Type) {
 239       case GL_INT: return int_types_direct[size];
 240       case GL_SHORT: return short_types_direct[size];
 241       case GL_BYTE: return byte_types_direct[size];
 242       case GL_UNSIGNED_INT: return uint_types_direct[size];
 243       case GL_UNSIGNED_SHORT: return ushort_types_direct[size];
 244       case GL_UNSIGNED_BYTE: return ubyte_types_direct[size];
 245       default: assert(0); return 0;
 246       }
 247    } else if (glarray->Type == GL_UNSIGNED_INT_10F_11F_11F_REV) {
 248       return BRW_SURFACEFORMAT_R11G11B10_FLOAT;
 249    } else if (glarray->Normalized) {
 250       switch (glarray->Type) {
 251       case GL_DOUBLE: return double_types[size];
 252       case GL_FLOAT: return float_types[size];
 253       case GL_HALF_FLOAT: return half_float_types[size];
 254       case GL_INT: return int_types_norm[size];
 255       case GL_SHORT: return short_types_norm[size];
 256       case GL_BYTE: return byte_types_norm[size];
 257       case GL_UNSIGNED_INT: return uint_types_norm[size];
 258       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 259       case GL_UNSIGNED_BYTE:
 260          if (glarray->Format == GL_BGRA) {
 261             /* See GL_EXT_vertex_array_bgra */
 262             assert(size == 4);
 263             return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
 264          }
 265          else {
 266             return ubyte_types_norm[size];
 267          }
 268       case GL_FIXED:
 269          if (brw->gen >= 8 || brw->is_haswell)
 270             return fixed_point_types[size];
 271
 272          /* This produces GL_FIXED inputs as values between INT32_MIN and
 273           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 274           */
 275          return int_types_scale[size];
 276       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 277        * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
 278        * like to use here, so upload everything as UINT and fix
 279        * it in the shader
 280        */
 281       case GL_INT_2_10_10_10_REV:
 282          assert(size == 4);
 283          if (brw->gen >= 8 || brw->is_haswell) {
 284             return glarray->Format == GL_BGRA
 285                ? BRW_SURFACEFORMAT_B10G10R10A2_SNORM
 286                : BRW_SURFACEFORMAT_R10G10B10A2_SNORM;
 287          }
 288          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 289       case GL_UNSIGNED_INT_2_10_10_10_REV:
 290          assert(size == 4);
 291          if (brw->gen >= 8 || brw->is_haswell) {
 292             return glarray->Format == GL_BGRA
 293                ? BRW_SURFACEFORMAT_B10G10R10A2_UNORM
 294                : BRW_SURFACEFORMAT_R10G10B10A2_UNORM;
 295          }
 296          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 297       default: assert(0); return 0;
 298       }
 299    }
 300    else {
 301       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 302        * W/A: the hardware doesn't really support the formats we'd
 303        * like to use here, so upload everything as UINT and fix
 304        * it in the shader
 305        */
 306       if (glarray->Type == GL_INT_2_10_10_10_REV) {
 307          assert(size == 4);
 308          if (brw->gen >= 8 || brw->is_haswell) {
 309             return glarray->Format == GL_BGRA
 310                ? BRW_SURFACEFORMAT_B10G10R10A2_SSCALED
 311                : BRW_SURFACEFORMAT_R10G10B10A2_SSCALED;
 312          }
 313          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 314       } else if (glarray->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 315          assert(size == 4);
 316          if (brw->gen >= 8 || brw->is_haswell) {
 317             return glarray->Format == GL_BGRA
 318                ? BRW_SURFACEFORMAT_B10G10R10A2_USCALED
 319                : BRW_SURFACEFORMAT_R10G10B10A2_USCALED;
 320          }
 321          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 322       }
 323       assert(glarray->Format == GL_RGBA); /* sanity check */
 324       switch (glarray->Type) {
 325       case GL_DOUBLE: return double_types[size];
 326       case GL_FLOAT: return float_types[size];
 327       case GL_HALF_FLOAT: return half_float_types[size];
 328       case GL_INT: return int_types_scale[size];
 329       case GL_SHORT: return short_types_scale[size];
 330       case GL_BYTE: return byte_types_scale[size];
 331       case GL_UNSIGNED_INT: return uint_types_scale[size];
 332       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 333       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 334       case GL_FIXED:
 335          if (brw->gen >= 8 || brw->is_haswell)
 336             return fixed_point_types[size];
 337
 338          /* This produces GL_FIXED inputs as values between INT32_MIN and
 339           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 340           */
 341          return int_types_scale[size];
 342       default: assert(0); return 0;
 343       }
 344    }
 345 }
 346
 347 unsigned
 348 brw_get_index_type(GLenum type)
 349 {
 350    switch (type) {
 351    case GL_UNSIGNED_BYTE:  return BRW_INDEX_BYTE;
 352    case GL_UNSIGNED_SHORT: return BRW_INDEX_WORD;
 353    case GL_UNSIGNED_INT:   return BRW_INDEX_DWORD;
 354    default: assert(0); return 0;
 355    }
 356 }
 357
 358 static void
 359 copy_array_to_vbo_array(struct brw_context *brw,
 360                         struct brw_vertex_element *element,
 361                         int min, int max,
 362                         struct brw_vertex_buffer *buffer,
 363                         GLuint dst_stride)
 364 {
 365    const int src_stride = element->glarray->StrideB;
 366
 367    /* If the source stride is zero, we just want to upload the current
 368     * attribute once and set the buffer's stride to 0.  There's no need
 369     * to replicate it out.
 370     */
 371    if (src_stride == 0) {
 372       intel_upload_data(brw, element->glarray->Ptr,
 373                         element->glarray->_ElementSize,
 374                         element->glarray->_ElementSize,
 375                         &buffer->bo, &buffer->offset);
 376
 377       buffer->stride = 0;
 378       return;
 379    }
 380
 381    const unsigned char *src = element->glarray->Ptr + min * src_stride;
 382    int count = max - min + 1;
 383    GLuint size = count * dst_stride;
 384
 385    if (dst_stride == src_stride) {
 386       intel_upload_data(brw, src, size, dst_stride,
 387                         &buffer->bo, &buffer->offset);
 388    } else {
 389       char * const map = intel_upload_map(brw, size, dst_stride);
 390       char *dst = map;
 391
 392       while (count--) {
 393          memcpy(dst, src, dst_stride);
 394          src += src_stride;
 395          dst += dst_stride;
 396       }
 397       intel_upload_unmap(brw, map, size, dst_stride,
 398                          &buffer->bo, &buffer->offset);
 399    }
 400    buffer->stride = dst_stride;
 401 }
 402
 403 static void brw_prepare_vertices(struct brw_context *brw)
 404 {
 405    struct gl_context *ctx = &brw->ctx;
 406    /* CACHE_NEW_VS_PROG */
 407    GLbitfield64 vs_inputs = brw->vs.prog_data->inputs_read;
 408    const unsigned char *ptr = NULL;
 409    GLuint interleaved = 0;
 410    unsigned int min_index = brw->vb.min_index + brw->basevertex;
 411    unsigned int max_index = brw->vb.max_index + brw->basevertex;
 412    int delta, i, j;
 413
 414    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 415    GLuint nr_uploads = 0;
 416
 417    /* _NEW_POLYGON
 418     *
 419     * On gen6+, edge flags don't end up in the VUE (either in or out of the
 420     * VS).  Instead, they're uploaded as the last vertex element, and the data
 421     * is passed sideband through the fixed function units.  So, we need to
 422     * prepare the vertex buffer for it, but it's not present in inputs_read.
 423     */
 424    if (brw->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
 425                            ctx->Polygon.BackMode != GL_FILL)) {
 426       vs_inputs |= VERT_BIT_EDGEFLAG;
 427    }
 428
 429    if (0)
 430       printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
 431
 432    /* Accumulate the list of enabled arrays. */
 433    brw->vb.nr_enabled = 0;
 434    while (vs_inputs) {
 435       GLuint i = ffsll(vs_inputs) - 1;
 436       struct brw_vertex_element *input = &brw->vb.inputs[i];
 437
 438       vs_inputs &= ~BITFIELD64_BIT(i);
 439       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 440    }
 441
 442    if (brw->vb.nr_enabled == 0)
 443       return;
 444
 445    if (brw->vb.nr_buffers)
 446       return;
 447
 448    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 449       struct brw_vertex_element *input = brw->vb.enabled[i];
 450       const struct gl_client_array *glarray = input->glarray;
 451
 452       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 453          struct intel_buffer_object *intel_buffer =
 454             intel_buffer_object(glarray->BufferObj);
 455          int k;
 456
 457          /* If we have a VB set to be uploaded for this buffer object
 458           * already, reuse that VB state so that we emit fewer
 459           * relocations.
 460           */
 461          for (k = 0; k < i; k++) {
 462             const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
 463             if (glarray->BufferObj == other->BufferObj &&
 464                 glarray->StrideB == other->StrideB &&
 465                 glarray->InstanceDivisor == other->InstanceDivisor &&
 466                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 467             {
 468                input->buffer = brw->vb.enabled[k]->buffer;
 469                input->offset = glarray->Ptr - other->Ptr;
 470                break;
 471             }
 472          }
 473          if (k == i) {
 474             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 475
 476             /* Named buffer object: Just reference its contents directly. */
 477             buffer->offset = (uintptr_t)glarray->Ptr;
 478             buffer->stride = glarray->StrideB;
 479             buffer->step_rate = glarray->InstanceDivisor;
 480
 481             uint32_t offset, size;
 482             if (glarray->InstanceDivisor) {
 483                offset = buffer->offset;
 484                size = (buffer->stride * ((brw->num_instances /
 485                                           glarray->InstanceDivisor) - 1) +
 486                        glarray->_ElementSize);
 487             } else {
 488                if (min_index == -1) {
 489                   offset = 0;
 490                   size = intel_buffer->Base.Size;
 491                } else {
 492                   offset = buffer->offset + min_index * buffer->stride;
 493                   size = (buffer->stride * (max_index - min_index) +
 494                           glarray->_ElementSize);
 495                }
 496             }
 497             buffer->bo = intel_bufferobj_buffer(brw, intel_buffer,
 498                                                 offset, size);
 499             drm_intel_bo_reference(buffer->bo);
 500
 501             input->buffer = j++;
 502             input->offset = 0;
 503          }
 504
 505          /* This is a common place to reach if the user mistakenly supplies
 506           * a pointer in place of a VBO offset.  If we just let it go through,
 507           * we may end up dereferencing a pointer beyond the bounds of the
 508           * GTT.  We would hope that the VBO's max_index would save us, but
 509           * Mesa appears to hand us min/max values not clipped to the
 510           * array object's _MaxElement, and _MaxElement frequently appears
 511           * to be wrong anyway.
 512           *
 513           * The VBO spec allows application termination in this case, and it's
 514           * probably a service to the poor programmer to do so rather than
 515           * trying to just not render.
 516           */
 517          assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
 518       } else {
 519          /* Queue the buffer object up to be uploaded in the next pass,
 520           * when we've decided if we're doing interleaved or not.
 521           */
 522          if (nr_uploads == 0) {
 523             interleaved = glarray->StrideB;
 524             ptr = glarray->Ptr;
 525          }
 526          else if (interleaved != glarray->StrideB ||
 527                   glarray->Ptr < ptr ||
 528                   (uintptr_t)(glarray->Ptr - ptr) + glarray->_ElementSize > interleaved)
 529          {
 530             /* If our stride is different from the first attribute's stride,
 531              * or if the first attribute's stride didn't cover our element,
 532              * disable the interleaved upload optimization.  The second case
 533              * can most commonly occur in cases where there is a single vertex
 534              * and, for example, the data is stored on the application's
 535              * stack.
 536              *
 537              * NOTE: This will also disable the optimization in cases where
 538              * the data is in a different order than the array indices.
 539              * Something like:
 540              *
 541              *     float data[...];
 542              *     glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
 543              *     glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
 544              */
 545             interleaved = 0;
 546          }
 547
 548          upload[nr_uploads++] = input;
 549       }
 550    }
 551
 552    /* If we need to upload all the arrays, then we can trim those arrays to
 553     * only the used elements [min_index, max_index] so long as we adjust all
 554     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 555     */
 556    brw->vb.start_vertex_bias = 0;
 557    delta = min_index;
 558    if (nr_uploads == brw->vb.nr_enabled) {
 559       brw->vb.start_vertex_bias = -delta;
 560       delta = 0;
 561    }
 562
 563    /* Handle any arrays to be uploaded. */
 564    if (nr_uploads > 1) {
 565       if (interleaved) {
 566          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 567          /* All uploads are interleaved, so upload the arrays together as
 568           * interleaved.  First, upload the contents and set up upload[0].
 569           */
 570          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 571                                  buffer, interleaved);
 572          buffer->offset -= delta * interleaved;
 573
 574          for (i = 0; i < nr_uploads; i++) {
 575             /* Then, just point upload[i] at upload[0]'s buffer. */
 576             upload[i]->offset =
 577                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 578             upload[i]->buffer = j;
 579          }
 580          j++;
 581
 582          nr_uploads = 0;
 583       }
 584    }
 585    /* Upload non-interleaved arrays */
 586    for (i = 0; i < nr_uploads; i++) {
 587       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 588       if (upload[i]->glarray->InstanceDivisor == 0) {
 589          copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 590                                  buffer, upload[i]->glarray->_ElementSize);
 591       } else {
 592          /* This is an instanced attribute, since its InstanceDivisor
 593           * is not zero. Therefore, its data will be stepped after the
 594           * instanced draw has been run InstanceDivisor times.
 595           */
 596          uint32_t instanced_attr_max_index =
 597             (brw->num_instances - 1) / upload[i]->glarray->InstanceDivisor;
 598          copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
 599                                  buffer, upload[i]->glarray->_ElementSize);
 600       }
 601       buffer->offset -= delta * buffer->stride;
 602       buffer->step_rate = upload[i]->glarray->InstanceDivisor;
 603       upload[i]->buffer = j++;
 604       upload[i]->offset = 0;
 605    }
 606
 607    brw->vb.nr_buffers = j;
 608 }
 609
 610 static void brw_emit_vertices(struct brw_context *brw)
 611 {
 612    struct gl_context *ctx = &brw->ctx;
 613    GLuint i, nr_elements;
 614
 615    brw_prepare_vertices(brw);
 616
 617    brw_emit_query_begin(brw);
 618
 619    nr_elements = brw->vb.nr_enabled + brw->vs.prog_data->uses_vertexid;
 620
 621    /* If the VS doesn't read any inputs (calculating vertex position from
 622     * a state variable for some reason, for example), emit a single pad
 623     * VERTEX_ELEMENT struct and bail.
 624     *
 625     * The stale VB state stays in place, but they don't do anything unless
 626     * a VE loads from them.
 627     */
 628    if (nr_elements == 0) {
 629       BEGIN_BATCH(3);
 630       OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | 1);
 631       if (brw->gen >= 6) {
 632          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 633                    GEN6_VE0_VALID |
 634                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 635                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 636       } else {
 637          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 638                    BRW_VE0_VALID |
 639                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 640                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 641       }
 642       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 643                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 644                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 645                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 646       CACHED_BATCH();
 647       return;
 648    }
 649
 650    /* Now emit VB and VEP state packets.
 651     */
 652
 653    if (brw->vb.nr_buffers) {
 654       if (brw->gen >= 6) {
 655          assert(brw->vb.nr_buffers <= 33);
 656       } else {
 657          assert(brw->vb.nr_buffers <= 17);
 658       }
 659
 660       BEGIN_BATCH(1 + 4*brw->vb.nr_buffers);
 661       OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (4*brw->vb.nr_buffers - 1));
 662       for (i = 0; i < brw->vb.nr_buffers; i++) {
 663          struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 664          uint32_t dw0;
 665
 666          if (brw->gen >= 6) {
 667             dw0 = buffer->step_rate
 668                      ? GEN6_VB0_ACCESS_INSTANCEDATA
 669                      : GEN6_VB0_ACCESS_VERTEXDATA;
 670             dw0 |= i << GEN6_VB0_INDEX_SHIFT;
 671          } else {
 672             dw0 = buffer->step_rate
 673                      ? BRW_VB0_ACCESS_INSTANCEDATA
 674                      : BRW_VB0_ACCESS_VERTEXDATA;
 675             dw0 |= i << BRW_VB0_INDEX_SHIFT;
 676          }
 677
 678          if (brw->gen >= 7)
 679             dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
 680
 681          if (brw->gen == 7)
 682             dw0 |= GEN7_MOCS_L3 << 16;
 683
 684          WARN_ONCE(buffer->stride >= (brw->gen >= 5 ? 2048 : 2047),
 685                    "VBO stride %d too large, bad rendering may occur\n",
 686                    buffer->stride);
 687          OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT));
 688          OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset);
 689          if (brw->gen >= 5) {
 690             OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1);
 691          } else
 692             OUT_BATCH(0);
 693          OUT_BATCH(buffer->step_rate);
 694       }
 695       ADVANCE_BATCH();
 696    }
 697
 698    /* The hardware allows one more VERTEX_ELEMENTS than VERTEX_BUFFERS, presumably
 699     * for VertexID/InstanceID.
 700     */
 701    if (brw->gen >= 6) {
 702       assert(nr_elements <= 34);
 703    } else {
 704       assert(nr_elements <= 18);
 705    }
 706
 707    struct brw_vertex_element *gen6_edgeflag_input = NULL;
 708
 709    BEGIN_BATCH(1 + nr_elements * 2);
 710    OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (2 * nr_elements - 1));
 711    for (i = 0; i < brw->vb.nr_enabled; i++) {
 712       struct brw_vertex_element *input = brw->vb.enabled[i];
 713       uint32_t format = brw_get_vertex_surface_type(brw, input->glarray);
 714       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 715       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 716       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 717       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 718
 719       /* The gen4 driver expects edgeflag to come in as a float, and passes
 720        * that float on to the tests in the clipper.  Mesa's current vertex
 721        * attribute value for EdgeFlag is stored as a float, which works out.
 722        * glEdgeFlagPointer, on the other hand, gives us an unnormalized
 723        * integer ubyte.  Just rewrite that to convert to a float.
 724        */
 725       if (input->attrib == VERT_ATTRIB_EDGEFLAG) {
 726          /* Gen6+ passes edgeflag as sideband along with the vertex, instead
 727           * of in the VUE.  We have to upload it sideband as the last vertex
 728           * element according to the B-Spec.
 729           */
 730          if (brw->gen >= 6) {
 731             gen6_edgeflag_input = input;
 732             continue;
 733          }
 734
 735          if (format == BRW_SURFACEFORMAT_R8_UINT)
 736             format = BRW_SURFACEFORMAT_R8_SSCALED;
 737       }
 738
 739       switch (input->glarray->Size) {
 740       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 741       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 742       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 743       case 3: comp3 = input->glarray->Integer ? BRW_VE1_COMPONENT_STORE_1_INT
 744                                               : BRW_VE1_COMPONENT_STORE_1_FLT;
 745          break;
 746       }
 747
 748       if (brw->gen >= 6) {
 749          OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
 750                    GEN6_VE0_VALID |
 751                    (format << BRW_VE0_FORMAT_SHIFT) |
 752                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 753       } else {
 754          OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) |
 755                    BRW_VE0_VALID |
 756                    (format << BRW_VE0_FORMAT_SHIFT) |
 757                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 758       }
 759
 760       if (brw->gen >= 5)
 761           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 762                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 763                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 764                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 765       else
 766           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 767                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 768                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 769                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 770                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 771    }
 772
 773    if (brw->gen >= 6 && gen6_edgeflag_input) {
 774       uint32_t format =
 775          brw_get_vertex_surface_type(brw, gen6_edgeflag_input->glarray);
 776
 777       OUT_BATCH((gen6_edgeflag_input->buffer << GEN6_VE0_INDEX_SHIFT) |
 778                 GEN6_VE0_VALID |
 779                 GEN6_VE0_EDGE_FLAG_ENABLE |
 780                 (format << BRW_VE0_FORMAT_SHIFT) |
 781                 (gen6_edgeflag_input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 782       OUT_BATCH((BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_0_SHIFT) |
 783                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 784                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 785                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_3_SHIFT));
 786    }
 787
 788    if (brw->vs.prog_data->uses_vertexid) {
 789       uint32_t dw0 = 0, dw1 = 0;
 790
 791       dw1 = ((BRW_VE1_COMPONENT_STORE_VID << BRW_VE1_COMPONENT_0_SHIFT) |
 792              (BRW_VE1_COMPONENT_STORE_IID << BRW_VE1_COMPONENT_1_SHIFT) |
 793              (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 794              (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_3_SHIFT));
 795
 796       if (brw->gen >= 6) {
 797          dw0 |= GEN6_VE0_VALID;
 798       } else {
 799          dw0 |= BRW_VE0_VALID;
 800          dw1 |= (i * 4) << BRW_VE1_DST_OFFSET_SHIFT;
 801       }
 802
 803       /* Note that for gl_VertexID, gl_InstanceID, and gl_PrimitiveID values,
 804        * the format is ignored and the value is always int.
 805        */
 806
 807       OUT_BATCH(dw0);
 808       OUT_BATCH(dw1);
 809    }
 810
 811    CACHED_BATCH();
 812 }
 813
 814 const struct brw_tracked_state brw_vertices = {
 815    .dirty = {
 816       .mesa = _NEW_POLYGON,
 817       .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
 818       .cache = CACHE_NEW_VS_PROG,
 819    },
 820    .emit = brw_emit_vertices,
 821 };
 822
 823 static void brw_upload_indices(struct brw_context *brw)
 824 {
 825    struct gl_context *ctx = &brw->ctx;
 826    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 827    GLuint ib_size;
 828    drm_intel_bo *bo = NULL;
 829    struct gl_buffer_object *bufferobj;
 830    GLuint offset;
 831    GLuint ib_type_size;
 832
 833    if (index_buffer == NULL)
 834       return;
 835
 836    ib_type_size = _mesa_sizeof_type(index_buffer->type);
 837    ib_size = ib_type_size * index_buffer->count;
 838    bufferobj = index_buffer->obj;
 839
 840    /* Turn into a proper VBO:
 841     */
 842    if (!_mesa_is_bufferobj(bufferobj)) {
 843
 844       /* Get new bufferobj, offset:
 845        */
 846       intel_upload_data(brw, index_buffer->ptr, ib_size, ib_type_size,
 847                         &bo, &offset);
 848       brw->ib.start_vertex_offset = offset / ib_type_size;
 849    } else {
 850       offset = (GLuint) (unsigned long) index_buffer->ptr;
 851
 852       /* If the index buffer isn't aligned to its element size, we have to
 853        * rebase it into a temporary.
 854        */
 855        if ((ib_type_size - 1) & offset) {
 856           perf_debug("copying index buffer to a temporary to work around "
 857                      "misaligned offset %d\n", offset);
 858
 859           GLubyte *map = ctx->Driver.MapBufferRange(ctx,
 860                                                     offset,
 861                                                     ib_size,
 862                                                     GL_MAP_READ_BIT,
 863                                                     bufferobj);
 864
 865           intel_upload_data(brw, map, ib_size, ib_type_size, &bo, &offset);
 866           brw->ib.start_vertex_offset = offset / ib_type_size;
 867
 868           ctx->Driver.UnmapBuffer(ctx, bufferobj);
 869        } else {
 870           /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
 871            * the index buffer state when we're just moving the start index
 872            * of our drawing.
 873            */
 874           brw->ib.start_vertex_offset = offset / ib_type_size;
 875
 876           bo = intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 877                                       offset, ib_size);
 878           drm_intel_bo_reference(bo);
 879        }
 880    }
 881
 882    if (brw->ib.bo != bo) {
 883       drm_intel_bo_unreference(brw->ib.bo);
 884       brw->ib.bo = bo;
 885
 886       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 887    } else {
 888       drm_intel_bo_unreference(bo);
 889    }
 890
 891    if (index_buffer->type != brw->ib.type) {
 892       brw->ib.type = index_buffer->type;
 893       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 894    }
 895 }
 896
 897 const struct brw_tracked_state brw_indices = {
 898    .dirty = {
 899       .mesa = 0,
 900       .brw = BRW_NEW_INDICES,
 901       .cache = 0,
 902    },
 903    .emit = brw_upload_indices,
 904 };
 905
 906 static void brw_emit_index_buffer(struct brw_context *brw)
 907 {
 908    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 909    GLuint cut_index_setting;
 910
 911    if (index_buffer == NULL)
 912       return;
 913
 914    if (brw->prim_restart.enable_cut_index && !brw->is_haswell) {
 915       cut_index_setting = BRW_CUT_INDEX_ENABLE;
 916    } else {
 917       cut_index_setting = 0;
 918    }
 919
 920    BEGIN_BATCH(3);
 921    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 922              cut_index_setting |
 923              brw_get_index_type(index_buffer->type) << 8 |
 924              1);
 925    OUT_RELOC(brw->ib.bo,
 926              I915_GEM_DOMAIN_VERTEX, 0,
 927              0);
 928    OUT_RELOC(brw->ib.bo,
 929              I915_GEM_DOMAIN_VERTEX, 0,
 930              brw->ib.bo->size - 1);
 931    ADVANCE_BATCH();
 932 }
 933
 934 const struct brw_tracked_state brw_index_buffer = {
 935    .dirty = {
 936       .mesa = 0,
 937       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 938       .cache = 0,
 939    },
 940    .emit = brw_emit_index_buffer,
 941 };