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[5] = {
  41    0,
  42    BRW_SURFACEFORMAT_R64_FLOAT,
  43    BRW_SURFACEFORMAT_R64G64_FLOAT,
  44    BRW_SURFACEFORMAT_R64G64B64_FLOAT,
  45    BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
  46 };
  47
  48 static const GLuint float_types[5] = {
  49    0,
  50    BRW_SURFACEFORMAT_R32_FLOAT,
  51    BRW_SURFACEFORMAT_R32G32_FLOAT,
  52    BRW_SURFACEFORMAT_R32G32B32_FLOAT,
  53    BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
  54 };
  55
  56 static const GLuint half_float_types[5] = {
  57    0,
  58    BRW_SURFACEFORMAT_R16_FLOAT,
  59    BRW_SURFACEFORMAT_R16G16_FLOAT,
  60    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT,
  61    BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
  62 };
  63
  64 static const GLuint fixed_point_types[5] = {
  65    0,
  66    BRW_SURFACEFORMAT_R32_SFIXED,
  67    BRW_SURFACEFORMAT_R32G32_SFIXED,
  68    BRW_SURFACEFORMAT_R32G32B32_SFIXED,
  69    BRW_SURFACEFORMAT_R32G32B32A32_SFIXED,
  70 };
  71
  72 static const GLuint uint_types_direct[5] = {
  73    0,
  74    BRW_SURFACEFORMAT_R32_UINT,
  75    BRW_SURFACEFORMAT_R32G32_UINT,
  76    BRW_SURFACEFORMAT_R32G32B32_UINT,
  77    BRW_SURFACEFORMAT_R32G32B32A32_UINT
  78 };
  79
  80 static const GLuint uint_types_norm[5] = {
  81    0,
  82    BRW_SURFACEFORMAT_R32_UNORM,
  83    BRW_SURFACEFORMAT_R32G32_UNORM,
  84    BRW_SURFACEFORMAT_R32G32B32_UNORM,
  85    BRW_SURFACEFORMAT_R32G32B32A32_UNORM
  86 };
  87
  88 static const GLuint uint_types_scale[5] = {
  89    0,
  90    BRW_SURFACEFORMAT_R32_USCALED,
  91    BRW_SURFACEFORMAT_R32G32_USCALED,
  92    BRW_SURFACEFORMAT_R32G32B32_USCALED,
  93    BRW_SURFACEFORMAT_R32G32B32A32_USCALED
  94 };
  95
  96 static const GLuint int_types_direct[5] = {
  97    0,
  98    BRW_SURFACEFORMAT_R32_SINT,
  99    BRW_SURFACEFORMAT_R32G32_SINT,
 100    BRW_SURFACEFORMAT_R32G32B32_SINT,
 101    BRW_SURFACEFORMAT_R32G32B32A32_SINT
 102 };
 103
 104 static const GLuint int_types_norm[5] = {
 105    0,
 106    BRW_SURFACEFORMAT_R32_SNORM,
 107    BRW_SURFACEFORMAT_R32G32_SNORM,
 108    BRW_SURFACEFORMAT_R32G32B32_SNORM,
 109    BRW_SURFACEFORMAT_R32G32B32A32_SNORM
 110 };
 111
 112 static const GLuint int_types_scale[5] = {
 113    0,
 114    BRW_SURFACEFORMAT_R32_SSCALED,
 115    BRW_SURFACEFORMAT_R32G32_SSCALED,
 116    BRW_SURFACEFORMAT_R32G32B32_SSCALED,
 117    BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
 118 };
 119
 120 static const GLuint ushort_types_direct[5] = {
 121    0,
 122    BRW_SURFACEFORMAT_R16_UINT,
 123    BRW_SURFACEFORMAT_R16G16_UINT,
 124    BRW_SURFACEFORMAT_R16G16B16A16_UINT,
 125    BRW_SURFACEFORMAT_R16G16B16A16_UINT
 126 };
 127
 128 static const GLuint ushort_types_norm[5] = {
 129    0,
 130    BRW_SURFACEFORMAT_R16_UNORM,
 131    BRW_SURFACEFORMAT_R16G16_UNORM,
 132    BRW_SURFACEFORMAT_R16G16B16_UNORM,
 133    BRW_SURFACEFORMAT_R16G16B16A16_UNORM
 134 };
 135
 136 static const GLuint ushort_types_scale[5] = {
 137    0,
 138    BRW_SURFACEFORMAT_R16_USCALED,
 139    BRW_SURFACEFORMAT_R16G16_USCALED,
 140    BRW_SURFACEFORMAT_R16G16B16_USCALED,
 141    BRW_SURFACEFORMAT_R16G16B16A16_USCALED
 142 };
 143
 144 static const GLuint short_types_direct[5] = {
 145    0,
 146    BRW_SURFACEFORMAT_R16_SINT,
 147    BRW_SURFACEFORMAT_R16G16_SINT,
 148    BRW_SURFACEFORMAT_R16G16B16A16_SINT,
 149    BRW_SURFACEFORMAT_R16G16B16A16_SINT
 150 };
 151
 152 static const GLuint short_types_norm[5] = {
 153    0,
 154    BRW_SURFACEFORMAT_R16_SNORM,
 155    BRW_SURFACEFORMAT_R16G16_SNORM,
 156    BRW_SURFACEFORMAT_R16G16B16_SNORM,
 157    BRW_SURFACEFORMAT_R16G16B16A16_SNORM
 158 };
 159
 160 static const GLuint short_types_scale[5] = {
 161    0,
 162    BRW_SURFACEFORMAT_R16_SSCALED,
 163    BRW_SURFACEFORMAT_R16G16_SSCALED,
 164    BRW_SURFACEFORMAT_R16G16B16_SSCALED,
 165    BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
 166 };
 167
 168 static const GLuint ubyte_types_direct[5] = {
 169    0,
 170    BRW_SURFACEFORMAT_R8_UINT,
 171    BRW_SURFACEFORMAT_R8G8_UINT,
 172    BRW_SURFACEFORMAT_R8G8B8A8_UINT,
 173    BRW_SURFACEFORMAT_R8G8B8A8_UINT
 174 };
 175
 176 static const GLuint ubyte_types_norm[5] = {
 177    0,
 178    BRW_SURFACEFORMAT_R8_UNORM,
 179    BRW_SURFACEFORMAT_R8G8_UNORM,
 180    BRW_SURFACEFORMAT_R8G8B8_UNORM,
 181    BRW_SURFACEFORMAT_R8G8B8A8_UNORM
 182 };
 183
 184 static const GLuint ubyte_types_scale[5] = {
 185    0,
 186    BRW_SURFACEFORMAT_R8_USCALED,
 187    BRW_SURFACEFORMAT_R8G8_USCALED,
 188    BRW_SURFACEFORMAT_R8G8B8_USCALED,
 189    BRW_SURFACEFORMAT_R8G8B8A8_USCALED
 190 };
 191
 192 static const GLuint byte_types_direct[5] = {
 193    0,
 194    BRW_SURFACEFORMAT_R8_SINT,
 195    BRW_SURFACEFORMAT_R8G8_SINT,
 196    BRW_SURFACEFORMAT_R8G8B8A8_SINT,
 197    BRW_SURFACEFORMAT_R8G8B8A8_SINT
 198 };
 199
 200 static const GLuint byte_types_norm[5] = {
 201    0,
 202    BRW_SURFACEFORMAT_R8_SNORM,
 203    BRW_SURFACEFORMAT_R8G8_SNORM,
 204    BRW_SURFACEFORMAT_R8G8B8_SNORM,
 205    BRW_SURFACEFORMAT_R8G8B8A8_SNORM
 206 };
 207
 208 static const GLuint byte_types_scale[5] = {
 209    0,
 210    BRW_SURFACEFORMAT_R8_SSCALED,
 211    BRW_SURFACEFORMAT_R8G8_SSCALED,
 212    BRW_SURFACEFORMAT_R8G8B8_SSCALED,
 213    BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
 214 };
 215
 216
 217 /**
 218  * Given vertex array type/size/format/normalized info, return
 219  * the appopriate hardware surface type.
 220  * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
 221  */
 222 unsigned
 223 brw_get_vertex_surface_type(struct brw_context *brw,
 224                             const struct gl_client_array *glarray)
 225 {
 226    int size = glarray->Size;
 227
 228    if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
 229       fprintf(stderr, "type %s size %d normalized %d\n",
 230               _mesa_enum_to_string(glarray->Type),
 231               glarray->Size, glarray->Normalized);
 232
 233    if (glarray->Integer) {
 234       assert(glarray->Format == GL_RGBA); /* sanity check */
 235       switch (glarray->Type) {
 236       case GL_INT: return int_types_direct[size];
 237       case GL_SHORT: return short_types_direct[size];
 238       case GL_BYTE: return byte_types_direct[size];
 239       case GL_UNSIGNED_INT: return uint_types_direct[size];
 240       case GL_UNSIGNED_SHORT: return ushort_types_direct[size];
 241       case GL_UNSIGNED_BYTE: return ubyte_types_direct[size];
 242       default: unreachable("not reached");
 243       }
 244    } else if (glarray->Type == GL_UNSIGNED_INT_10F_11F_11F_REV) {
 245       return BRW_SURFACEFORMAT_R11G11B10_FLOAT;
 246    } else if (glarray->Normalized) {
 247       switch (glarray->Type) {
 248       case GL_DOUBLE: return double_types[size];
 249       case GL_FLOAT: return float_types[size];
 250       case GL_HALF_FLOAT: return half_float_types[size];
 251       case GL_INT: return int_types_norm[size];
 252       case GL_SHORT: return short_types_norm[size];
 253       case GL_BYTE: return byte_types_norm[size];
 254       case GL_UNSIGNED_INT: return uint_types_norm[size];
 255       case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
 256       case GL_UNSIGNED_BYTE:
 257          if (glarray->Format == GL_BGRA) {
 258             /* See GL_EXT_vertex_array_bgra */
 259             assert(size == 4);
 260             return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
 261          }
 262          else {
 263             return ubyte_types_norm[size];
 264          }
 265       case GL_FIXED:
 266          if (brw->gen >= 8 || brw->is_haswell)
 267             return fixed_point_types[size];
 268
 269          /* This produces GL_FIXED inputs as values between INT32_MIN and
 270           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 271           */
 272          return int_types_scale[size];
 273       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 274        * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
 275        * like to use here, so upload everything as UINT and fix
 276        * it in the shader
 277        */
 278       case GL_INT_2_10_10_10_REV:
 279          assert(size == 4);
 280          if (brw->gen >= 8 || brw->is_haswell) {
 281             return glarray->Format == GL_BGRA
 282                ? BRW_SURFACEFORMAT_B10G10R10A2_SNORM
 283                : BRW_SURFACEFORMAT_R10G10B10A2_SNORM;
 284          }
 285          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 286       case GL_UNSIGNED_INT_2_10_10_10_REV:
 287          assert(size == 4);
 288          if (brw->gen >= 8 || brw->is_haswell) {
 289             return glarray->Format == GL_BGRA
 290                ? BRW_SURFACEFORMAT_B10G10R10A2_UNORM
 291                : BRW_SURFACEFORMAT_R10G10B10A2_UNORM;
 292          }
 293          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 294       default: unreachable("not reached");
 295       }
 296    }
 297    else {
 298       /* See GL_ARB_vertex_type_2_10_10_10_rev.
 299        * W/A: the hardware doesn't really support the formats we'd
 300        * like to use here, so upload everything as UINT and fix
 301        * it in the shader
 302        */
 303       if (glarray->Type == GL_INT_2_10_10_10_REV) {
 304          assert(size == 4);
 305          if (brw->gen >= 8 || brw->is_haswell) {
 306             return glarray->Format == GL_BGRA
 307                ? BRW_SURFACEFORMAT_B10G10R10A2_SSCALED
 308                : BRW_SURFACEFORMAT_R10G10B10A2_SSCALED;
 309          }
 310          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 311       } else if (glarray->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
 312          assert(size == 4);
 313          if (brw->gen >= 8 || brw->is_haswell) {
 314             return glarray->Format == GL_BGRA
 315                ? BRW_SURFACEFORMAT_B10G10R10A2_USCALED
 316                : BRW_SURFACEFORMAT_R10G10B10A2_USCALED;
 317          }
 318          return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
 319       }
 320       assert(glarray->Format == GL_RGBA); /* sanity check */
 321       switch (glarray->Type) {
 322       case GL_DOUBLE: return double_types[size];
 323       case GL_FLOAT: return float_types[size];
 324       case GL_HALF_FLOAT: return half_float_types[size];
 325       case GL_INT: return int_types_scale[size];
 326       case GL_SHORT: return short_types_scale[size];
 327       case GL_BYTE: return byte_types_scale[size];
 328       case GL_UNSIGNED_INT: return uint_types_scale[size];
 329       case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
 330       case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
 331       case GL_FIXED:
 332          if (brw->gen >= 8 || brw->is_haswell)
 333             return fixed_point_types[size];
 334
 335          /* This produces GL_FIXED inputs as values between INT32_MIN and
 336           * INT32_MAX, which will be scaled down by 1/65536 by the VS.
 337           */
 338          return int_types_scale[size];
 339       default: unreachable("not reached");
 340       }
 341    }
 342 }
 343
 344 static void
 345 copy_array_to_vbo_array(struct brw_context *brw,
 346                         struct brw_vertex_element *element,
 347                         int min, int max,
 348                         struct brw_vertex_buffer *buffer,
 349                         GLuint dst_stride)
 350 {
 351    const int src_stride = element->glarray->StrideB;
 352
 353    /* If the source stride is zero, we just want to upload the current
 354     * attribute once and set the buffer's stride to 0.  There's no need
 355     * to replicate it out.
 356     */
 357    if (src_stride == 0) {
 358       intel_upload_data(brw, element->glarray->Ptr,
 359                         element->glarray->_ElementSize,
 360                         element->glarray->_ElementSize,
 361                         &buffer->bo, &buffer->offset);
 362
 363       buffer->stride = 0;
 364       return;
 365    }
 366
 367    const unsigned char *src = element->glarray->Ptr + min * src_stride;
 368    int count = max - min + 1;
 369    GLuint size = count * dst_stride;
 370    uint8_t *dst = intel_upload_space(brw, size, dst_stride,
 371                                      &buffer->bo, &buffer->offset);
 372
 373    if (dst_stride == src_stride) {
 374       memcpy(dst, src, size);
 375    } else {
 376       while (count--) {
 377          memcpy(dst, src, dst_stride);
 378          src += src_stride;
 379          dst += dst_stride;
 380       }
 381    }
 382    buffer->stride = dst_stride;
 383 }
 384
 385 void
 386 brw_prepare_vertices(struct brw_context *brw)
 387 {
 388    struct gl_context *ctx = &brw->ctx;
 389    /* BRW_NEW_VS_PROG_DATA */
 390    GLbitfield64 vs_inputs = brw->vs.prog_data->inputs_read;
 391    const unsigned char *ptr = NULL;
 392    GLuint interleaved = 0;
 393    unsigned int min_index = brw->vb.min_index + brw->basevertex;
 394    unsigned int max_index = brw->vb.max_index + brw->basevertex;
 395    unsigned i;
 396    int delta, j;
 397
 398    struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
 399    GLuint nr_uploads = 0;
 400
 401    /* _NEW_POLYGON
 402     *
 403     * On gen6+, edge flags don't end up in the VUE (either in or out of the
 404     * VS).  Instead, they're uploaded as the last vertex element, and the data
 405     * is passed sideband through the fixed function units.  So, we need to
 406     * prepare the vertex buffer for it, but it's not present in inputs_read.
 407     */
 408    if (brw->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
 409                            ctx->Polygon.BackMode != GL_FILL)) {
 410       vs_inputs |= VERT_BIT_EDGEFLAG;
 411    }
 412
 413    if (0)
 414       fprintf(stderr, "%s %d..%d\n", __func__, min_index, max_index);
 415
 416    /* Accumulate the list of enabled arrays. */
 417    brw->vb.nr_enabled = 0;
 418    while (vs_inputs) {
 419       GLuint index = ffsll(vs_inputs) - 1;
 420       struct brw_vertex_element *input = &brw->vb.inputs[index];
 421
 422       vs_inputs &= ~BITFIELD64_BIT(index);
 423       brw->vb.enabled[brw->vb.nr_enabled++] = input;
 424    }
 425
 426    if (brw->vb.nr_enabled == 0)
 427       return;
 428
 429    if (brw->vb.nr_buffers)
 430       return;
 431
 432    for (i = j = 0; i < brw->vb.nr_enabled; i++) {
 433       struct brw_vertex_element *input = brw->vb.enabled[i];
 434       const struct gl_client_array *glarray = input->glarray;
 435
 436       if (_mesa_is_bufferobj(glarray->BufferObj)) {
 437          struct intel_buffer_object *intel_buffer =
 438             intel_buffer_object(glarray->BufferObj);
 439          unsigned k;
 440
 441          /* If we have a VB set to be uploaded for this buffer object
 442           * already, reuse that VB state so that we emit fewer
 443           * relocations.
 444           */
 445          for (k = 0; k < i; k++) {
 446             const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
 447             if (glarray->BufferObj == other->BufferObj &&
 448                 glarray->StrideB == other->StrideB &&
 449                 glarray->InstanceDivisor == other->InstanceDivisor &&
 450                 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
 451             {
 452                input->buffer = brw->vb.enabled[k]->buffer;
 453                input->offset = glarray->Ptr - other->Ptr;
 454                break;
 455             }
 456          }
 457          if (k == i) {
 458             struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 459
 460             /* Named buffer object: Just reference its contents directly. */
 461             buffer->offset = (uintptr_t)glarray->Ptr;
 462             buffer->stride = glarray->StrideB;
 463             buffer->step_rate = glarray->InstanceDivisor;
 464
 465             uint32_t offset, size;
 466             if (glarray->InstanceDivisor) {
 467                offset = buffer->offset;
 468                size = (buffer->stride * ((brw->num_instances /
 469                                           glarray->InstanceDivisor) - 1) +
 470                        glarray->_ElementSize);
 471             } else {
 472                if (min_index == -1) {
 473                   offset = 0;
 474                   size = intel_buffer->Base.Size;
 475                } else {
 476                   offset = buffer->offset + min_index * buffer->stride;
 477                   size = (buffer->stride * (max_index - min_index) +
 478                           glarray->_ElementSize);
 479                }
 480             }
 481             buffer->bo = intel_bufferobj_buffer(brw, intel_buffer,
 482                                                 offset, size);
 483             drm_intel_bo_reference(buffer->bo);
 484
 485             input->buffer = j++;
 486             input->offset = 0;
 487          }
 488
 489          /* This is a common place to reach if the user mistakenly supplies
 490           * a pointer in place of a VBO offset.  If we just let it go through,
 491           * we may end up dereferencing a pointer beyond the bounds of the
 492           * GTT.
 493           *
 494           * The VBO spec allows application termination in this case, and it's
 495           * probably a service to the poor programmer to do so rather than
 496           * trying to just not render.
 497           */
 498          assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
 499       } else {
 500          /* Queue the buffer object up to be uploaded in the next pass,
 501           * when we've decided if we're doing interleaved or not.
 502           */
 503          if (nr_uploads == 0) {
 504             interleaved = glarray->StrideB;
 505             ptr = glarray->Ptr;
 506          }
 507          else if (interleaved != glarray->StrideB ||
 508                   glarray->Ptr < ptr ||
 509                   (uintptr_t)(glarray->Ptr - ptr) + glarray->_ElementSize > interleaved)
 510          {
 511             /* If our stride is different from the first attribute's stride,
 512              * or if the first attribute's stride didn't cover our element,
 513              * disable the interleaved upload optimization.  The second case
 514              * can most commonly occur in cases where there is a single vertex
 515              * and, for example, the data is stored on the application's
 516              * stack.
 517              *
 518              * NOTE: This will also disable the optimization in cases where
 519              * the data is in a different order than the array indices.
 520              * Something like:
 521              *
 522              *     float data[...];
 523              *     glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
 524              *     glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
 525              */
 526             interleaved = 0;
 527          }
 528
 529          upload[nr_uploads++] = input;
 530       }
 531    }
 532
 533    /* If we need to upload all the arrays, then we can trim those arrays to
 534     * only the used elements [min_index, max_index] so long as we adjust all
 535     * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
 536     */
 537    brw->vb.start_vertex_bias = 0;
 538    delta = min_index;
 539    if (nr_uploads == brw->vb.nr_enabled) {
 540       brw->vb.start_vertex_bias = -delta;
 541       delta = 0;
 542    }
 543
 544    /* Handle any arrays to be uploaded. */
 545    if (nr_uploads > 1) {
 546       if (interleaved) {
 547          struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 548          /* All uploads are interleaved, so upload the arrays together as
 549           * interleaved.  First, upload the contents and set up upload[0].
 550           */
 551          copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
 552                                  buffer, interleaved);
 553          buffer->offset -= delta * interleaved;
 554
 555          for (i = 0; i < nr_uploads; i++) {
 556             /* Then, just point upload[i] at upload[0]'s buffer. */
 557             upload[i]->offset =
 558                ((const unsigned char *)upload[i]->glarray->Ptr - ptr);
 559             upload[i]->buffer = j;
 560          }
 561          j++;
 562
 563          nr_uploads = 0;
 564       }
 565    }
 566    /* Upload non-interleaved arrays */
 567    for (i = 0; i < nr_uploads; i++) {
 568       struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
 569       if (upload[i]->glarray->InstanceDivisor == 0) {
 570          copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
 571                                  buffer, upload[i]->glarray->_ElementSize);
 572       } else {
 573          /* This is an instanced attribute, since its InstanceDivisor
 574           * is not zero. Therefore, its data will be stepped after the
 575           * instanced draw has been run InstanceDivisor times.
 576           */
 577          uint32_t instanced_attr_max_index =
 578             (brw->num_instances - 1) / upload[i]->glarray->InstanceDivisor;
 579          copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
 580                                  buffer, upload[i]->glarray->_ElementSize);
 581       }
 582       buffer->offset -= delta * buffer->stride;
 583       buffer->step_rate = upload[i]->glarray->InstanceDivisor;
 584       upload[i]->buffer = j++;
 585       upload[i]->offset = 0;
 586    }
 587
 588    brw->vb.nr_buffers = j;
 589 }
 590
 591 void
 592 brw_prepare_shader_draw_parameters(struct brw_context *brw)
 593 {
 594    /* For non-indirect draws, upload gl_BaseVertex. */
 595    if ((brw->vs.prog_data->uses_basevertex ||
 596         brw->vs.prog_data->uses_baseinstance) &&
 597        brw->draw.draw_params_bo == NULL) {
 598       intel_upload_data(brw, &brw->draw.params, sizeof(brw->draw.params), 4,
 599                         &brw->draw.draw_params_bo,
 600                         &brw->draw.draw_params_offset);
 601    }
 602 }
 603
 604 /**
 605  * Emit a VERTEX_BUFFER_STATE entry (part of 3DSTATE_VERTEX_BUFFERS).
 606  */
 607 static uint32_t *
 608 emit_vertex_buffer_state(struct brw_context *brw,
 609                          unsigned buffer_nr,
 610                          drm_intel_bo *bo,
 611                          unsigned bo_ending_address,
 612                          unsigned bo_offset,
 613                          unsigned stride,
 614                          unsigned step_rate,
 615                          uint32_t *__map)
 616 {
 617    struct gl_context *ctx = &brw->ctx;
 618    uint32_t dw0;
 619
 620    if (brw->gen >= 6) {
 621       dw0 = (buffer_nr << GEN6_VB0_INDEX_SHIFT) |
 622             (step_rate ? GEN6_VB0_ACCESS_INSTANCEDATA
 623                        : GEN6_VB0_ACCESS_VERTEXDATA);
 624    } else {
 625       dw0 = (buffer_nr << BRW_VB0_INDEX_SHIFT) |
 626             (step_rate ? BRW_VB0_ACCESS_INSTANCEDATA
 627                        : BRW_VB0_ACCESS_VERTEXDATA);
 628    }
 629
 630    if (brw->gen >= 7)
 631       dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
 632
 633    if (brw->gen == 7)
 634       dw0 |= GEN7_MOCS_L3 << 16;
 635
 636    WARN_ONCE(stride >= (brw->gen >= 5 ? 2048 : 2047),
 637              "VBO stride %d too large, bad rendering may occur\n",
 638              stride);
 639    OUT_BATCH(dw0 | (stride << BRW_VB0_PITCH_SHIFT));
 640    OUT_RELOC(bo, I915_GEM_DOMAIN_VERTEX, 0, bo_offset);
 641    if (brw->gen >= 5) {
 642       OUT_RELOC(bo, I915_GEM_DOMAIN_VERTEX, 0, bo_ending_address);
 643    } else {
 644       OUT_BATCH(0);
 645    }
 646    OUT_BATCH(step_rate);
 647
 648    return __map;
 649 }
 650 #define EMIT_VERTEX_BUFFER_STATE(...) __map = emit_vertex_buffer_state(__VA_ARGS__, __map)
 651
 652 static void
 653 brw_emit_vertices(struct brw_context *brw)
 654 {
 655    GLuint i;
 656
 657    brw_prepare_vertices(brw);
 658    brw_prepare_shader_draw_parameters(brw);
 659
 660    brw_emit_query_begin(brw);
 661
 662    unsigned nr_elements = brw->vb.nr_enabled;
 663    if (brw->vs.prog_data->uses_vertexid || brw->vs.prog_data->uses_instanceid ||
 664        brw->vs.prog_data->uses_basevertex || brw->vs.prog_data->uses_baseinstance)
 665       ++nr_elements;
 666
 667    /* If the VS doesn't read any inputs (calculating vertex position from
 668     * a state variable for some reason, for example), emit a single pad
 669     * VERTEX_ELEMENT struct and bail.
 670     *
 671     * The stale VB state stays in place, but they don't do anything unless
 672     * a VE loads from them.
 673     */
 674    if (nr_elements == 0) {
 675       BEGIN_BATCH(3);
 676       OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | 1);
 677       if (brw->gen >= 6) {
 678          OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
 679                    GEN6_VE0_VALID |
 680                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 681                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 682       } else {
 683          OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
 684                    BRW_VE0_VALID |
 685                    (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
 686                    (0 << BRW_VE0_SRC_OFFSET_SHIFT));
 687       }
 688       OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
 689                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 690                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 691                 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
 692       ADVANCE_BATCH();
 693       return;
 694    }
 695
 696    /* Now emit VB and VEP state packets.
 697     */
 698
 699    const bool uses_draw_params =
 700       brw->vs.prog_data->uses_basevertex ||
 701       brw->vs.prog_data->uses_baseinstance;
 702    const unsigned nr_buffers = brw->vb.nr_buffers + uses_draw_params;
 703
 704    if (nr_buffers) {
 705       if (brw->gen >= 6) {
 706          assert(nr_buffers <= 33);
 707       } else {
 708          assert(nr_buffers <= 17);
 709       }
 710
 711       BEGIN_BATCH(1 + 4 * nr_buffers);
 712       OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (4 * nr_buffers - 1));
 713       for (i = 0; i < brw->vb.nr_buffers; i++) {
 714          struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
 715          EMIT_VERTEX_BUFFER_STATE(brw, i, buffer->bo, buffer->bo->size - 1,
 716                                   buffer->offset, buffer->stride,
 717                                   buffer->step_rate);
 718
 719       }
 720
 721       if (uses_draw_params) {
 722          EMIT_VERTEX_BUFFER_STATE(brw, brw->vb.nr_buffers,
 723                                   brw->draw.draw_params_bo,
 724                                   brw->draw.draw_params_bo->size - 1,
 725                                   brw->draw.draw_params_offset,
 726                                   0,  /* stride */
 727                                   0); /* step rate */
 728       }
 729       ADVANCE_BATCH();
 730    }
 731
 732    /* The hardware allows one more VERTEX_ELEMENTS than VERTEX_BUFFERS, presumably
 733     * for VertexID/InstanceID.
 734     */
 735    if (brw->gen >= 6) {
 736       assert(nr_elements <= 34);
 737    } else {
 738       assert(nr_elements <= 18);
 739    }
 740
 741    struct brw_vertex_element *gen6_edgeflag_input = NULL;
 742
 743    BEGIN_BATCH(1 + nr_elements * 2);
 744    OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (2 * nr_elements - 1));
 745    for (i = 0; i < brw->vb.nr_enabled; i++) {
 746       struct brw_vertex_element *input = brw->vb.enabled[i];
 747       uint32_t format = brw_get_vertex_surface_type(brw, input->glarray);
 748       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 749       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 750       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
 751       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
 752
 753       if (input == &brw->vb.inputs[VERT_ATTRIB_EDGEFLAG]) {
 754          /* Gen6+ passes edgeflag as sideband along with the vertex, instead
 755           * of in the VUE.  We have to upload it sideband as the last vertex
 756           * element according to the B-Spec.
 757           */
 758          if (brw->gen >= 6) {
 759             gen6_edgeflag_input = input;
 760             continue;
 761          }
 762       }
 763
 764       switch (input->glarray->Size) {
 765       case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
 766       case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
 767       case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
 768       case 3: comp3 = input->glarray->Integer ? BRW_VE1_COMPONENT_STORE_1_INT
 769                                               : BRW_VE1_COMPONENT_STORE_1_FLT;
 770          break;
 771       }
 772
 773       if (brw->gen >= 6) {
 774          OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
 775                    GEN6_VE0_VALID |
 776                    (format << BRW_VE0_FORMAT_SHIFT) |
 777                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 778       } else {
 779          OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) |
 780                    BRW_VE0_VALID |
 781                    (format << BRW_VE0_FORMAT_SHIFT) |
 782                    (input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 783       }
 784
 785       if (brw->gen >= 5)
 786           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 787                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 788                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 789                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
 790       else
 791           OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 792                     (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 793                     (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 794                     (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
 795                     ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
 796    }
 797
 798    if (brw->vs.prog_data->uses_vertexid || brw->vs.prog_data->uses_instanceid ||
 799        brw->vs.prog_data->uses_basevertex || brw->vs.prog_data->uses_baseinstance) {
 800       uint32_t dw0 = 0, dw1 = 0;
 801       uint32_t comp0 = BRW_VE1_COMPONENT_STORE_0;
 802       uint32_t comp1 = BRW_VE1_COMPONENT_STORE_0;
 803       uint32_t comp2 = BRW_VE1_COMPONENT_STORE_0;
 804       uint32_t comp3 = BRW_VE1_COMPONENT_STORE_0;
 805
 806       if (brw->vs.prog_data->uses_basevertex)
 807          comp0 = BRW_VE1_COMPONENT_STORE_SRC;
 808
 809       if (brw->vs.prog_data->uses_baseinstance)
 810          comp1 = BRW_VE1_COMPONENT_STORE_SRC;
 811
 812       if (brw->vs.prog_data->uses_vertexid)
 813          comp2 = BRW_VE1_COMPONENT_STORE_VID;
 814
 815       if (brw->vs.prog_data->uses_instanceid)
 816          comp3 = BRW_VE1_COMPONENT_STORE_IID;
 817
 818       dw1 = (comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
 819             (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
 820             (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
 821             (comp3 << BRW_VE1_COMPONENT_3_SHIFT);
 822
 823       if (brw->gen >= 6) {
 824          dw0 |= GEN6_VE0_VALID |
 825                 brw->vb.nr_buffers << GEN6_VE0_INDEX_SHIFT |
 826                 BRW_SURFACEFORMAT_R32G32_UINT << BRW_VE0_FORMAT_SHIFT;
 827       } else {
 828          dw0 |= BRW_VE0_VALID |
 829                 brw->vb.nr_buffers << BRW_VE0_INDEX_SHIFT |
 830                 BRW_SURFACEFORMAT_R32G32_UINT << BRW_VE0_FORMAT_SHIFT;
 831          dw1 |= (i * 4) << BRW_VE1_DST_OFFSET_SHIFT;
 832       }
 833
 834       /* Note that for gl_VertexID, gl_InstanceID, and gl_PrimitiveID values,
 835        * the format is ignored and the value is always int.
 836        */
 837
 838       OUT_BATCH(dw0);
 839       OUT_BATCH(dw1);
 840    }
 841
 842    if (brw->gen >= 6 && gen6_edgeflag_input) {
 843       uint32_t format =
 844          brw_get_vertex_surface_type(brw, gen6_edgeflag_input->glarray);
 845
 846       OUT_BATCH((gen6_edgeflag_input->buffer << GEN6_VE0_INDEX_SHIFT) |
 847                 GEN6_VE0_VALID |
 848                 GEN6_VE0_EDGE_FLAG_ENABLE |
 849                 (format << BRW_VE0_FORMAT_SHIFT) |
 850                 (gen6_edgeflag_input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
 851       OUT_BATCH((BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_0_SHIFT) |
 852                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
 853                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
 854                 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_3_SHIFT));
 855    }
 856
 857    ADVANCE_BATCH();
 858 }
 859
 860 const struct brw_tracked_state brw_vertices = {
 861    .dirty = {
 862       .mesa = _NEW_POLYGON,
 863       .brw = BRW_NEW_BATCH |
 864              BRW_NEW_VERTICES |
 865              BRW_NEW_VS_PROG_DATA,
 866    },
 867    .emit = brw_emit_vertices,
 868 };
 869
 870 static void
 871 brw_upload_indices(struct brw_context *brw)
 872 {
 873    struct gl_context *ctx = &brw->ctx;
 874    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 875    GLuint ib_size;
 876    drm_intel_bo *old_bo = brw->ib.bo;
 877    struct gl_buffer_object *bufferobj;
 878    GLuint offset;
 879    GLuint ib_type_size;
 880
 881    if (index_buffer == NULL)
 882       return;
 883
 884    ib_type_size = _mesa_sizeof_type(index_buffer->type);
 885    ib_size = ib_type_size * index_buffer->count;
 886    bufferobj = index_buffer->obj;
 887
 888    /* Turn into a proper VBO:
 889     */
 890    if (!_mesa_is_bufferobj(bufferobj)) {
 891       /* Get new bufferobj, offset:
 892        */
 893       intel_upload_data(brw, index_buffer->ptr, ib_size, ib_type_size,
 894                         &brw->ib.bo, &offset);
 895    } else {
 896       offset = (GLuint) (unsigned long) index_buffer->ptr;
 897
 898       /* If the index buffer isn't aligned to its element size, we have to
 899        * rebase it into a temporary.
 900        */
 901       if ((ib_type_size - 1) & offset) {
 902          perf_debug("copying index buffer to a temporary to work around "
 903                     "misaligned offset %d\n", offset);
 904
 905          GLubyte *map = ctx->Driver.MapBufferRange(ctx,
 906                                                    offset,
 907                                                    ib_size,
 908                                                    GL_MAP_READ_BIT,
 909                                                    bufferobj,
 910                                                    MAP_INTERNAL);
 911
 912          intel_upload_data(brw, map, ib_size, ib_type_size,
 913                            &brw->ib.bo, &offset);
 914
 915          ctx->Driver.UnmapBuffer(ctx, bufferobj, MAP_INTERNAL);
 916       } else {
 917          drm_intel_bo *bo =
 918             intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
 919                                    offset, ib_size);
 920          if (bo != brw->ib.bo) {
 921             drm_intel_bo_unreference(brw->ib.bo);
 922             brw->ib.bo = bo;
 923             drm_intel_bo_reference(bo);
 924          }
 925       }
 926    }
 927
 928    /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
 929     * the index buffer state when we're just moving the start index
 930     * of our drawing.
 931     */
 932    brw->ib.start_vertex_offset = offset / ib_type_size;
 933
 934    if (brw->ib.bo != old_bo)
 935       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 936
 937    if (index_buffer->type != brw->ib.type) {
 938       brw->ib.type = index_buffer->type;
 939       brw->ctx.NewDriverState |= BRW_NEW_INDEX_BUFFER;
 940    }
 941 }
 942
 943 const struct brw_tracked_state brw_indices = {
 944    .dirty = {
 945       .mesa = 0,
 946       .brw = BRW_NEW_INDICES,
 947    },
 948    .emit = brw_upload_indices,
 949 };
 950
 951 static void
 952 brw_emit_index_buffer(struct brw_context *brw)
 953 {
 954    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
 955    GLuint cut_index_setting;
 956
 957    if (index_buffer == NULL)
 958       return;
 959
 960    if (brw->prim_restart.enable_cut_index && !brw->is_haswell) {
 961       cut_index_setting = BRW_CUT_INDEX_ENABLE;
 962    } else {
 963       cut_index_setting = 0;
 964    }
 965
 966    BEGIN_BATCH(3);
 967    OUT_BATCH(CMD_INDEX_BUFFER << 16 |
 968              cut_index_setting |
 969              brw_get_index_type(index_buffer->type) |
 970              1);
 971    OUT_RELOC(brw->ib.bo,
 972              I915_GEM_DOMAIN_VERTEX, 0,
 973              0);
 974    OUT_RELOC(brw->ib.bo,
 975              I915_GEM_DOMAIN_VERTEX, 0,
 976              brw->ib.bo->size - 1);
 977    ADVANCE_BATCH();
 978 }
 979
 980 const struct brw_tracked_state brw_index_buffer = {
 981    .dirty = {
 982       .mesa = 0,
 983       .brw = BRW_NEW_BATCH |
 984              BRW_NEW_INDEX_BUFFER,
 985    },
 986    .emit = brw_emit_index_buffer,
 987 };