src/gallium/drivers/svga/svga_draw.c

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #include "pipe/p_compiler.h"
  27 #include "util/u_inlines.h"
  28 #include "pipe/p_defines.h"
  29 #include "util/u_helpers.h"
  30 #include "util/u_memory.h"
  31 #include "util/u_math.h"
  32
  33 #include "svga_context.h"
  34 #include "svga_draw.h"
  35 #include "svga_draw_private.h"
  36 #include "svga_debug.h"
  37 #include "svga_screen.h"
  38 #include "svga_resource.h"
  39 #include "svga_resource_buffer.h"
  40 #include "svga_resource_texture.h"
  41 #include "svga_shader.h"
  42 #include "svga_surface.h"
  43 #include "svga_winsys.h"
  44 #include "svga_cmd.h"
  45
  46
  47 struct svga_hwtnl *
  48 svga_hwtnl_create(struct svga_context *svga)
  49 {
  50    struct svga_hwtnl *hwtnl = CALLOC_STRUCT(svga_hwtnl);
  51    if (!hwtnl)
  52       goto fail;
  53
  54    hwtnl->svga = svga;
  55
  56    hwtnl->cmd.swc = svga->swc;
  57
  58    return hwtnl;
  59
  60 fail:
  61    return NULL;
  62 }
  63
  64
  65 void
  66 svga_hwtnl_destroy(struct svga_hwtnl *hwtnl)
  67 {
  68    unsigned i, j;
  69
  70    for (i = 0; i < PIPE_PRIM_MAX; i++) {
  71       for (j = 0; j < IDX_CACHE_MAX; j++) {
  72          pipe_resource_reference(&hwtnl->index_cache[i][j].buffer, NULL);
  73       }
  74    }
  75
  76    for (i = 0; i < hwtnl->cmd.vbuf_count; i++)
  77       pipe_resource_reference(&hwtnl->cmd.vbufs[i].buffer, NULL);
  78
  79    for (i = 0; i < hwtnl->cmd.prim_count; i++)
  80       pipe_resource_reference(&hwtnl->cmd.prim_ib[i], NULL);
  81
  82    FREE(hwtnl);
  83 }
  84
  85
  86 void
  87 svga_hwtnl_set_flatshade(struct svga_hwtnl *hwtnl,
  88                          boolean flatshade, boolean flatshade_first)
  89 {
  90    struct svga_screen *svgascreen = svga_screen(hwtnl->svga->pipe.screen);
  91
  92    /* User-specified PV */
  93    hwtnl->api_pv = (flatshade && !flatshade_first) ? PV_LAST : PV_FIRST;
  94
  95    /* Device supported PV */
  96    if (svgascreen->haveProvokingVertex) {
  97       /* use the mode specified by the user */
  98       hwtnl->hw_pv = hwtnl->api_pv;
  99    }
 100    else {
 101       /* the device only support first provoking vertex */
 102       hwtnl->hw_pv = PV_FIRST;
 103    }
 104 }
 105
 106
 107 void
 108 svga_hwtnl_set_fillmode(struct svga_hwtnl *hwtnl, unsigned mode)
 109 {
 110    hwtnl->api_fillmode = mode;
 111 }
 112
 113
 114 void
 115 svga_hwtnl_vertex_decls(struct svga_hwtnl *hwtnl,
 116                         unsigned count,
 117                         const SVGA3dVertexDecl * decls,
 118                         const unsigned *buffer_indexes,
 119                         SVGA3dElementLayoutId layout_id)
 120 {
 121    assert(hwtnl->cmd.prim_count == 0);
 122    hwtnl->cmd.vdecl_count = count;
 123    hwtnl->cmd.vdecl_layout_id = layout_id;
 124    memcpy(hwtnl->cmd.vdecl, decls, count * sizeof(*decls));
 125    memcpy(hwtnl->cmd.vdecl_buffer_index, buffer_indexes,
 126           count * sizeof(unsigned));
 127 }
 128
 129
 130 /**
 131  * Specify vertex buffers for hardware drawing.
 132  */
 133 void
 134 svga_hwtnl_vertex_buffers(struct svga_hwtnl *hwtnl,
 135                           unsigned count, struct pipe_vertex_buffer *buffers)
 136 {
 137    util_set_vertex_buffers_count(hwtnl->cmd.vbufs,
 138                                  &hwtnl->cmd.vbuf_count, buffers, 0, count);
 139 }
 140
 141
 142 /**
 143  * Determine whether the specified buffer is referred in the primitive queue,
 144  * for which no commands have been written yet.
 145  */
 146 boolean
 147 svga_hwtnl_is_buffer_referred(struct svga_hwtnl *hwtnl,
 148                               struct pipe_resource *buffer)
 149 {
 150    unsigned i;
 151
 152    if (svga_buffer_is_user_buffer(buffer)) {
 153       return FALSE;
 154    }
 155
 156    if (!hwtnl->cmd.prim_count) {
 157       return FALSE;
 158    }
 159
 160    for (i = 0; i < hwtnl->cmd.vbuf_count; ++i) {
 161       if (hwtnl->cmd.vbufs[i].buffer == buffer) {
 162          return TRUE;
 163       }
 164    }
 165
 166    for (i = 0; i < hwtnl->cmd.prim_count; ++i) {
 167       if (hwtnl->cmd.prim_ib[i] == buffer) {
 168          return TRUE;
 169       }
 170    }
 171
 172    return FALSE;
 173 }
 174
 175
 176 static enum pipe_error
 177 draw_vgpu9(struct svga_hwtnl *hwtnl)
 178 {
 179    struct svga_winsys_context *swc = hwtnl->cmd.swc;
 180    struct svga_context *svga = hwtnl->svga;
 181    enum pipe_error ret;
 182    struct svga_winsys_surface *vb_handle[SVGA3D_INPUTREG_MAX];
 183    struct svga_winsys_surface *ib_handle[QSZ];
 184    struct svga_winsys_surface *handle;
 185    SVGA3dVertexDecl *vdecl;
 186    SVGA3dPrimitiveRange *prim;
 187    unsigned i;
 188
 189    for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
 190       unsigned j = hwtnl->cmd.vdecl_buffer_index[i];
 191       handle = svga_buffer_handle(svga, hwtnl->cmd.vbufs[j].buffer);
 192       if (!handle)
 193          return PIPE_ERROR_OUT_OF_MEMORY;
 194
 195       vb_handle[i] = handle;
 196    }
 197
 198    for (i = 0; i < hwtnl->cmd.prim_count; i++) {
 199       if (hwtnl->cmd.prim_ib[i]) {
 200          handle = svga_buffer_handle(svga, hwtnl->cmd.prim_ib[i]);
 201          if (!handle)
 202             return PIPE_ERROR_OUT_OF_MEMORY;
 203       }
 204       else
 205          handle = NULL;
 206
 207       ib_handle[i] = handle;
 208    }
 209
 210    if (svga->rebind.flags.rendertargets) {
 211       ret = svga_reemit_framebuffer_bindings(svga);
 212       if (ret != PIPE_OK) {
 213          return ret;
 214       }
 215    }
 216
 217    if (svga->rebind.flags.texture_samplers) {
 218       ret = svga_reemit_tss_bindings(svga);
 219       if (ret != PIPE_OK) {
 220          return ret;
 221       }
 222    }
 223
 224    if (svga->rebind.flags.vs) {
 225       ret = svga_reemit_vs_bindings(svga);
 226       if (ret != PIPE_OK) {
 227          return ret;
 228       }
 229    }
 230
 231    if (svga->rebind.flags.fs) {
 232       ret = svga_reemit_fs_bindings(svga);
 233       if (ret != PIPE_OK) {
 234          return ret;
 235       }
 236    }
 237
 238    SVGA_DBG(DEBUG_DMA, "draw to sid %p, %d prims\n",
 239             svga->curr.framebuffer.cbufs[0] ?
 240             svga_surface(svga->curr.framebuffer.cbufs[0])->handle : NULL,
 241             hwtnl->cmd.prim_count);
 242
 243    ret = SVGA3D_BeginDrawPrimitives(swc,
 244                                     &vdecl,
 245                                     hwtnl->cmd.vdecl_count,
 246                                     &prim, hwtnl->cmd.prim_count);
 247    if (ret != PIPE_OK)
 248       return ret;
 249
 250    memcpy(vdecl,
 251           hwtnl->cmd.vdecl,
 252           hwtnl->cmd.vdecl_count * sizeof hwtnl->cmd.vdecl[0]);
 253
 254    for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
 255       /* check for 4-byte alignment */
 256       assert(vdecl[i].array.offset % 4 == 0);
 257       assert(vdecl[i].array.stride % 4 == 0);
 258
 259       /* Given rangeHint is considered to be relative to indexBias, and
 260        * indexBias varies per primitive, we cannot accurately supply an
 261        * rangeHint when emitting more than one primitive per draw command.
 262        */
 263       if (hwtnl->cmd.prim_count == 1) {
 264          vdecl[i].rangeHint.first = hwtnl->cmd.min_index[0];
 265          vdecl[i].rangeHint.last = hwtnl->cmd.max_index[0] + 1;
 266       }
 267       else {
 268          vdecl[i].rangeHint.first = 0;
 269          vdecl[i].rangeHint.last = 0;
 270       }
 271
 272       swc->surface_relocation(swc,
 273                               &vdecl[i].array.surfaceId,
 274                               NULL, vb_handle[i], SVGA_RELOC_READ);
 275    }
 276
 277    memcpy(prim,
 278           hwtnl->cmd.prim, hwtnl->cmd.prim_count * sizeof hwtnl->cmd.prim[0]);
 279
 280    for (i = 0; i < hwtnl->cmd.prim_count; i++) {
 281       swc->surface_relocation(swc,
 282                               &prim[i].indexArray.surfaceId,
 283                               NULL, ib_handle[i], SVGA_RELOC_READ);
 284       pipe_resource_reference(&hwtnl->cmd.prim_ib[i], NULL);
 285    }
 286
 287    SVGA_FIFOCommitAll(swc);
 288
 289    hwtnl->cmd.prim_count = 0;
 290
 291    return PIPE_OK;
 292 }
 293
 294
 295 static SVGA3dSurfaceFormat
 296 xlate_index_format(unsigned indexWidth)
 297 {
 298    if (indexWidth == 2) {
 299       return SVGA3D_R16_UINT;
 300    }
 301    else if (indexWidth == 4) {
 302       return SVGA3D_R32_UINT;
 303    }
 304    else {
 305       assert(!"Bad indexWidth");
 306       return SVGA3D_R32_UINT;
 307    }
 308 }
 309
 310
 311 static enum pipe_error
 312 validate_sampler_resources(struct svga_context *svga)
 313 {
 314    unsigned shader;
 315
 316    assert(svga_have_vgpu10(svga));
 317
 318    for (shader = PIPE_SHADER_VERTEX; shader <= PIPE_SHADER_GEOMETRY; shader++) {
 319       unsigned count = svga->curr.num_sampler_views[shader];
 320       unsigned i;
 321       struct svga_winsys_surface *surfaces[PIPE_MAX_SAMPLERS];
 322       enum pipe_error ret;
 323
 324       /*
 325        * Reference bound sampler resources to ensure pending updates are
 326        * noticed by the device.
 327        */
 328       for (i = 0; i < count; i++) {
 329          struct svga_pipe_sampler_view *sv =
 330             svga_pipe_sampler_view(svga->curr.sampler_views[shader][i]);
 331
 332          if (sv) {
 333             if (sv->base.texture->target == PIPE_BUFFER) {
 334                surfaces[i] = svga_buffer_handle(svga, sv->base.texture);
 335             }
 336             else {
 337                surfaces[i] = svga_texture(sv->base.texture)->handle;
 338             }
 339          }
 340          else {
 341             surfaces[i] = NULL;
 342          }
 343       }
 344
 345       if (shader == PIPE_SHADER_FRAGMENT &&
 346           svga->curr.rast->templ.poly_stipple_enable) {
 347          const unsigned unit = svga->state.hw_draw.fs->pstipple_sampler_unit;
 348          struct svga_pipe_sampler_view *sv =
 349             svga->polygon_stipple.sampler_view;
 350
 351          assert(sv);
 352          surfaces[unit] = svga_texture(sv->base.texture)->handle;
 353          count = MAX2(count, unit+1);
 354       }
 355
 356       /* rebind the shader resources if needed */
 357       if (svga->rebind.flags.texture_samplers) {
 358          for (i = 0; i < count; i++) {
 359             if (surfaces[i]) {
 360                ret = svga->swc->resource_rebind(svga->swc,
 361                                                 surfaces[i],
 362                                                 NULL,
 363                                                 SVGA_RELOC_READ);
 364                if (ret != PIPE_OK)
 365                   return ret;
 366             }
 367          }
 368       }
 369    }
 370    svga->rebind.flags.texture_samplers = FALSE;
 371
 372    return PIPE_OK;
 373 }
 374
 375
 376 static enum pipe_error
 377 validate_constant_buffers(struct svga_context *svga)
 378 {
 379    unsigned shader;
 380
 381    assert(svga_have_vgpu10(svga));
 382
 383    for (shader = PIPE_SHADER_VERTEX; shader <= PIPE_SHADER_GEOMETRY; shader++) {
 384       enum pipe_error ret;
 385       struct svga_buffer *buffer;
 386       struct svga_winsys_surface *handle;
 387       unsigned enabled_constbufs;
 388
 389       /* Rebind the default constant buffer if needed */
 390       if (svga->rebind.flags.constbufs) {
 391          buffer = svga_buffer(svga->state.hw_draw.constbuf[shader]);
 392          if (buffer) {
 393             ret = svga->swc->resource_rebind(svga->swc,
 394                                              buffer->handle,
 395                                              NULL,
 396                                              SVGA_RELOC_READ);
 397             if (ret != PIPE_OK)
 398                return ret;
 399          }
 400       }
 401
 402       /*
 403        * Reference other bound constant buffers to ensure pending updates are
 404        * noticed by the device.
 405        */
 406       enabled_constbufs = svga->state.hw_draw.enabled_constbufs[shader] & ~1u;
 407       while (enabled_constbufs) {
 408          unsigned i = u_bit_scan(&enabled_constbufs);
 409          buffer = svga_buffer(svga->curr.constbufs[shader][i].buffer);
 410          if (buffer) {
 411             handle = svga_buffer_handle(svga, &buffer->b.b);
 412
 413             if (svga->rebind.flags.constbufs) {
 414                ret = svga->swc->resource_rebind(svga->swc,
 415                                                 handle,
 416                                                 NULL,
 417                                                 SVGA_RELOC_READ);
 418                if (ret != PIPE_OK)
 419                   return ret;
 420             }
 421          }
 422       }
 423    }
 424    svga->rebind.flags.constbufs = FALSE;
 425
 426    return PIPE_OK;
 427 }
 428
 429
 430 static enum pipe_error
 431 draw_vgpu10(struct svga_hwtnl *hwtnl,
 432             const SVGA3dPrimitiveRange *range,
 433             unsigned vcount,
 434             unsigned min_index,
 435             unsigned max_index, struct pipe_resource *ib,
 436             unsigned start_instance, unsigned instance_count)
 437 {
 438    struct svga_context *svga = hwtnl->svga;
 439    struct pipe_resource *vbuffers[SVGA3D_INPUTREG_MAX];
 440    struct svga_winsys_surface *vbuffer_handles[SVGA3D_INPUTREG_MAX];
 441    struct svga_winsys_surface *ib_handle;
 442    const unsigned vbuf_count = hwtnl->cmd.vbuf_count;
 443    enum pipe_error ret;
 444    unsigned i;
 445    boolean rebind_ib = FALSE;
 446    boolean rebind_vbuf = FALSE;
 447
 448    assert(svga_have_vgpu10(svga));
 449    assert(hwtnl->cmd.prim_count == 0);
 450
 451    /* We need to reemit all the current resource bindings along with the Draw
 452     * command to be sure that the referenced resources are available for the
 453     * Draw command, just in case the surfaces associated with the resources
 454     * are paged out.
 455     */
 456    if (svga->rebind.val) {
 457       ret = svga_rebind_framebuffer_bindings(svga);
 458       if (ret != PIPE_OK)
 459          return ret;
 460
 461       ret = svga_rebind_shaders(svga);
 462       if (ret != PIPE_OK)
 463          return ret;
 464
 465       /* Rebind stream output targets */
 466       ret = svga_rebind_stream_output_targets(svga);
 467       if (ret != PIPE_OK)
 468          return ret;
 469
 470       /* Force rebinding the index buffer when needed */
 471       rebind_ib = TRUE;
 472
 473       /* Force rebinding the vertex buffers */
 474       rebind_vbuf = TRUE;
 475    }
 476
 477    ret = validate_sampler_resources(svga);
 478    if (ret != PIPE_OK)
 479       return ret;
 480
 481    ret = validate_constant_buffers(svga);
 482    if (ret != PIPE_OK)
 483       return ret;
 484
 485    /* Get handle for each referenced vertex buffer */
 486    for (i = 0; i < vbuf_count; i++) {
 487       struct svga_buffer *sbuf = svga_buffer(hwtnl->cmd.vbufs[i].buffer);
 488
 489       if (sbuf) {
 490          assert(sbuf->key.flags & SVGA3D_SURFACE_BIND_VERTEX_BUFFER);
 491          vbuffer_handles[i] = svga_buffer_handle(svga, &sbuf->b.b);
 492          if (vbuffer_handles[i] == NULL)
 493             return PIPE_ERROR_OUT_OF_MEMORY;
 494          vbuffers[i] = &sbuf->b.b;
 495       }
 496       else {
 497          vbuffers[i] = NULL;
 498          vbuffer_handles[i] = NULL;
 499       }
 500    }
 501
 502    for (; i < svga->state.hw_draw.num_vbuffers; i++) {
 503       vbuffers[i] = NULL;
 504       vbuffer_handles[i] = NULL;
 505    }
 506
 507    /* Get handle for the index buffer */
 508    if (ib) {
 509       struct svga_buffer *sbuf = svga_buffer(ib);
 510
 511       assert(sbuf->key.flags & SVGA3D_SURFACE_BIND_INDEX_BUFFER);
 512       (void) sbuf; /* silence unused var warning */
 513
 514       ib_handle = svga_buffer_handle(svga, ib);
 515       if (!ib_handle)
 516          return PIPE_ERROR_OUT_OF_MEMORY;
 517    }
 518    else {
 519       ib_handle = NULL;
 520    }
 521
 522    /* setup vertex attribute input layout */
 523    if (svga->state.hw_draw.layout_id != hwtnl->cmd.vdecl_layout_id) {
 524       ret = SVGA3D_vgpu10_SetInputLayout(svga->swc,
 525                                          hwtnl->cmd.vdecl_layout_id);
 526       if (ret != PIPE_OK)
 527          return ret;
 528
 529       svga->state.hw_draw.layout_id = hwtnl->cmd.vdecl_layout_id;
 530    }
 531
 532    /* setup vertex buffers */
 533    {
 534       SVGA3dVertexBuffer vbuffer_attrs[PIPE_MAX_ATTRIBS];
 535
 536       memset(vbuffer_attrs, 0, sizeof(vbuffer_attrs));
 537
 538       for (i = 0; i < vbuf_count; i++) {
 539          vbuffer_attrs[i].stride = hwtnl->cmd.vbufs[i].stride;
 540          vbuffer_attrs[i].offset = hwtnl->cmd.vbufs[i].buffer_offset;
 541       }
 542
 543       /* If we haven't yet emitted a drawing command or if any
 544        * vertex buffer state is changing, issue that state now.
 545        */
 546       if (rebind_vbuf ||
 547           ((hwtnl->cmd.swc->hints & SVGA_HINT_FLAG_CAN_PRE_FLUSH) == 0) ||
 548           vbuf_count != svga->state.hw_draw.num_vbuffers ||
 549           memcmp(vbuffer_attrs, svga->state.hw_draw.vbuffer_attrs,
 550                  vbuf_count * sizeof(vbuffer_attrs[0])) ||
 551           memcmp(vbuffers, svga->state.hw_draw.vbuffers,
 552                  vbuf_count * sizeof(vbuffers[0]))) {
 553
 554          unsigned num_vbuffers;
 555
 556          /* get the max of the current bound vertex buffers count and
 557           * the to-be-bound vertex buffers count, so as to unbind
 558           * the unused vertex buffers.
 559           */
 560          num_vbuffers = MAX2(vbuf_count, svga->state.hw_draw.num_vbuffers);
 561
 562          if (num_vbuffers > 0) {
 563
 564             ret = SVGA3D_vgpu10_SetVertexBuffers(svga->swc, num_vbuffers,
 565                                                  0,    /* startBuffer */
 566                                                  vbuffer_attrs,
 567                                                  vbuffer_handles);
 568             if (ret != PIPE_OK)
 569                return ret;
 570
 571             svga->state.hw_draw.num_vbuffers = num_vbuffers;
 572             memcpy(svga->state.hw_draw.vbuffer_attrs, vbuffer_attrs,
 573                    num_vbuffers * sizeof(vbuffer_attrs[0]));
 574             for (i = 0; i < num_vbuffers; i++) {
 575                pipe_resource_reference(&svga->state.hw_draw.vbuffers[i],
 576                                        vbuffers[i]);
 577             }
 578          }
 579       }
 580    }
 581
 582    /* Set primitive type (line, tri, etc) */
 583    if (svga->state.hw_draw.topology != range->primType) {
 584       ret = SVGA3D_vgpu10_SetTopology(svga->swc, range->primType);
 585       if (ret != PIPE_OK)
 586          return ret;
 587
 588       svga->state.hw_draw.topology = range->primType;
 589    }
 590
 591    if (ib_handle) {
 592       /* indexed drawing */
 593       SVGA3dSurfaceFormat indexFormat = xlate_index_format(range->indexWidth);
 594
 595       /* setup index buffer */
 596       if (rebind_ib ||
 597           ib != svga->state.hw_draw.ib ||
 598           indexFormat != svga->state.hw_draw.ib_format ||
 599           range->indexArray.offset != svga->state.hw_draw.ib_offset) {
 600
 601          assert(indexFormat != SVGA3D_FORMAT_INVALID);
 602          ret = SVGA3D_vgpu10_SetIndexBuffer(svga->swc, ib_handle,
 603                                             indexFormat,
 604                                             range->indexArray.offset);
 605          if (ret != PIPE_OK)
 606             return ret;
 607
 608          pipe_resource_reference(&svga->state.hw_draw.ib, ib);
 609          svga->state.hw_draw.ib_format = indexFormat;
 610          svga->state.hw_draw.ib_offset = range->indexArray.offset;
 611       }
 612
 613       if (instance_count > 1) {
 614          ret = SVGA3D_vgpu10_DrawIndexedInstanced(svga->swc,
 615                                                   vcount,
 616                                                   instance_count,
 617                                                   0, /* startIndexLocation */
 618                                                   range->indexBias,
 619                                                   start_instance);
 620          if (ret != PIPE_OK)
 621             return ret;
 622       }
 623       else {
 624          /* non-instanced drawing */
 625          ret = SVGA3D_vgpu10_DrawIndexed(svga->swc,
 626                                          vcount,
 627                                          0,      /* startIndexLocation */
 628                                          range->indexBias);
 629          if (ret != PIPE_OK)
 630             return ret;
 631       }
 632    }
 633    else {
 634       /* non-indexed drawing */
 635       if (svga->state.hw_draw.ib_format != SVGA3D_FORMAT_INVALID ||
 636           svga->state.hw_draw.ib != NULL) {
 637          /* Unbind previously bound index buffer */
 638          ret = SVGA3D_vgpu10_SetIndexBuffer(svga->swc, NULL,
 639                                             SVGA3D_FORMAT_INVALID, 0);
 640          if (ret != PIPE_OK)
 641             return ret;
 642          pipe_resource_reference(&svga->state.hw_draw.ib, NULL);
 643          svga->state.hw_draw.ib_format = SVGA3D_FORMAT_INVALID;
 644       }
 645
 646       assert(svga->state.hw_draw.ib == NULL);
 647
 648       if (instance_count > 1) {
 649          ret = SVGA3D_vgpu10_DrawInstanced(svga->swc,
 650                                            vcount,
 651                                            instance_count,
 652                                            range->indexBias,
 653                                            start_instance);
 654          if (ret != PIPE_OK)
 655             return ret;
 656       }
 657       else {
 658          /* non-instanced */
 659          ret = SVGA3D_vgpu10_Draw(svga->swc,
 660                                   vcount,
 661                                   range->indexBias);
 662          if (ret != PIPE_OK)
 663             return ret;
 664       }
 665    }
 666
 667    hwtnl->cmd.prim_count = 0;
 668
 669    return PIPE_OK;
 670 }
 671
 672
 673
 674 /**
 675  * Emit any pending drawing commands to the command buffer.
 676  * When we receive VGPU9 drawing commands we accumulate them and don't
 677  * immediately emit them into the command buffer.
 678  * This function needs to be called before we change state that could
 679  * effect those pending draws.
 680  */
 681 enum pipe_error
 682 svga_hwtnl_flush(struct svga_hwtnl *hwtnl)
 683 {
 684    if (!svga_have_vgpu10(hwtnl->svga) && hwtnl->cmd.prim_count) {
 685       /* we only queue up primitive for VGPU9 */
 686       return draw_vgpu9(hwtnl);
 687    }
 688    return PIPE_OK;
 689 }
 690
 691
 692 void
 693 svga_hwtnl_set_index_bias(struct svga_hwtnl *hwtnl, int index_bias)
 694 {
 695    hwtnl->index_bias = index_bias;
 696 }
 697
 698
 699
 700 /***********************************************************************
 701  * Internal functions:
 702  */
 703
 704 /**
 705  * For debugging only.
 706  */
 707 static void
 708 check_draw_params(struct svga_hwtnl *hwtnl,
 709                   const SVGA3dPrimitiveRange *range,
 710                   unsigned min_index, unsigned max_index,
 711                   struct pipe_resource *ib)
 712 {
 713    unsigned i;
 714
 715    assert(!svga_have_vgpu10(hwtnl->svga));
 716
 717    for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
 718       unsigned j = hwtnl->cmd.vdecl_buffer_index[i];
 719       const struct pipe_vertex_buffer *vb = &hwtnl->cmd.vbufs[j];
 720       unsigned size = vb->buffer ? vb->buffer->width0 : 0;
 721       unsigned offset = hwtnl->cmd.vdecl[i].array.offset;
 722       unsigned stride = hwtnl->cmd.vdecl[i].array.stride;
 723       int index_bias = (int) range->indexBias + hwtnl->index_bias;
 724       unsigned width;
 725
 726       if (size == 0)
 727          continue;
 728
 729       assert(vb);
 730       assert(size);
 731       assert(offset < size);
 732       assert(min_index <= max_index);
 733       (void) width;
 734       (void) stride;
 735       (void) offset;
 736       (void) size;
 737
 738       switch (hwtnl->cmd.vdecl[i].identity.type) {
 739       case SVGA3D_DECLTYPE_FLOAT1:
 740          width = 4;
 741          break;
 742       case SVGA3D_DECLTYPE_FLOAT2:
 743          width = 4 * 2;
 744          break;
 745       case SVGA3D_DECLTYPE_FLOAT3:
 746          width = 4 * 3;
 747          break;
 748       case SVGA3D_DECLTYPE_FLOAT4:
 749          width = 4 * 4;
 750          break;
 751       case SVGA3D_DECLTYPE_D3DCOLOR:
 752          width = 4;
 753          break;
 754       case SVGA3D_DECLTYPE_UBYTE4:
 755          width = 1 * 4;
 756          break;
 757       case SVGA3D_DECLTYPE_SHORT2:
 758          width = 2 * 2;
 759          break;
 760       case SVGA3D_DECLTYPE_SHORT4:
 761          width = 2 * 4;
 762          break;
 763       case SVGA3D_DECLTYPE_UBYTE4N:
 764          width = 1 * 4;
 765          break;
 766       case SVGA3D_DECLTYPE_SHORT2N:
 767          width = 2 * 2;
 768          break;
 769       case SVGA3D_DECLTYPE_SHORT4N:
 770          width = 2 * 4;
 771          break;
 772       case SVGA3D_DECLTYPE_USHORT2N:
 773          width = 2 * 2;
 774          break;
 775       case SVGA3D_DECLTYPE_USHORT4N:
 776          width = 2 * 4;
 777          break;
 778       case SVGA3D_DECLTYPE_UDEC3:
 779          width = 4;
 780          break;
 781       case SVGA3D_DECLTYPE_DEC3N:
 782          width = 4;
 783          break;
 784       case SVGA3D_DECLTYPE_FLOAT16_2:
 785          width = 2 * 2;
 786          break;
 787       case SVGA3D_DECLTYPE_FLOAT16_4:
 788          width = 2 * 4;
 789          break;
 790       default:
 791          assert(0);
 792          width = 0;
 793          break;
 794       }
 795
 796       if (index_bias >= 0) {
 797          assert(offset + index_bias * stride + width <= size);
 798       }
 799
 800       /*
 801        * min_index/max_index are merely conservative guesses, so we can't
 802        * make buffer overflow detection based on their values.
 803        */
 804    }
 805
 806    assert(range->indexWidth == range->indexArray.stride);
 807
 808    if (ib) {
 809       MAYBE_UNUSED unsigned size = ib->width0;
 810       MAYBE_UNUSED unsigned offset = range->indexArray.offset;
 811       MAYBE_UNUSED unsigned stride = range->indexArray.stride;
 812       MAYBE_UNUSED unsigned count;
 813
 814       assert(size);
 815       assert(offset < size);
 816       assert(stride);
 817
 818       switch (range->primType) {
 819       case SVGA3D_PRIMITIVE_POINTLIST:
 820          count = range->primitiveCount;
 821          break;
 822       case SVGA3D_PRIMITIVE_LINELIST:
 823          count = range->primitiveCount * 2;
 824          break;
 825       case SVGA3D_PRIMITIVE_LINESTRIP:
 826          count = range->primitiveCount + 1;
 827          break;
 828       case SVGA3D_PRIMITIVE_TRIANGLELIST:
 829          count = range->primitiveCount * 3;
 830          break;
 831       case SVGA3D_PRIMITIVE_TRIANGLESTRIP:
 832          count = range->primitiveCount + 2;
 833          break;
 834       case SVGA3D_PRIMITIVE_TRIANGLEFAN:
 835          count = range->primitiveCount + 2;
 836          break;
 837       default:
 838          assert(0);
 839          count = 0;
 840          break;
 841       }
 842
 843       assert(offset + count * stride <= size);
 844    }
 845 }
 846
 847
 848 /**
 849  * All drawing filters down into this function, either directly
 850  * on the hardware path or after doing software vertex processing.
 851  */
 852 enum pipe_error
 853 svga_hwtnl_prim(struct svga_hwtnl *hwtnl,
 854                 const SVGA3dPrimitiveRange * range,
 855                 unsigned vcount,
 856                 unsigned min_index,
 857                 unsigned max_index, struct pipe_resource *ib,
 858                 unsigned start_instance, unsigned instance_count)
 859 {
 860    enum pipe_error ret = PIPE_OK;
 861
 862    if (svga_have_vgpu10(hwtnl->svga)) {
 863       /* draw immediately */
 864       ret = draw_vgpu10(hwtnl, range, vcount, min_index, max_index, ib,
 865                         start_instance, instance_count);
 866       if (ret != PIPE_OK) {
 867          svga_context_flush(hwtnl->svga, NULL);
 868          ret = draw_vgpu10(hwtnl, range, vcount, min_index, max_index, ib,
 869                            start_instance, instance_count);
 870          assert(ret == PIPE_OK);
 871       }
 872    }
 873    else {
 874       /* batch up drawing commands */
 875 #ifdef DEBUG
 876       check_draw_params(hwtnl, range, min_index, max_index, ib);
 877       assert(start_instance == 0);
 878       assert(instance_count <= 1);
 879 #else
 880       (void) check_draw_params;
 881 #endif
 882
 883       if (hwtnl->cmd.prim_count + 1 >= QSZ) {
 884          ret = svga_hwtnl_flush(hwtnl);
 885          if (ret != PIPE_OK)
 886             return ret;
 887       }
 888
 889       /* min/max indices are relative to bias */
 890       hwtnl->cmd.min_index[hwtnl->cmd.prim_count] = min_index;
 891       hwtnl->cmd.max_index[hwtnl->cmd.prim_count] = max_index;
 892
 893       hwtnl->cmd.prim[hwtnl->cmd.prim_count] = *range;
 894       hwtnl->cmd.prim[hwtnl->cmd.prim_count].indexBias += hwtnl->index_bias;
 895
 896       pipe_resource_reference(&hwtnl->cmd.prim_ib[hwtnl->cmd.prim_count], ib);
 897       hwtnl->cmd.prim_count++;
 898    }
 899
 900    return ret;
 901 }