src/gallium/drivers/svga/svga_screen.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 "util/u_memory.h"
  27 #include "util/u_inlines.h"
  28 #include "util/u_string.h"
  29 #include "util/u_math.h"
  30
  31 #include "svga_winsys.h"
  32 #include "svga_public.h"
  33 #include "svga_context.h"
  34 #include "svga_format.h"
  35 #include "svga_screen.h"
  36 #include "svga_resource_texture.h"
  37 #include "svga_resource.h"
  38 #include "svga_debug.h"
  39
  40 #include "svga3d_shaderdefs.h"
  41
  42
  43 #ifdef DEBUG
  44 int SVGA_DEBUG = 0;
  45
  46 static const struct debug_named_value svga_debug_flags[] = {
  47    { "dma",      DEBUG_DMA, NULL },
  48    { "tgsi",     DEBUG_TGSI, NULL },
  49    { "pipe",     DEBUG_PIPE, NULL },
  50    { "state",    DEBUG_STATE, NULL },
  51    { "screen",   DEBUG_SCREEN, NULL },
  52    { "tex",      DEBUG_TEX, NULL },
  53    { "swtnl",    DEBUG_SWTNL, NULL },
  54    { "const",    DEBUG_CONSTS, NULL },
  55    { "viewport", DEBUG_VIEWPORT, NULL },
  56    { "views",    DEBUG_VIEWS, NULL },
  57    { "perf",     DEBUG_PERF, NULL },
  58    { "flush",    DEBUG_FLUSH, NULL },
  59    { "sync",     DEBUG_SYNC, NULL },
  60    { "cache",    DEBUG_CACHE, NULL },
  61    DEBUG_NAMED_VALUE_END
  62 };
  63 #endif
  64
  65 static const char *
  66 svga_get_vendor( struct pipe_screen *pscreen )
  67 {
  68    return "VMware, Inc.";
  69 }
  70
  71
  72 static const char *
  73 svga_get_name( struct pipe_screen *pscreen )
  74 {
  75    const char *build = "", *llvm = "", *mutex = "";
  76    static char name[100];
  77 #ifdef DEBUG
  78    /* Only return internal details in the DEBUG version:
  79     */
  80    build = "build: DEBUG;";
  81    mutex = "mutex: " PIPE_ATOMIC ";";
  82 #ifdef HAVE_LLVM
  83    llvm = "LLVM;";
  84 #endif
  85 #else
  86    build = "build: RELEASE;";
  87 #endif
  88
  89    util_snprintf(name, sizeof(name), "SVGA3D; %s %s %s", build, mutex, llvm);
  90    return name;
  91 }
  92
  93
  94
  95
  96 static float
  97 svga_get_paramf(struct pipe_screen *screen, enum pipe_capf param)
  98 {
  99    struct svga_screen *svgascreen = svga_screen(screen);
 100    struct svga_winsys_screen *sws = svgascreen->sws;
 101    SVGA3dDevCapResult result;
 102
 103    switch (param) {
 104    case PIPE_CAPF_MAX_LINE_WIDTH:
 105       /* fall-through */
 106    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 107       return 7.0;
 108
 109    case PIPE_CAPF_MAX_POINT_WIDTH:
 110       /* fall-through */
 111    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 112       /* Keep this to a reasonable size to avoid failures in
 113        * conform/pntaa.c:
 114        */
 115       return SVGA_MAX_POINTSIZE;
 116
 117    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 118       if(!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_ANISOTROPY, &result))
 119          return 4.0;
 120       return result.u;
 121
 122    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 123       return 16.0;
 124
 125    default:
 126       return 0;
 127    }
 128 }
 129
 130
 131 static int
 132 svga_get_param(struct pipe_screen *screen, enum pipe_cap param)
 133 {
 134    struct svga_screen *svgascreen = svga_screen(screen);
 135    struct svga_winsys_screen *sws = svgascreen->sws;
 136    SVGA3dDevCapResult result;
 137
 138    switch (param) {
 139    case PIPE_CAP_MAX_COMBINED_SAMPLERS:
 140       return 16;
 141    case PIPE_CAP_NPOT_TEXTURES:
 142       return 1;
 143    case PIPE_CAP_TWO_SIDED_STENCIL:
 144       return 1;
 145    case PIPE_CAP_ANISOTROPIC_FILTER:
 146       return 1;
 147    case PIPE_CAP_POINT_SPRITE:
 148       return 1;
 149    case PIPE_CAP_MAX_RENDER_TARGETS:
 150       if(!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_RENDER_TARGETS, &result))
 151          return 1;
 152       if(!result.u)
 153          return 1;
 154       return MIN2(result.u, PIPE_MAX_COLOR_BUFS);
 155    case PIPE_CAP_OCCLUSION_QUERY:
 156       return 1;
 157    case PIPE_CAP_TIMER_QUERY:
 158       return 0;
 159    case PIPE_CAP_TEXTURE_SHADOW_MAP:
 160       return 1;
 161    case PIPE_CAP_TEXTURE_SWIZZLE:
 162       return 1;
 163
 164    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 165       {
 166          unsigned levels = SVGA_MAX_TEXTURE_LEVELS;
 167          if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_WIDTH, &result))
 168             levels = MIN2(util_logbase2(result.u) + 1, levels);
 169          else
 170             levels = 12 /* 2048x2048 */;
 171          if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT, &result))
 172             levels = MIN2(util_logbase2(result.u) + 1, levels);
 173          else
 174             levels = 12 /* 2048x2048 */;
 175          return levels;
 176       }
 177
 178    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 179       if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VOLUME_EXTENT, &result))
 180          return 8;  /* max 128x128x128 */
 181       return MIN2(util_logbase2(result.u) + 1, SVGA_MAX_TEXTURE_LEVELS);
 182
 183    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 184       /*
 185        * No mechanism to query the host, and at least limited to 2048x2048 on
 186        * certain hardware.
 187        */
 188       return MIN2(screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS),
 189                   12 /* 2048x2048 */);
 190
 191    case PIPE_CAP_BLEND_EQUATION_SEPARATE: /* req. for GL 1.5 */
 192       return 1;
 193
 194    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 195    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 196       return 1;
 197    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 198    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 199       return 0;
 200
 201    case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
 202       return 1;
 203    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 204       return 0;
 205
 206    default:
 207       return 0;
 208    }
 209 }
 210
 211 static int svga_get_shader_param(struct pipe_screen *screen, unsigned shader, enum pipe_shader_cap param)
 212 {
 213    struct svga_screen *svgascreen = svga_screen(screen);
 214    struct svga_winsys_screen *sws = svgascreen->sws;
 215    SVGA3dDevCapResult result;
 216
 217    switch (shader)
 218    {
 219    case PIPE_SHADER_FRAGMENT:
 220       switch (param)
 221       {
 222       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 223       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 224       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 225       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 226          return 512;
 227       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 228          return SVGA3D_MAX_NESTING_LEVEL;
 229       case PIPE_SHADER_CAP_MAX_INPUTS:
 230          return 10;
 231       case PIPE_SHADER_CAP_MAX_CONSTS:
 232          return 224;
 233       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 234          return 1;
 235       case PIPE_SHADER_CAP_MAX_TEMPS:
 236          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_FRAGMENT_SHADER_TEMPS, &result))
 237             return 32;
 238          return result.u;
 239       case PIPE_SHADER_CAP_MAX_ADDRS:
 240       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 241          /*
 242           * Although PS 3.0 has some addressing abilities it can only represent
 243           * loops that can be statically determined and unrolled. Given we can
 244           * only handle a subset of the cases that the state tracker already
 245           * does it is better to defer loop unrolling to the state tracker.
 246           */
 247          return 0;
 248       case PIPE_SHADER_CAP_MAX_PREDS:
 249          return 1;
 250       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 251          return 1;
 252       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 253       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 254       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 255          return 0;
 256       case PIPE_SHADER_CAP_SUBROUTINES:
 257          return 0;
 258       case PIPE_SHADER_CAP_INTEGERS:
 259          return 0;
 260       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 261          return 16;
 262       case PIPE_SHADER_CAP_OUTPUT_READ:
 263          return 0;
 264       }
 265       break;
 266    case PIPE_SHADER_VERTEX:
 267       switch (param)
 268       {
 269       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 270       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 271          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_INSTRUCTIONS, &result))
 272             return 512;
 273          return result.u;
 274       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 275       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 276          /* XXX: until we have vertex texture support */
 277          return 0;
 278       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 279          return SVGA3D_MAX_NESTING_LEVEL;
 280       case PIPE_SHADER_CAP_MAX_INPUTS:
 281          return 16;
 282       case PIPE_SHADER_CAP_MAX_CONSTS:
 283          return 256;
 284       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 285          return 1;
 286       case PIPE_SHADER_CAP_MAX_TEMPS:
 287          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_TEMPS, &result))
 288             return 32;
 289          return result.u;
 290       case PIPE_SHADER_CAP_MAX_ADDRS:
 291          return 1;
 292       case PIPE_SHADER_CAP_MAX_PREDS:
 293          return 1;
 294       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 295          return 1;
 296       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 297       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 298          return 1;
 299       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 300          return 0;
 301       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 302          return 1;
 303       case PIPE_SHADER_CAP_SUBROUTINES:
 304          return 0;
 305       case PIPE_SHADER_CAP_INTEGERS:
 306          return 0;
 307       case PIPE_SHADER_CAP_OUTPUT_READ:
 308          return 0;
 309       default:
 310          break;
 311       }
 312       break;
 313    default:
 314       break;
 315    }
 316    return 0;
 317 }
 318
 319
 320 static boolean
 321 svga_is_format_supported( struct pipe_screen *screen,
 322                           enum pipe_format format,
 323                           enum pipe_texture_target target,
 324                           unsigned sample_count,
 325                           unsigned tex_usage)
 326 {
 327    struct svga_screen *ss = svga_screen(screen);
 328    SVGA3dSurfaceFormat svga_format;
 329    SVGA3dSurfaceFormatCaps caps;
 330    SVGA3dSurfaceFormatCaps mask;
 331
 332    assert(tex_usage);
 333
 334    if (sample_count > 1) {
 335       return FALSE;
 336    }
 337
 338    svga_format = svga_translate_format(ss, format, tex_usage);
 339    if (svga_format == SVGA3D_FORMAT_INVALID) {
 340       return FALSE;
 341    }
 342
 343    /*
 344     * Override host capabilities, so that we end up with the same
 345     * visuals for all virtual hardware implementations.
 346     */
 347
 348    if (tex_usage & PIPE_BIND_DISPLAY_TARGET) {
 349       switch (svga_format) {
 350       case SVGA3D_A8R8G8B8:
 351       case SVGA3D_X8R8G8B8:
 352       case SVGA3D_R5G6B5:
 353          break;
 354
 355       /* Often unsupported/problematic. This means we end up with the same
 356        * visuals for all virtual hardware implementations.
 357        */
 358       case PIPE_FORMAT_B4G4R4A4_UNORM:
 359       case PIPE_FORMAT_B5G5R5A1_UNORM:
 360          return FALSE;
 361
 362       default:
 363          return FALSE;
 364       }
 365    }
 366
 367    /*
 368     * Query the host capabilities.
 369     */
 370
 371    svga_get_format_cap(ss, svga_format, &caps);
 372
 373    mask.value = 0;
 374    if (tex_usage & PIPE_BIND_RENDER_TARGET) {
 375       mask.offscreenRenderTarget = 1;
 376    }
 377    if (tex_usage & PIPE_BIND_DEPTH_STENCIL) {
 378       mask.zStencil = 1;
 379    }
 380    if (tex_usage & PIPE_BIND_SAMPLER_VIEW) {
 381       mask.texture = 1;
 382    }
 383
 384    return (caps.value & mask.value) == mask.value;
 385 }
 386
 387
 388 static void
 389 svga_fence_reference(struct pipe_screen *screen,
 390                      struct pipe_fence_handle **ptr,
 391                      struct pipe_fence_handle *fence)
 392 {
 393    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 394    sws->fence_reference(sws, ptr, fence);
 395 }
 396
 397
 398 static boolean
 399 svga_fence_signalled(struct pipe_screen *screen,
 400                      struct pipe_fence_handle *fence)
 401 {
 402    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 403    return sws->fence_signalled(sws, fence, 0) == 0;
 404 }
 405
 406
 407 static boolean
 408 svga_fence_finish(struct pipe_screen *screen,
 409                   struct pipe_fence_handle *fence,
 410                   uint64_t timeout)
 411 {
 412    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 413
 414    SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "%s fence_ptr %p\n",
 415             __FUNCTION__, fence);
 416
 417    return sws->fence_finish(sws, fence, 0) == 0;
 418 }
 419
 420
 421 static void
 422 svga_destroy_screen( struct pipe_screen *screen )
 423 {
 424    struct svga_screen *svgascreen = svga_screen(screen);
 425
 426    svga_screen_cache_cleanup(svgascreen);
 427
 428    pipe_mutex_destroy(svgascreen->swc_mutex);
 429    pipe_mutex_destroy(svgascreen->tex_mutex);
 430
 431    svgascreen->sws->destroy(svgascreen->sws);
 432
 433    FREE(svgascreen);
 434 }
 435
 436
 437 /**
 438  * Create a new svga_screen object
 439  */
 440 struct pipe_screen *
 441 svga_screen_create(struct svga_winsys_screen *sws)
 442 {
 443    struct svga_screen *svgascreen;
 444    struct pipe_screen *screen;
 445    SVGA3dDevCapResult result;
 446    boolean use_vs30, use_ps30;
 447
 448 #ifdef DEBUG
 449    SVGA_DEBUG = debug_get_flags_option("SVGA_DEBUG", svga_debug_flags, 0 );
 450 #endif
 451
 452    svgascreen = CALLOC_STRUCT(svga_screen);
 453    if (!svgascreen)
 454       goto error1;
 455
 456    svgascreen->debug.force_level_surface_view =
 457       debug_get_bool_option("SVGA_FORCE_LEVEL_SURFACE_VIEW", FALSE);
 458    svgascreen->debug.force_surface_view =
 459       debug_get_bool_option("SVGA_FORCE_SURFACE_VIEW", FALSE);
 460    svgascreen->debug.force_sampler_view =
 461       debug_get_bool_option("SVGA_FORCE_SAMPLER_VIEW", FALSE);
 462    svgascreen->debug.no_surface_view =
 463       debug_get_bool_option("SVGA_NO_SURFACE_VIEW", FALSE);
 464    svgascreen->debug.no_sampler_view =
 465       debug_get_bool_option("SVGA_NO_SAMPLER_VIEW", FALSE);
 466
 467    screen = &svgascreen->screen;
 468
 469    screen->destroy = svga_destroy_screen;
 470    screen->get_name = svga_get_name;
 471    screen->get_vendor = svga_get_vendor;
 472    screen->get_param = svga_get_param;
 473    screen->get_shader_param = svga_get_shader_param;
 474    screen->get_paramf = svga_get_paramf;
 475    screen->is_format_supported = svga_is_format_supported;
 476    screen->context_create = svga_context_create;
 477    screen->fence_reference = svga_fence_reference;
 478    screen->fence_signalled = svga_fence_signalled;
 479    screen->fence_finish = svga_fence_finish;
 480    svgascreen->sws = sws;
 481
 482    svga_init_screen_resource_functions(svgascreen);
 483
 484    if (sws->get_hw_version) {
 485       svgascreen->hw_version = sws->get_hw_version(sws);
 486    } else {
 487       svgascreen->hw_version = SVGA3D_HWVERSION_WS65_B1;
 488    }
 489
 490    use_ps30 =
 491       sws->get_cap(sws, SVGA3D_DEVCAP_FRAGMENT_SHADER_VERSION, &result) &&
 492       result.u >= SVGA3DPSVERSION_30 ? TRUE : FALSE;
 493
 494    use_vs30 =
 495       sws->get_cap(sws, SVGA3D_DEVCAP_VERTEX_SHADER_VERSION, &result) &&
 496       result.u >= SVGA3DVSVERSION_30 ? TRUE : FALSE;
 497
 498    /* we require Shader model 3.0 or later */
 499    if (!use_ps30 || !use_vs30)
 500       goto error2;
 501
 502    /*
 503     * The D16, D24X8, and D24S8 formats always do an implicit shadow compare
 504     * when sampled from, where as the DF16, DF24, and D24S8_INT do not.  So
 505     * we prefer the later when available.
 506     *
 507     * This mimics hardware vendors extensions for D3D depth sampling. See also
 508     * http://aras-p.info/texts/D3D9GPUHacks.html
 509     */
 510
 511    {
 512       boolean has_df16, has_df24, has_d24s8_int;
 513       SVGA3dSurfaceFormatCaps caps;
 514       SVGA3dSurfaceFormatCaps mask;
 515       mask.value = 0;
 516       mask.zStencil = 1;
 517       mask.texture = 1;
 518
 519       svgascreen->depth.z16 = SVGA3D_Z_D16;
 520       svgascreen->depth.x8z24 = SVGA3D_Z_D24X8;
 521       svgascreen->depth.s8z24 = SVGA3D_Z_D24S8;
 522
 523       svga_get_format_cap(svgascreen, SVGA3D_Z_DF16, &caps);
 524       has_df16 = (caps.value & mask.value) == mask.value;
 525
 526       svga_get_format_cap(svgascreen, SVGA3D_Z_DF24, &caps);
 527       has_df24 = (caps.value & mask.value) == mask.value;
 528
 529       svga_get_format_cap(svgascreen, SVGA3D_Z_D24S8_INT, &caps);
 530       has_d24s8_int = (caps.value & mask.value) == mask.value;
 531
 532       /* XXX: We might want some other logic here.
 533        * Like if we only have d24s8_int we should
 534        * emulate the other formats with that.
 535        */
 536       if (has_df16) {
 537          svgascreen->depth.z16 = SVGA3D_Z_DF16;
 538       }
 539       if (has_df24) {
 540          svgascreen->depth.x8z24 = SVGA3D_Z_DF24;
 541       }
 542       if (has_d24s8_int) {
 543          svgascreen->depth.s8z24 = SVGA3D_Z_D24S8_INT;
 544       }
 545    }
 546
 547    pipe_mutex_init(svgascreen->tex_mutex);
 548    pipe_mutex_init(svgascreen->swc_mutex);
 549
 550    svga_screen_cache_init(svgascreen);
 551
 552    return screen;
 553 error2:
 554    FREE(svgascreen);
 555 error1:
 556    return NULL;
 557 }
 558
 559 struct svga_winsys_screen *
 560 svga_winsys_screen(struct pipe_screen *screen)
 561 {
 562    return svga_screen(screen)->sws;
 563 }
 564
 565 #ifdef DEBUG
 566 struct svga_screen *
 567 svga_screen(struct pipe_screen *screen)
 568 {
 569    assert(screen);
 570    assert(screen->destroy == svga_destroy_screen);
 571    return (struct svga_screen *)screen;
 572 }
 573 #endif