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_cap 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_CAP_MAX_LINE_WIDTH:
 105       /* fall-through */
 106    case PIPE_CAP_MAX_LINE_WIDTH_AA:
 107       return 7.0;
 108
 109    case PIPE_CAP_MAX_POINT_WIDTH:
 110       /* fall-through */
 111    case PIPE_CAP_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_CAP_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_CAP_MAX_TEXTURE_LOD_BIAS:
 123       return 16.0;
 124
 125    case PIPE_CAP_MAX_COMBINED_SAMPLERS:
 126       return 16;
 127    case PIPE_CAP_NPOT_TEXTURES:
 128       return 1;
 129    case PIPE_CAP_TWO_SIDED_STENCIL:
 130       return 1;
 131    case PIPE_CAP_GLSL:
 132       return svgascreen->use_ps30 && svgascreen->use_vs30;
 133    case PIPE_CAP_ANISOTROPIC_FILTER:
 134       return 1;
 135    case PIPE_CAP_POINT_SPRITE:
 136       return 1;
 137    case PIPE_CAP_MAX_RENDER_TARGETS:
 138       if(!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_RENDER_TARGETS, &result))
 139          return 1;
 140       if(!result.u)
 141          return 1;
 142       return MIN2(result.u, PIPE_MAX_COLOR_BUFS);
 143    case PIPE_CAP_OCCLUSION_QUERY:
 144       return 1;
 145    case PIPE_CAP_TIMER_QUERY:
 146       return 0;
 147    case PIPE_CAP_TEXTURE_SHADOW_MAP:
 148       return 1;
 149    case PIPE_CAP_TEXTURE_SWIZZLE:
 150       return 1;
 151
 152    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 153       {
 154          unsigned levels = SVGA_MAX_TEXTURE_LEVELS;
 155          if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_WIDTH, &result))
 156             levels = MIN2(util_logbase2(result.u) + 1, levels);
 157          else
 158             levels = 12 /* 2048x2048 */;
 159          if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT, &result))
 160             levels = MIN2(util_logbase2(result.u) + 1, levels);
 161          else
 162             levels = 12 /* 2048x2048 */;
 163          return levels;
 164       }
 165
 166    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 167       if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VOLUME_EXTENT, &result))
 168          return 8;  /* max 128x128x128 */
 169       return MIN2(util_logbase2(result.u) + 1, SVGA_MAX_TEXTURE_LEVELS);
 170
 171    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 172       /*
 173        * No mechanism to query the host, and at least limited to 2048x2048 on
 174        * certain hardware.
 175        */
 176       return MIN2(screen->get_paramf(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS),
 177                   12.0 /* 2048x2048 */);
 178
 179    case PIPE_CAP_BLEND_EQUATION_SEPARATE: /* req. for GL 1.5 */
 180       return 1;
 181
 182    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 183    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 184       return 1;
 185    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 186    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 187       return 0;
 188
 189    case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
 190       return 1;
 191    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 192       return 0;
 193
 194    default:
 195       return 0;
 196    }
 197 }
 198
 199
 200 /* This is a fairly pointless interface
 201  */
 202 static int
 203 svga_get_param(struct pipe_screen *screen, enum pipe_cap param)
 204 {
 205    return (int) svga_get_paramf( screen, param );
 206 }
 207
 208 static int svga_get_shader_param(struct pipe_screen *screen, unsigned shader, enum pipe_shader_cap param)
 209 {
 210    struct svga_screen *svgascreen = svga_screen(screen);
 211    struct svga_winsys_screen *sws = svgascreen->sws;
 212    SVGA3dDevCapResult result;
 213
 214    switch (shader)
 215    {
 216    case PIPE_SHADER_FRAGMENT:
 217       switch (param)
 218       {
 219       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 220       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 221       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 222       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 223          return svgascreen->use_ps30 ? 512 : 96;
 224       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 225          return SVGA3D_MAX_NESTING_LEVEL;
 226       case PIPE_SHADER_CAP_MAX_INPUTS:
 227          return 10;
 228       case PIPE_SHADER_CAP_MAX_CONSTS:
 229          return svgascreen->use_ps30 ? 224 : 16;
 230       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 231          return 1;
 232       case PIPE_SHADER_CAP_MAX_TEMPS:
 233          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_FRAGMENT_SHADER_TEMPS, &result))
 234             return svgascreen->use_ps30 ? 32 : 12;
 235          return result.u;
 236       case PIPE_SHADER_CAP_MAX_ADDRS:
 237       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 238          /*
 239           * Although PS 3.0 has some addressing abilities it can only represent
 240           * loops that can be statically determined and unrolled. Given we can
 241           * only handle a subset of the cases that the state tracker already
 242           * does it is better to defer loop unrolling to the state tracker.
 243           */
 244          return 0;
 245       case PIPE_SHADER_CAP_MAX_PREDS:
 246          return svgascreen->use_ps30 ? 1 : 0;
 247       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 248          return 1;
 249       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 250       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 251       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 252          return 0;
 253       case PIPE_SHADER_CAP_SUBROUTINES:
 254          return 0;
 255       case PIPE_SHADER_CAP_INTEGERS:
 256          return 0;
 257       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 258          return 16;
 259       }
 260       break;
 261    case PIPE_SHADER_VERTEX:
 262       switch (param)
 263       {
 264       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 265       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 266          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_INSTRUCTIONS, &result))
 267             return svgascreen->use_vs30 ? 512 : 256;
 268          return result.u;
 269       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 270       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 271          /* XXX: until we have vertex texture support */
 272          return 0;
 273       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 274          return SVGA3D_MAX_NESTING_LEVEL;
 275       case PIPE_SHADER_CAP_MAX_INPUTS:
 276          return 16;
 277       case PIPE_SHADER_CAP_MAX_CONSTS:
 278          return 256;
 279       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 280          return 1;
 281       case PIPE_SHADER_CAP_MAX_TEMPS:
 282          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_TEMPS, &result))
 283             return svgascreen->use_vs30 ? 32 : 12;
 284          return result.u;
 285       case PIPE_SHADER_CAP_MAX_ADDRS:
 286          return svgascreen->use_vs30 ? 1 : 0;
 287       case PIPE_SHADER_CAP_MAX_PREDS:
 288          return svgascreen->use_vs30 ? 1 : 0;
 289       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 290          return 1;
 291       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 292       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 293          return svgascreen->use_vs30 ? 1 : 0;
 294       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 295          return 0;
 296       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 297          return 1;
 298       case PIPE_SHADER_CAP_SUBROUTINES:
 299          return 0;
 300       case PIPE_SHADER_CAP_INTEGERS:
 301          return 0;
 302       default:
 303          break;
 304       }
 305       break;
 306    default:
 307       break;
 308    }
 309    return 0;
 310 }
 311
 312
 313 static boolean
 314 svga_is_format_supported( struct pipe_screen *screen,
 315                           enum pipe_format format,
 316                           enum pipe_texture_target target,
 317                           unsigned sample_count,
 318                           unsigned tex_usage)
 319 {
 320    struct svga_screen *ss = svga_screen(screen);
 321    SVGA3dSurfaceFormat svga_format;
 322    SVGA3dSurfaceFormatCaps caps;
 323    SVGA3dSurfaceFormatCaps mask;
 324
 325    assert(tex_usage);
 326
 327    if (sample_count > 1) {
 328       return FALSE;
 329    }
 330
 331    svga_format = svga_translate_format(ss, format, tex_usage);
 332    if (svga_format == SVGA3D_FORMAT_INVALID) {
 333       return FALSE;
 334    }
 335
 336    /*
 337     * Override host capabilities, so that we end up with the same
 338     * visuals for all virtual hardware implementations.
 339     */
 340
 341    if (tex_usage & PIPE_BIND_DISPLAY_TARGET) {
 342       switch (svga_format) {
 343       case SVGA3D_A8R8G8B8:
 344       case SVGA3D_X8R8G8B8:
 345       case SVGA3D_R5G6B5:
 346          break;
 347
 348       /* Often unsupported/problematic. This means we end up with the same
 349        * visuals for all virtual hardware implementations.
 350        */
 351       case PIPE_FORMAT_B4G4R4A4_UNORM:
 352       case PIPE_FORMAT_B5G5R5A1_UNORM:
 353          return FALSE;
 354
 355       default:
 356          return FALSE;
 357       }
 358    }
 359
 360    /*
 361     * Query the host capabilities.
 362     */
 363
 364    svga_get_format_cap(ss, svga_format, &caps);
 365
 366    mask.value = 0;
 367    if (tex_usage & PIPE_BIND_RENDER_TARGET) {
 368       mask.offscreenRenderTarget = 1;
 369    }
 370    if (tex_usage & PIPE_BIND_DEPTH_STENCIL) {
 371       mask.zStencil = 1;
 372    }
 373    if (tex_usage & PIPE_BIND_SAMPLER_VIEW) {
 374       mask.texture = 1;
 375    }
 376
 377    return (caps.value & mask.value) == mask.value;
 378 }
 379
 380
 381 static void
 382 svga_fence_reference(struct pipe_screen *screen,
 383                      struct pipe_fence_handle **ptr,
 384                      struct pipe_fence_handle *fence)
 385 {
 386    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 387    sws->fence_reference(sws, ptr, fence);
 388 }
 389
 390
 391 static boolean
 392 svga_fence_signalled(struct pipe_screen *screen,
 393                      struct pipe_fence_handle *fence)
 394 {
 395    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 396    return sws->fence_signalled(sws, fence, 0) == 0;
 397 }
 398
 399
 400 static boolean
 401 svga_fence_finish(struct pipe_screen *screen,
 402                   struct pipe_fence_handle *fence,
 403                   uint64_t timeout)
 404 {
 405    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 406
 407    SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "%s fence_ptr %p\n",
 408             __FUNCTION__, fence);
 409
 410    return sws->fence_finish(sws, fence, 0) == 0;
 411 }
 412
 413
 414 static void
 415 svga_destroy_screen( struct pipe_screen *screen )
 416 {
 417    struct svga_screen *svgascreen = svga_screen(screen);
 418
 419    svga_screen_cache_cleanup(svgascreen);
 420
 421    pipe_mutex_destroy(svgascreen->swc_mutex);
 422    pipe_mutex_destroy(svgascreen->tex_mutex);
 423
 424    svgascreen->sws->destroy(svgascreen->sws);
 425
 426    FREE(svgascreen);
 427 }
 428
 429
 430 /**
 431  * Create a new svga_screen object
 432  */
 433 struct pipe_screen *
 434 svga_screen_create(struct svga_winsys_screen *sws)
 435 {
 436    struct svga_screen *svgascreen;
 437    struct pipe_screen *screen;
 438    SVGA3dDevCapResult result;
 439
 440 #ifdef DEBUG
 441    SVGA_DEBUG = debug_get_flags_option("SVGA_DEBUG", svga_debug_flags, 0 );
 442 #endif
 443
 444    svgascreen = CALLOC_STRUCT(svga_screen);
 445    if (!svgascreen)
 446       goto error1;
 447
 448    svgascreen->debug.force_level_surface_view =
 449       debug_get_bool_option("SVGA_FORCE_LEVEL_SURFACE_VIEW", FALSE);
 450    svgascreen->debug.force_surface_view =
 451       debug_get_bool_option("SVGA_FORCE_SURFACE_VIEW", FALSE);
 452    svgascreen->debug.force_sampler_view =
 453       debug_get_bool_option("SVGA_FORCE_SAMPLER_VIEW", FALSE);
 454    svgascreen->debug.no_surface_view =
 455       debug_get_bool_option("SVGA_NO_SURFACE_VIEW", FALSE);
 456    svgascreen->debug.no_sampler_view =
 457       debug_get_bool_option("SVGA_NO_SAMPLER_VIEW", FALSE);
 458
 459    screen = &svgascreen->screen;
 460
 461    screen->destroy = svga_destroy_screen;
 462    screen->get_name = svga_get_name;
 463    screen->get_vendor = svga_get_vendor;
 464    screen->get_param = svga_get_param;
 465    screen->get_shader_param = svga_get_shader_param;
 466    screen->get_paramf = svga_get_paramf;
 467    screen->is_format_supported = svga_is_format_supported;
 468    screen->context_create = svga_context_create;
 469    screen->fence_reference = svga_fence_reference;
 470    screen->fence_signalled = svga_fence_signalled;
 471    screen->fence_finish = svga_fence_finish;
 472    svgascreen->sws = sws;
 473
 474    svga_init_screen_resource_functions(svgascreen);
 475
 476    if (sws->get_hw_version) {
 477       svgascreen->hw_version = sws->get_hw_version(sws);
 478    } else {
 479       svgascreen->hw_version = SVGA3D_HWVERSION_WS65_B1;
 480    }
 481
 482    svgascreen->use_ps30 =
 483       sws->get_cap(sws, SVGA3D_DEVCAP_FRAGMENT_SHADER_VERSION, &result) &&
 484       result.u >= SVGA3DPSVERSION_30 ? TRUE : FALSE;
 485
 486    svgascreen->use_vs30 =
 487       sws->get_cap(sws, SVGA3D_DEVCAP_VERTEX_SHADER_VERSION, &result) &&
 488       result.u >= SVGA3DVSVERSION_30 ? TRUE : FALSE;
 489
 490    /*
 491     * The D16, D24X8, and D24S8 formats always do an implicit shadow compare
 492     * when sampled from, where as the DF16, DF24, and D24S8_INT do not.  So
 493     * we prefer the later when available.
 494     *
 495     * This mimics hardware vendors extensions for D3D depth sampling. See also
 496     * http://aras-p.info/texts/D3D9GPUHacks.html
 497     */
 498
 499    {
 500       boolean has_df16, has_df24, has_d24s8_int;
 501       SVGA3dSurfaceFormatCaps caps;
 502       SVGA3dSurfaceFormatCaps mask;
 503       mask.value = 0;
 504       mask.zStencil = 1;
 505       mask.texture = 1;
 506
 507       svgascreen->depth.z16 = SVGA3D_Z_D16;
 508       svgascreen->depth.x8z24 = SVGA3D_Z_D24X8;
 509       svgascreen->depth.s8z24 = SVGA3D_Z_D24S8;
 510
 511       svga_get_format_cap(svgascreen, SVGA3D_Z_DF16, &caps);
 512       has_df16 = (caps.value & mask.value) == mask.value;
 513
 514       svga_get_format_cap(svgascreen, SVGA3D_Z_DF24, &caps);
 515       has_df24 = (caps.value & mask.value) == mask.value;
 516
 517       svga_get_format_cap(svgascreen, SVGA3D_Z_D24S8_INT, &caps);
 518       has_d24s8_int = (caps.value & mask.value) == mask.value;
 519
 520       /* XXX: We might want some other logic here.
 521        * Like if we only have d24s8_int we should
 522        * emulate the other formats with that.
 523        */
 524       if (has_df16) {
 525          svgascreen->depth.z16 = SVGA3D_Z_DF16;
 526       }
 527       if (has_df24) {
 528          svgascreen->depth.x8z24 = SVGA3D_Z_DF24;
 529       }
 530       if (has_d24s8_int) {
 531          svgascreen->depth.s8z24 = SVGA3D_Z_D24S8_INT;
 532       }
 533    }
 534
 535 #if 1
 536    /* Shader model 2.0 is unsupported at the moment. */
 537    if(!svgascreen->use_ps30 || !svgascreen->use_vs30)
 538       goto error2;
 539 #else
 540    if(debug_get_bool_option("SVGA_NO_SM30", FALSE))
 541       svgascreen->use_vs30 = svgascreen->use_ps30 = FALSE;
 542 #endif
 543
 544    pipe_mutex_init(svgascreen->tex_mutex);
 545    pipe_mutex_init(svgascreen->swc_mutex);
 546
 547    svga_screen_cache_init(svgascreen);
 548
 549    return screen;
 550 error2:
 551    FREE(svgascreen);
 552 error1:
 553    return NULL;
 554 }
 555
 556 struct svga_winsys_screen *
 557 svga_winsys_screen(struct pipe_screen *screen)
 558 {
 559    return svga_screen(screen)->sws;
 560 }
 561
 562 #ifdef DEBUG
 563 struct svga_screen *
 564 svga_screen(struct pipe_screen *screen)
 565 {
 566    assert(screen);
 567    assert(screen->destroy == svga_destroy_screen);
 568    return (struct svga_screen *)screen;
 569 }
 570 #endif