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 15.0;
 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    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 207       return 0;
 208
 209    default:
 210       return 0;
 211    }
 212 }
 213
 214 static int svga_get_shader_param(struct pipe_screen *screen, unsigned shader, enum pipe_shader_cap param)
 215 {
 216    struct svga_screen *svgascreen = svga_screen(screen);
 217    struct svga_winsys_screen *sws = svgascreen->sws;
 218    SVGA3dDevCapResult result;
 219
 220    switch (shader)
 221    {
 222    case PIPE_SHADER_FRAGMENT:
 223       switch (param)
 224       {
 225       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 226       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 227       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 228       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 229          return 512;
 230       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 231          return SVGA3D_MAX_NESTING_LEVEL;
 232       case PIPE_SHADER_CAP_MAX_INPUTS:
 233          return 10;
 234       case PIPE_SHADER_CAP_MAX_CONSTS:
 235          return 224;
 236       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 237          return 1;
 238       case PIPE_SHADER_CAP_MAX_TEMPS:
 239          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_FRAGMENT_SHADER_TEMPS, &result))
 240             return 32;
 241          return MIN2(result.u, SVGA3D_TEMPREG_MAX);
 242       case PIPE_SHADER_CAP_MAX_ADDRS:
 243       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 244          /*
 245           * Although PS 3.0 has some addressing abilities it can only represent
 246           * loops that can be statically determined and unrolled. Given we can
 247           * only handle a subset of the cases that the state tracker already
 248           * does it is better to defer loop unrolling to the state tracker.
 249           */
 250          return 0;
 251       case PIPE_SHADER_CAP_MAX_PREDS:
 252          return 1;
 253       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 254          return 1;
 255       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 256       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 257       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 258          return 0;
 259       case PIPE_SHADER_CAP_SUBROUTINES:
 260          return 0;
 261       case PIPE_SHADER_CAP_INTEGERS:
 262          return 0;
 263       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 264          return 16;
 265       }
 266       break;
 267    case PIPE_SHADER_VERTEX:
 268       switch (param)
 269       {
 270       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 271       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 272          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_INSTRUCTIONS, &result))
 273             return 512;
 274          return result.u;
 275       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 276       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 277          /* XXX: until we have vertex texture support */
 278          return 0;
 279       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 280          return SVGA3D_MAX_NESTING_LEVEL;
 281       case PIPE_SHADER_CAP_MAX_INPUTS:
 282          return 16;
 283       case PIPE_SHADER_CAP_MAX_CONSTS:
 284          return 256;
 285       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 286          return 1;
 287       case PIPE_SHADER_CAP_MAX_TEMPS:
 288          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_TEMPS, &result))
 289             return 32;
 290          return MIN2(result.u, SVGA3D_TEMPREG_MAX);
 291       case PIPE_SHADER_CAP_MAX_ADDRS:
 292          return 1;
 293       case PIPE_SHADER_CAP_MAX_PREDS:
 294          return 1;
 295       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 296          return 1;
 297       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 298       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 299          return 1;
 300       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 301          return 0;
 302       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 303          return 1;
 304       case PIPE_SHADER_CAP_SUBROUTINES:
 305          return 0;
 306       case PIPE_SHADER_CAP_INTEGERS:
 307          return 0;
 308       default:
 309          break;
 310       }
 311       break;
 312    default:
 313       break;
 314    }
 315    return 0;
 316 }
 317
 318
 319 static boolean
 320 svga_is_format_supported( struct pipe_screen *screen,
 321                           enum pipe_format format,
 322                           enum pipe_texture_target target,
 323                           unsigned sample_count,
 324                           unsigned tex_usage)
 325 {
 326    struct svga_screen *ss = svga_screen(screen);
 327    SVGA3dSurfaceFormat svga_format;
 328    SVGA3dSurfaceFormatCaps caps;
 329    SVGA3dSurfaceFormatCaps mask;
 330
 331    assert(tex_usage);
 332
 333    if (sample_count > 1) {
 334       return FALSE;
 335    }
 336
 337    svga_format = svga_translate_format(ss, format, tex_usage);
 338    if (svga_format == SVGA3D_FORMAT_INVALID) {
 339       return FALSE;
 340    }
 341
 342    /*
 343     * Override host capabilities, so that we end up with the same
 344     * visuals for all virtual hardware implementations.
 345     */
 346
 347    if (tex_usage & PIPE_BIND_DISPLAY_TARGET) {
 348       switch (svga_format) {
 349       case SVGA3D_A8R8G8B8:
 350       case SVGA3D_X8R8G8B8:
 351       case SVGA3D_R5G6B5:
 352          break;
 353
 354       /* Often unsupported/problematic. This means we end up with the same
 355        * visuals for all virtual hardware implementations.
 356        */
 357       case SVGA3D_A4R4G4B4:
 358       case SVGA3D_A1R5G5B5:
 359          return FALSE;
 360
 361       default:
 362          return FALSE;
 363       }
 364    }
 365
 366    /*
 367     * Query the host capabilities.
 368     */
 369
 370    svga_get_format_cap(ss, svga_format, &caps);
 371
 372    mask.value = 0;
 373    if (tex_usage & PIPE_BIND_RENDER_TARGET) {
 374       mask.offscreenRenderTarget = 1;
 375    }
 376    if (tex_usage & PIPE_BIND_DEPTH_STENCIL) {
 377       mask.zStencil = 1;
 378    }
 379    if (tex_usage & PIPE_BIND_SAMPLER_VIEW) {
 380       mask.texture = 1;
 381    }
 382
 383    return (caps.value & mask.value) == mask.value;
 384 }
 385
 386
 387 static void
 388 svga_fence_reference(struct pipe_screen *screen,
 389                      struct pipe_fence_handle **ptr,
 390                      struct pipe_fence_handle *fence)
 391 {
 392    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 393    sws->fence_reference(sws, ptr, fence);
 394 }
 395
 396
 397 static boolean
 398 svga_fence_signalled(struct pipe_screen *screen,
 399                      struct pipe_fence_handle *fence)
 400 {
 401    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 402    return sws->fence_signalled(sws, fence, 0) == 0;
 403 }
 404
 405
 406 static boolean
 407 svga_fence_finish(struct pipe_screen *screen,
 408                   struct pipe_fence_handle *fence,
 409                   uint64_t timeout)
 410 {
 411    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 412
 413    SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "%s fence_ptr %p\n",
 414             __FUNCTION__, fence);
 415
 416    return sws->fence_finish(sws, fence, 0) == 0;
 417 }
 418
 419
 420 static void
 421 svga_destroy_screen( struct pipe_screen *screen )
 422 {
 423    struct svga_screen *svgascreen = svga_screen(screen);
 424
 425    svga_screen_cache_cleanup(svgascreen);
 426
 427    pipe_mutex_destroy(svgascreen->swc_mutex);
 428    pipe_mutex_destroy(svgascreen->tex_mutex);
 429
 430    svgascreen->sws->destroy(svgascreen->sws);
 431
 432    FREE(svgascreen);
 433 }
 434
 435
 436 /**
 437  * Create a new svga_screen object
 438  */
 439 struct pipe_screen *
 440 svga_screen_create(struct svga_winsys_screen *sws)
 441 {
 442    struct svga_screen *svgascreen;
 443    struct pipe_screen *screen;
 444    SVGA3dDevCapResult result;
 445    boolean use_vs30, use_ps30;
 446
 447 #ifdef DEBUG
 448    SVGA_DEBUG = debug_get_flags_option("SVGA_DEBUG", svga_debug_flags, 0 );
 449 #endif
 450
 451    svgascreen = CALLOC_STRUCT(svga_screen);
 452    if (!svgascreen)
 453       goto error1;
 454
 455    svgascreen->debug.force_level_surface_view =
 456       debug_get_bool_option("SVGA_FORCE_LEVEL_SURFACE_VIEW", FALSE);
 457    svgascreen->debug.force_surface_view =
 458       debug_get_bool_option("SVGA_FORCE_SURFACE_VIEW", FALSE);
 459    svgascreen->debug.force_sampler_view =
 460       debug_get_bool_option("SVGA_FORCE_SAMPLER_VIEW", FALSE);
 461    svgascreen->debug.no_surface_view =
 462       debug_get_bool_option("SVGA_NO_SURFACE_VIEW", FALSE);
 463    svgascreen->debug.no_sampler_view =
 464       debug_get_bool_option("SVGA_NO_SAMPLER_VIEW", FALSE);
 465
 466    screen = &svgascreen->screen;
 467
 468    screen->destroy = svga_destroy_screen;
 469    screen->get_name = svga_get_name;
 470    screen->get_vendor = svga_get_vendor;
 471    screen->get_param = svga_get_param;
 472    screen->get_shader_param = svga_get_shader_param;
 473    screen->get_paramf = svga_get_paramf;
 474    screen->is_format_supported = svga_is_format_supported;
 475    screen->context_create = svga_context_create;
 476    screen->fence_reference = svga_fence_reference;
 477    screen->fence_signalled = svga_fence_signalled;
 478    screen->fence_finish = svga_fence_finish;
 479    svgascreen->sws = sws;
 480
 481    svga_init_screen_resource_functions(svgascreen);
 482
 483    if (sws->get_hw_version) {
 484       svgascreen->hw_version = sws->get_hw_version(sws);
 485    } else {
 486       svgascreen->hw_version = SVGA3D_HWVERSION_WS65_B1;
 487    }
 488
 489    use_ps30 =
 490       sws->get_cap(sws, SVGA3D_DEVCAP_FRAGMENT_SHADER_VERSION, &result) &&
 491       result.u >= SVGA3DPSVERSION_30 ? TRUE : FALSE;
 492
 493    use_vs30 =
 494       sws->get_cap(sws, SVGA3D_DEVCAP_VERTEX_SHADER_VERSION, &result) &&
 495       result.u >= SVGA3DVSVERSION_30 ? TRUE : FALSE;
 496
 497    /* we require Shader model 3.0 or later */
 498    if (!use_ps30 || !use_vs30)
 499       goto error2;
 500
 501    /*
 502     * The D16, D24X8, and D24S8 formats always do an implicit shadow compare
 503     * when sampled from, where as the DF16, DF24, and D24S8_INT do not.  So
 504     * we prefer the later when available.
 505     *
 506     * This mimics hardware vendors extensions for D3D depth sampling. See also
 507     * http://aras-p.info/texts/D3D9GPUHacks.html
 508     */
 509
 510    {
 511       boolean has_df16, has_df24, has_d24s8_int;
 512       SVGA3dSurfaceFormatCaps caps;
 513       SVGA3dSurfaceFormatCaps mask;
 514       mask.value = 0;
 515       mask.zStencil = 1;
 516       mask.texture = 1;
 517
 518       svgascreen->depth.z16 = SVGA3D_Z_D16;
 519       svgascreen->depth.x8z24 = SVGA3D_Z_D24X8;
 520       svgascreen->depth.s8z24 = SVGA3D_Z_D24S8;
 521
 522       svga_get_format_cap(svgascreen, SVGA3D_Z_DF16, &caps);
 523       has_df16 = (caps.value & mask.value) == mask.value;
 524
 525       svga_get_format_cap(svgascreen, SVGA3D_Z_DF24, &caps);
 526       has_df24 = (caps.value & mask.value) == mask.value;
 527
 528       svga_get_format_cap(svgascreen, SVGA3D_Z_D24S8_INT, &caps);
 529       has_d24s8_int = (caps.value & mask.value) == mask.value;
 530
 531       /* XXX: We might want some other logic here.
 532        * Like if we only have d24s8_int we should
 533        * emulate the other formats with that.
 534        */
 535       if (has_df16) {
 536          svgascreen->depth.z16 = SVGA3D_Z_DF16;
 537       }
 538       if (has_df24) {
 539          svgascreen->depth.x8z24 = SVGA3D_Z_DF24;
 540       }
 541       if (has_d24s8_int) {
 542          svgascreen->depth.s8z24 = SVGA3D_Z_D24S8_INT;
 543       }
 544    }
 545
 546    pipe_mutex_init(svgascreen->tex_mutex);
 547    pipe_mutex_init(svgascreen->swc_mutex);
 548
 549    svga_screen_cache_init(svgascreen);
 550
 551    return screen;
 552 error2:
 553    FREE(svgascreen);
 554 error1:
 555    return NULL;
 556 }
 557
 558 struct svga_winsys_screen *
 559 svga_winsys_screen(struct pipe_screen *screen)
 560 {
 561    return svga_screen(screen)->sws;
 562 }
 563
 564 #ifdef DEBUG
 565 struct svga_screen *
 566 svga_screen(struct pipe_screen *screen)
 567 {
 568    assert(screen);
 569    assert(screen->destroy == svga_destroy_screen);
 570    return (struct svga_screen *)screen;
 571 }
 572 #endif