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_TEXTURE_IMAGE_UNITS:
 126       return 16;
 127    case PIPE_CAP_MAX_COMBINED_SAMPLERS:
 128       return 16;
 129    case PIPE_CAP_NPOT_TEXTURES:
 130       return 1;
 131    case PIPE_CAP_TWO_SIDED_STENCIL:
 132       return 1;
 133    case PIPE_CAP_GLSL:
 134       return svgascreen->use_ps30 && svgascreen->use_vs30;
 135    case PIPE_CAP_ANISOTROPIC_FILTER:
 136       return 1;
 137    case PIPE_CAP_POINT_SPRITE:
 138       return 1;
 139    case PIPE_CAP_MAX_RENDER_TARGETS:
 140       if(!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_RENDER_TARGETS, &result))
 141          return 1;
 142       if(!result.u)
 143          return 1;
 144       return MIN2(result.u, PIPE_MAX_COLOR_BUFS);
 145    case PIPE_CAP_OCCLUSION_QUERY:
 146       return 1;
 147    case PIPE_CAP_TIMER_QUERY:
 148       return 0;
 149    case PIPE_CAP_TEXTURE_SHADOW_MAP:
 150       return 1;
 151    case PIPE_CAP_TEXTURE_SWIZZLE:
 152       return 1;
 153
 154    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 155       {
 156          unsigned levels = SVGA_MAX_TEXTURE_LEVELS;
 157          if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_WIDTH, &result))
 158             levels = MIN2(util_logbase2(result.u) + 1, levels);
 159          else
 160             levels = 12 /* 2048x2048 */;
 161          if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT, &result))
 162             levels = MIN2(util_logbase2(result.u) + 1, levels);
 163          else
 164             levels = 12 /* 2048x2048 */;
 165          return levels;
 166       }
 167
 168    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 169       if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VOLUME_EXTENT, &result))
 170          return 8;  /* max 128x128x128 */
 171       return MIN2(util_logbase2(result.u) + 1, SVGA_MAX_TEXTURE_LEVELS);
 172
 173    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 174       /*
 175        * No mechanism to query the host, and at least limited to 2048x2048 on
 176        * certain hardware.
 177        */
 178       return MIN2(screen->get_paramf(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS),
 179                   12.0 /* 2048x2048 */);
 180
 181    case PIPE_CAP_TEXTURE_MIRROR_REPEAT: /* req. for GL 1.4 */
 182       return 1;
 183
 184    case PIPE_CAP_BLEND_EQUATION_SEPARATE: /* req. for GL 1.5 */
 185       return 1;
 186
 187    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 188    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 189       return 1;
 190    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 191    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 192       return 0;
 193
 194    case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
 195       return 1;
 196    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 197       return 0;
 198
 199    default:
 200       return 0;
 201    }
 202 }
 203
 204
 205 /* This is a fairly pointless interface
 206  */
 207 static int
 208 svga_get_param(struct pipe_screen *screen, enum pipe_cap param)
 209 {
 210    return (int) svga_get_paramf( screen, param );
 211 }
 212
 213 static int svga_get_shader_param(struct pipe_screen *screen, unsigned shader, enum pipe_shader_cap param)
 214 {
 215    struct svga_screen *svgascreen = svga_screen(screen);
 216    struct svga_winsys_screen *sws = svgascreen->sws;
 217    SVGA3dDevCapResult result;
 218
 219    switch (shader)
 220    {
 221    case PIPE_SHADER_FRAGMENT:
 222       switch (param)
 223       {
 224       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 225       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 226       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 227       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 228          return svgascreen->use_ps30 ? 512 : 96;
 229       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 230          return SVGA3D_MAX_NESTING_LEVEL;
 231       case PIPE_SHADER_CAP_MAX_INPUTS:
 232          return 10;
 233       case PIPE_SHADER_CAP_MAX_CONSTS:
 234          return svgascreen->use_ps30 ? 224 : 16;
 235       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 236          return 1;
 237       case PIPE_SHADER_CAP_MAX_TEMPS:
 238          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_FRAGMENT_SHADER_TEMPS, &result))
 239             return svgascreen->use_ps30 ? 32 : 12;
 240          return result.u;
 241       case PIPE_SHADER_CAP_MAX_ADDRS:
 242       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 243          /*
 244           * Although PS 3.0 has some addressing abilities it can only represent
 245           * loops that can be statically determined and unrolled. Given we can
 246           * only handle a subset of the cases that the state tracker already
 247           * does it is better to defer loop unrolling to the state tracker.
 248           */
 249          return 0;
 250       case PIPE_SHADER_CAP_MAX_PREDS:
 251          return svgascreen->use_ps30 ? 1 : 0;
 252       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 253          return 1;
 254       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 255       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 256       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 257          return 0;
 258       case PIPE_SHADER_CAP_SUBROUTINES:
 259          return 0;
 260       case PIPE_SHADER_CAP_INTEGERS:
 261          return 0;
 262       }
 263       break;
 264    case PIPE_SHADER_VERTEX:
 265       switch (param)
 266       {
 267       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 268       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 269          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_INSTRUCTIONS, &result))
 270             return svgascreen->use_vs30 ? 512 : 256;
 271          return result.u;
 272       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 273       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 274          /* XXX: until we have vertex texture support */
 275          return 0;
 276       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 277          return SVGA3D_MAX_NESTING_LEVEL;
 278       case PIPE_SHADER_CAP_MAX_INPUTS:
 279          return 16;
 280       case PIPE_SHADER_CAP_MAX_CONSTS:
 281          return 256;
 282       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 283          return 1;
 284       case PIPE_SHADER_CAP_MAX_TEMPS:
 285          if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VERTEX_SHADER_TEMPS, &result))
 286             return svgascreen->use_vs30 ? 32 : 12;
 287          return result.u;
 288       case PIPE_SHADER_CAP_MAX_ADDRS:
 289          return svgascreen->use_vs30 ? 1 : 0;
 290       case PIPE_SHADER_CAP_MAX_PREDS:
 291          return svgascreen->use_vs30 ? 1 : 0;
 292       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 293          return 1;
 294       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 295       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 296          return svgascreen->use_vs30 ? 1 : 0;
 297       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 298          return 0;
 299       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 300          return 1;
 301       case PIPE_SHADER_CAP_SUBROUTINES:
 302          return 0;
 303       case PIPE_SHADER_CAP_INTEGERS:
 304          return 0;
 305       default:
 306          break;
 307       }
 308       break;
 309    default:
 310       break;
 311    }
 312    return 0;
 313 }
 314
 315
 316 static boolean
 317 svga_is_format_supported( struct pipe_screen *screen,
 318                           enum pipe_format format,
 319                           enum pipe_texture_target target,
 320                           unsigned sample_count,
 321                           unsigned tex_usage)
 322 {
 323    struct svga_screen *ss = svga_screen(screen);
 324    SVGA3dSurfaceFormat svga_format;
 325    SVGA3dSurfaceFormatCaps caps;
 326    SVGA3dSurfaceFormatCaps mask;
 327
 328    assert(tex_usage);
 329
 330    if (sample_count > 1) {
 331       return FALSE;
 332    }
 333
 334    svga_format = svga_translate_format(ss, format, tex_usage);
 335    if (svga_format == SVGA3D_FORMAT_INVALID) {
 336       return FALSE;
 337    }
 338
 339    /*
 340     * Override host capabilities, so that we end up with the same
 341     * visuals for all virtual hardware implementations.
 342     */
 343
 344    if (tex_usage & PIPE_BIND_DISPLAY_TARGET) {
 345       switch (svga_format) {
 346       case SVGA3D_A8R8G8B8:
 347       case SVGA3D_X8R8G8B8:
 348       case SVGA3D_R5G6B5:
 349          break;
 350
 351       /* Often unsupported/problematic. This means we end up with the same
 352        * visuals for all virtual hardware implementations.
 353        */
 354       case PIPE_FORMAT_B4G4R4A4_UNORM:
 355       case PIPE_FORMAT_B5G5R5A1_UNORM:
 356          return FALSE;
 357
 358       default:
 359          return FALSE;
 360       }
 361    }
 362
 363    /*
 364     * Query the host capabilities.
 365     */
 366
 367    svga_get_format_cap(ss, svga_format, &caps);
 368
 369    mask.value = 0;
 370    if (tex_usage & PIPE_BIND_RENDER_TARGET) {
 371       mask.offscreenRenderTarget = 1;
 372    }
 373    if (tex_usage & PIPE_BIND_DEPTH_STENCIL) {
 374       mask.zStencil = 1;
 375    }
 376    if (tex_usage & PIPE_BIND_SAMPLER_VIEW) {
 377       mask.texture = 1;
 378    }
 379
 380    return (caps.value & mask.value) == mask.value;
 381 }
 382
 383
 384 static void
 385 svga_fence_reference(struct pipe_screen *screen,
 386                      struct pipe_fence_handle **ptr,
 387                      struct pipe_fence_handle *fence)
 388 {
 389    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 390    sws->fence_reference(sws, ptr, fence);
 391 }
 392
 393
 394 static boolean
 395 svga_fence_signalled(struct pipe_screen *screen,
 396                      struct pipe_fence_handle *fence)
 397 {
 398    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 399    return sws->fence_signalled(sws, fence, 0) == 0;
 400 }
 401
 402
 403 static boolean
 404 svga_fence_finish(struct pipe_screen *screen,
 405                   struct pipe_fence_handle *fence,
 406                   uint64_t timeout)
 407 {
 408    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
 409
 410    SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "%s fence_ptr %p\n",
 411             __FUNCTION__, fence);
 412
 413    return sws->fence_finish(sws, fence, 0) == 0;
 414 }
 415
 416
 417 static void
 418 svga_destroy_screen( struct pipe_screen *screen )
 419 {
 420    struct svga_screen *svgascreen = svga_screen(screen);
 421
 422    svga_screen_cache_cleanup(svgascreen);
 423
 424    pipe_mutex_destroy(svgascreen->swc_mutex);
 425    pipe_mutex_destroy(svgascreen->tex_mutex);
 426
 427    svgascreen->sws->destroy(svgascreen->sws);
 428
 429    FREE(svgascreen);
 430 }
 431
 432
 433 /**
 434  * Create a new svga_screen object
 435  */
 436 struct pipe_screen *
 437 svga_screen_create(struct svga_winsys_screen *sws)
 438 {
 439    struct svga_screen *svgascreen;
 440    struct pipe_screen *screen;
 441    SVGA3dDevCapResult result;
 442
 443 #ifdef DEBUG
 444    SVGA_DEBUG = debug_get_flags_option("SVGA_DEBUG", svga_debug_flags, 0 );
 445 #endif
 446
 447    svgascreen = CALLOC_STRUCT(svga_screen);
 448    if (!svgascreen)
 449       goto error1;
 450
 451    svgascreen->debug.force_level_surface_view =
 452       debug_get_bool_option("SVGA_FORCE_LEVEL_SURFACE_VIEW", FALSE);
 453    svgascreen->debug.force_surface_view =
 454       debug_get_bool_option("SVGA_FORCE_SURFACE_VIEW", FALSE);
 455    svgascreen->debug.force_sampler_view =
 456       debug_get_bool_option("SVGA_FORCE_SAMPLER_VIEW", FALSE);
 457    svgascreen->debug.no_surface_view =
 458       debug_get_bool_option("SVGA_NO_SURFACE_VIEW", FALSE);
 459    svgascreen->debug.no_sampler_view =
 460       debug_get_bool_option("SVGA_NO_SAMPLER_VIEW", FALSE);
 461
 462    screen = &svgascreen->screen;
 463
 464    screen->destroy = svga_destroy_screen;
 465    screen->get_name = svga_get_name;
 466    screen->get_vendor = svga_get_vendor;
 467    screen->get_param = svga_get_param;
 468    screen->get_shader_param = svga_get_shader_param;
 469    screen->get_paramf = svga_get_paramf;
 470    screen->is_format_supported = svga_is_format_supported;
 471    screen->context_create = svga_context_create;
 472    screen->fence_reference = svga_fence_reference;
 473    screen->fence_signalled = svga_fence_signalled;
 474    screen->fence_finish = svga_fence_finish;
 475    svgascreen->sws = sws;
 476
 477    svga_init_screen_resource_functions(svgascreen);
 478
 479    if (sws->get_hw_version) {
 480       svgascreen->hw_version = sws->get_hw_version(sws);
 481    } else {
 482       svgascreen->hw_version = SVGA3D_HWVERSION_WS65_B1;
 483    }
 484
 485    svgascreen->use_ps30 =
 486       sws->get_cap(sws, SVGA3D_DEVCAP_FRAGMENT_SHADER_VERSION, &result) &&
 487       result.u >= SVGA3DPSVERSION_30 ? TRUE : FALSE;
 488
 489    svgascreen->use_vs30 =
 490       sws->get_cap(sws, SVGA3D_DEVCAP_VERTEX_SHADER_VERSION, &result) &&
 491       result.u >= SVGA3DVSVERSION_30 ? TRUE : FALSE;
 492
 493    /*
 494     * The D16, D24X8, and D24S8 formats always do an implicit shadow compare
 495     * when sampled from, where as the DF16, DF24, and D24S8_INT do not.  So
 496     * we prefer the later when available.
 497     *
 498     * This mimics hardware vendors extensions for D3D depth sampling. See also
 499     * http://aras-p.info/texts/D3D9GPUHacks.html
 500     */
 501
 502    {
 503       boolean has_df16, has_df24, has_d24s8_int;
 504       SVGA3dSurfaceFormatCaps caps;
 505       SVGA3dSurfaceFormatCaps mask;
 506       mask.value = 0;
 507       mask.zStencil = 1;
 508       mask.texture = 1;
 509
 510       svgascreen->depth.z16 = SVGA3D_Z_D16;
 511       svgascreen->depth.x8z24 = SVGA3D_Z_D24X8;
 512       svgascreen->depth.s8z24 = SVGA3D_Z_D24S8;
 513
 514       svga_get_format_cap(svgascreen, SVGA3D_Z_DF16, &caps);
 515       has_df16 = (caps.value & mask.value) == mask.value;
 516
 517       svga_get_format_cap(svgascreen, SVGA3D_Z_DF24, &caps);
 518       has_df24 = (caps.value & mask.value) == mask.value;
 519
 520       svga_get_format_cap(svgascreen, SVGA3D_Z_D24S8_INT, &caps);
 521       has_d24s8_int = (caps.value & mask.value) == mask.value;
 522
 523       /* XXX: We might want some other logic here.
 524        * Like if we only have d24s8_int we should
 525        * emulate the other formats with that.
 526        */
 527       if (has_df16) {
 528          svgascreen->depth.z16 = SVGA3D_Z_DF16;
 529       }
 530       if (has_df24) {
 531          svgascreen->depth.x8z24 = SVGA3D_Z_DF24;
 532       }
 533       if (has_d24s8_int) {
 534          svgascreen->depth.s8z24 = SVGA3D_Z_D24S8_INT;
 535       }
 536    }
 537
 538 #if 1
 539    /* Shader model 2.0 is unsupported at the moment. */
 540    if(!svgascreen->use_ps30 || !svgascreen->use_vs30)
 541       goto error2;
 542 #else
 543    if(debug_get_bool_option("SVGA_NO_SM30", FALSE))
 544       svgascreen->use_vs30 = svgascreen->use_ps30 = FALSE;
 545 #endif
 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