src/gallium/drivers/nouveau/nv30/nv30_screen.c

   1 /*
   2  * Copyright 2012 Red Hat Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20  * OTHER DEALINGS IN THE SOFTWARE.
  21  *
  22  * Authors: Ben Skeggs
  23  *
  24  */
  25
  26 #include <xf86drm.h>
  27 #include <nouveau_drm.h>
  28 #include "util/u_format.h"
  29 #include "util/u_format_s3tc.h"
  30
  31 #include "nv_object.xml.h"
  32 #include "nv_m2mf.xml.h"
  33 #include "nv30/nv30-40_3d.xml.h"
  34 #include "nv30/nv01_2d.xml.h"
  35
  36 #include "nouveau_fence.h"
  37 #include "nv30/nv30_screen.h"
  38 #include "nv30/nv30_context.h"
  39 #include "nv30/nv30_resource.h"
  40 #include "nv30/nv30_format.h"
  41
  42 #define RANKINE_0397_CHIPSET 0x00000003
  43 #define RANKINE_0497_CHIPSET 0x000001e0
  44 #define RANKINE_0697_CHIPSET 0x00000010
  45 #define CURIE_4097_CHIPSET   0x00000baf
  46 #define CURIE_4497_CHIPSET   0x00005450
  47 #define CURIE_4497_CHIPSET6X 0x00000088
  48
  49 static int
  50 nv30_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
  51 {
  52    struct nv30_screen *screen = nv30_screen(pscreen);
  53    struct nouveau_object *eng3d = screen->eng3d;
  54    struct nouveau_device *dev = nouveau_screen(pscreen)->device;
  55
  56    switch (param) {
  57    /* non-boolean capabilities */
  58    case PIPE_CAP_MAX_RENDER_TARGETS:
  59       return (eng3d->oclass >= NV40_3D_CLASS) ? 4 : 1;
  60    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
  61       return 13;
  62    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
  63       return 10;
  64    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
  65       return 13;
  66    case PIPE_CAP_GLSL_FEATURE_LEVEL:
  67       return 120;
  68    case PIPE_CAP_ENDIANNESS:
  69       return PIPE_ENDIAN_LITTLE;
  70    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
  71       return 16;
  72    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
  73       return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
  74    case PIPE_CAP_MAX_VIEWPORTS:
  75       return 1;
  76    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
  77       return 2048;
  78    /* supported capabilities */
  79    case PIPE_CAP_TWO_SIDED_STENCIL:
  80    case PIPE_CAP_ANISOTROPIC_FILTER:
  81    case PIPE_CAP_POINT_SPRITE:
  82    case PIPE_CAP_OCCLUSION_QUERY:
  83    case PIPE_CAP_QUERY_TIME_ELAPSED:
  84    case PIPE_CAP_QUERY_TIMESTAMP:
  85    case PIPE_CAP_TEXTURE_SHADOW_MAP:
  86    case PIPE_CAP_TEXTURE_SWIZZLE:
  87    case PIPE_CAP_DEPTH_CLIP_DISABLE:
  88    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
  89    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
  90    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
  91    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
  92    case PIPE_CAP_TGSI_TEXCOORD:
  93    case PIPE_CAP_USER_CONSTANT_BUFFERS:
  94    case PIPE_CAP_USER_INDEX_BUFFERS:
  95    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
  96    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
  97    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
  98    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
  99    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 100       return 1;
 101    /* nv35 capabilities */
 102    case PIPE_CAP_DEPTH_BOUNDS_TEST:
 103       return eng3d->oclass == NV35_3D_CLASS || eng3d->oclass >= NV40_3D_CLASS;
 104    /* nv4x capabilities */
 105    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 106    case PIPE_CAP_NPOT_TEXTURES:
 107    case PIPE_CAP_CONDITIONAL_RENDER:
 108    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 109    case PIPE_CAP_PRIMITIVE_RESTART:
 110       return (eng3d->oclass >= NV40_3D_CLASS) ? 1 : 0;
 111    /* unsupported */
 112    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 113    case PIPE_CAP_SM3:
 114    case PIPE_CAP_INDEP_BLEND_ENABLE:
 115    case PIPE_CAP_INDEP_BLEND_FUNC:
 116    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 117    case PIPE_CAP_SHADER_STENCIL_EXPORT:
 118    case PIPE_CAP_TGSI_INSTANCEID:
 119    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR: /* XXX: yes? */
 120    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 121    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 122    case PIPE_CAP_MIN_TEXEL_OFFSET:
 123    case PIPE_CAP_MAX_TEXEL_OFFSET:
 124    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 125    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 126    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 127    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 128    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 129    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 130    case PIPE_CAP_MAX_VERTEX_STREAMS:
 131    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 132    case PIPE_CAP_TEXTURE_BARRIER:
 133    case PIPE_CAP_SEAMLESS_CUBE_MAP:
 134    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 135    case PIPE_CAP_CUBE_MAP_ARRAY:
 136    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 137    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 138    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 139    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 140    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 141    case PIPE_CAP_START_INSTANCE:
 142    case PIPE_CAP_TEXTURE_MULTISAMPLE:
 143    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 144    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 145    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 146    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 147    case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 148    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 149    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 150    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 151    case PIPE_CAP_TEXTURE_GATHER_SM5:
 152    case PIPE_CAP_FAKE_SW_MSAA:
 153    case PIPE_CAP_TEXTURE_QUERY_LOD:
 154    case PIPE_CAP_SAMPLE_SHADING:
 155    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 156    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 157    case PIPE_CAP_USER_VERTEX_BUFFERS:
 158    case PIPE_CAP_COMPUTE:
 159    case PIPE_CAP_DRAW_INDIRECT:
 160    case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 161    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 162    case PIPE_CAP_SAMPLER_VIEW_TARGET:
 163    case PIPE_CAP_CLIP_HALFZ:
 164    case PIPE_CAP_VERTEXID_NOBASE:
 165    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 166    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 167    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 168    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 169    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 170    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 171    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 172    case PIPE_CAP_TGSI_TXQS:
 173    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 174    case PIPE_CAP_SHAREABLE_SHADERS:
 175    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 176       return 0;
 177
 178    case PIPE_CAP_VENDOR_ID:
 179       return 0x10de;
 180    case PIPE_CAP_DEVICE_ID: {
 181       uint64_t device_id;
 182       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
 183          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
 184          return -1;
 185       }
 186       return device_id;
 187    }
 188    case PIPE_CAP_ACCELERATED:
 189       return 1;
 190    case PIPE_CAP_VIDEO_MEMORY:
 191       return dev->vram_size >> 20;
 192    case PIPE_CAP_UMA:
 193       return 0;
 194    }
 195
 196    debug_printf("unknown param %d\n", param);
 197    return 0;
 198 }
 199
 200 static float
 201 nv30_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
 202 {
 203    struct nv30_screen *screen = nv30_screen(pscreen);
 204    struct nouveau_object *eng3d = screen->eng3d;
 205
 206    switch (param) {
 207    case PIPE_CAPF_MAX_LINE_WIDTH:
 208    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 209       return 10.0;
 210    case PIPE_CAPF_MAX_POINT_WIDTH:
 211    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 212       return 64.0;
 213    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 214       return (eng3d->oclass >= NV40_3D_CLASS) ? 16.0 : 8.0;
 215    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 216       return 15.0;
 217    default:
 218       debug_printf("unknown paramf %d\n", param);
 219       return 0;
 220    }
 221 }
 222
 223 static int
 224 nv30_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
 225                              enum pipe_shader_cap param)
 226 {
 227    struct nv30_screen *screen = nv30_screen(pscreen);
 228    struct nouveau_object *eng3d = screen->eng3d;
 229
 230    switch (shader) {
 231    case PIPE_SHADER_VERTEX:
 232       switch (param) {
 233       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 234       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 235          return (eng3d->oclass >= NV40_3D_CLASS) ? 512 : 256;
 236       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 237       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 238          return (eng3d->oclass >= NV40_3D_CLASS) ? 512 : 0;
 239       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 240          return 0;
 241       case PIPE_SHADER_CAP_MAX_INPUTS:
 242       case PIPE_SHADER_CAP_MAX_OUTPUTS:
 243          return 16;
 244       case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 245          return ((eng3d->oclass >= NV40_3D_CLASS) ? (468 - 6): (256 - 6)) * sizeof(float[4]);
 246       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 247          return 1;
 248       case PIPE_SHADER_CAP_MAX_TEMPS:
 249          return (eng3d->oclass >= NV40_3D_CLASS) ? 32 : 13;
 250       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 251       case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 252          return 0;
 253       case PIPE_SHADER_CAP_MAX_PREDS:
 254       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 255       case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 256       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 257       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 258       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 259       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 260       case PIPE_SHADER_CAP_SUBROUTINES:
 261       case PIPE_SHADER_CAP_INTEGERS:
 262       case PIPE_SHADER_CAP_DOUBLES:
 263       case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 264       case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 265       case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 266       case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 267          return 0;
 268       case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 269          return 32;
 270       default:
 271          debug_printf("unknown vertex shader param %d\n", param);
 272          return 0;
 273       }
 274       break;
 275    case PIPE_SHADER_FRAGMENT:
 276       switch (param) {
 277       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 278       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 279       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 280       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 281          return 4096;
 282       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 283          return 0;
 284       case PIPE_SHADER_CAP_MAX_INPUTS:
 285          return 8; /* should be possible to do 10 with nv4x */
 286       case PIPE_SHADER_CAP_MAX_OUTPUTS:
 287          return 4;
 288       case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 289          return ((eng3d->oclass >= NV40_3D_CLASS) ? 224 : 32) * sizeof(float[4]);
 290       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 291          return 1;
 292       case PIPE_SHADER_CAP_MAX_TEMPS:
 293          return 32;
 294       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 295       case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 296          return 16;
 297       case PIPE_SHADER_CAP_MAX_PREDS:
 298       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 299       case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 300       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 301       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 302       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 303       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 304       case PIPE_SHADER_CAP_SUBROUTINES:
 305       case PIPE_SHADER_CAP_DOUBLES:
 306       case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 307       case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 308       case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 309       case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 310          return 0;
 311       case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 312          return 32;
 313       default:
 314          debug_printf("unknown fragment shader param %d\n", param);
 315          return 0;
 316       }
 317       break;
 318    default:
 319       return 0;
 320    }
 321 }
 322
 323 static boolean
 324 nv30_screen_is_format_supported(struct pipe_screen *pscreen,
 325                                 enum pipe_format format,
 326                                 enum pipe_texture_target target,
 327                                 unsigned sample_count,
 328                                 unsigned bindings)
 329 {
 330    if (sample_count > nv30_screen(pscreen)->max_sample_count)
 331       return false;
 332
 333    if (!(0x00000017 & (1 << sample_count)))
 334       return false;
 335
 336    if (!util_format_is_supported(format, bindings)) {
 337       return false;
 338    }
 339
 340    /* transfers & shared are always supported */
 341    bindings &= ~(PIPE_BIND_TRANSFER_READ |
 342                  PIPE_BIND_TRANSFER_WRITE |
 343                  PIPE_BIND_SHARED);
 344
 345    return (nv30_format_info(pscreen, format)->bindings & bindings) == bindings;
 346 }
 347
 348 static void
 349 nv30_screen_fence_emit(struct pipe_screen *pscreen, uint32_t *sequence)
 350 {
 351    struct nv30_screen *screen = nv30_screen(pscreen);
 352    struct nouveau_pushbuf *push = screen->base.pushbuf;
 353
 354    *sequence = ++screen->base.fence.sequence;
 355
 356    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 3);
 357    PUSH_DATA (push, NV30_3D_FENCE_OFFSET |
 358               (2 /* size */ << 18) | (7 /* subchan */ << 13));
 359    PUSH_DATA (push, 0);
 360    PUSH_DATA (push, *sequence);
 361 }
 362
 363 static uint32_t
 364 nv30_screen_fence_update(struct pipe_screen *pscreen)
 365 {
 366    struct nv30_screen *screen = nv30_screen(pscreen);
 367    struct nv04_notify *fence = screen->fence->data;
 368    return *(uint32_t *)((char *)screen->notify->map + fence->offset);
 369 }
 370
 371 static void
 372 nv30_screen_destroy(struct pipe_screen *pscreen)
 373 {
 374    struct nv30_screen *screen = nv30_screen(pscreen);
 375
 376    if (!nouveau_drm_screen_unref(&screen->base))
 377       return;
 378
 379    if (screen->base.fence.current) {
 380       struct nouveau_fence *current = NULL;
 381
 382       /* nouveau_fence_wait will create a new current fence, so wait on the
 383        * _current_ one, and remove both.
 384        */
 385       nouveau_fence_ref(screen->base.fence.current, &current);
 386       nouveau_fence_wait(current, NULL);
 387       nouveau_fence_ref(NULL, &current);
 388       nouveau_fence_ref(NULL, &screen->base.fence.current);
 389    }
 390
 391    nouveau_bo_ref(NULL, &screen->notify);
 392
 393    nouveau_heap_destroy(&screen->query_heap);
 394    nouveau_heap_destroy(&screen->vp_exec_heap);
 395    nouveau_heap_destroy(&screen->vp_data_heap);
 396
 397    nouveau_object_del(&screen->query);
 398    nouveau_object_del(&screen->fence);
 399    nouveau_object_del(&screen->ntfy);
 400
 401    nouveau_object_del(&screen->sifm);
 402    nouveau_object_del(&screen->swzsurf);
 403    nouveau_object_del(&screen->surf2d);
 404    nouveau_object_del(&screen->m2mf);
 405    nouveau_object_del(&screen->eng3d);
 406    nouveau_object_del(&screen->null);
 407
 408    nouveau_screen_fini(&screen->base);
 409    FREE(screen);
 410 }
 411
 412 #define FAIL_SCREEN_INIT(str, err)                    \
 413    do {                                               \
 414       NOUVEAU_ERR(str, err);                          \
 415       nv30_screen_destroy(pscreen);                   \
 416       return NULL;                                    \
 417    } while(0)
 418
 419 struct pipe_screen *
 420 nv30_screen_create(struct nouveau_device *dev)
 421 {
 422    struct nv30_screen *screen = CALLOC_STRUCT(nv30_screen);
 423    struct pipe_screen *pscreen;
 424    struct nouveau_pushbuf *push;
 425    struct nv04_fifo *fifo;
 426    unsigned oclass = 0;
 427    int ret, i;
 428
 429    if (!screen)
 430       return NULL;
 431
 432    switch (dev->chipset & 0xf0) {
 433    case 0x30:
 434       if (RANKINE_0397_CHIPSET & (1 << (dev->chipset & 0x0f)))
 435          oclass = NV30_3D_CLASS;
 436       else
 437       if (RANKINE_0697_CHIPSET & (1 << (dev->chipset & 0x0f)))
 438          oclass = NV34_3D_CLASS;
 439       else
 440       if (RANKINE_0497_CHIPSET & (1 << (dev->chipset & 0x0f)))
 441          oclass = NV35_3D_CLASS;
 442       break;
 443    case 0x40:
 444       if (CURIE_4097_CHIPSET & (1 << (dev->chipset & 0x0f)))
 445          oclass = NV40_3D_CLASS;
 446       else
 447       if (CURIE_4497_CHIPSET & (1 << (dev->chipset & 0x0f)))
 448          oclass = NV44_3D_CLASS;
 449       break;
 450    case 0x60:
 451       if (CURIE_4497_CHIPSET6X & (1 << (dev->chipset & 0x0f)))
 452          oclass = NV44_3D_CLASS;
 453       break;
 454    default:
 455       break;
 456    }
 457
 458    if (!oclass) {
 459       NOUVEAU_ERR("unknown 3d class for 0x%02x\n", dev->chipset);
 460       FREE(screen);
 461       return NULL;
 462    }
 463
 464    /*
 465     * Some modern apps try to use msaa without keeping in mind the
 466     * restrictions on videomem of older cards. Resulting in dmesg saying:
 467     * [ 1197.850642] nouveau E[soffice.bin[3785]] fail ttm_validate
 468     * [ 1197.850648] nouveau E[soffice.bin[3785]] validating bo list
 469     * [ 1197.850654] nouveau E[soffice.bin[3785]] validate: -12
 470     *
 471     * Because we are running out of video memory, after which the program
 472     * using the msaa visual freezes, and eventually the entire system freezes.
 473     *
 474     * To work around this we do not allow msaa visauls by default and allow
 475     * the user to override this via NV30_MAX_MSAA.
 476     */
 477    screen->max_sample_count = debug_get_num_option("NV30_MAX_MSAA", 0);
 478    if (screen->max_sample_count > 4)
 479       screen->max_sample_count = 4;
 480
 481    pscreen = &screen->base.base;
 482    pscreen->destroy = nv30_screen_destroy;
 483    pscreen->get_param = nv30_screen_get_param;
 484    pscreen->get_paramf = nv30_screen_get_paramf;
 485    pscreen->get_shader_param = nv30_screen_get_shader_param;
 486    pscreen->context_create = nv30_context_create;
 487    pscreen->is_format_supported = nv30_screen_is_format_supported;
 488    nv30_resource_screen_init(pscreen);
 489    nouveau_screen_init_vdec(&screen->base);
 490
 491    screen->base.fence.emit = nv30_screen_fence_emit;
 492    screen->base.fence.update = nv30_screen_fence_update;
 493
 494    ret = nouveau_screen_init(&screen->base, dev);
 495    if (ret)
 496       FAIL_SCREEN_INIT("nv30_screen_init failed: %d\n", ret);
 497
 498    screen->base.vidmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
 499    screen->base.sysmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
 500    if (oclass == NV40_3D_CLASS) {
 501       screen->base.vidmem_bindings |= PIPE_BIND_INDEX_BUFFER;
 502       screen->base.sysmem_bindings |= PIPE_BIND_INDEX_BUFFER;
 503    }
 504
 505    fifo = screen->base.channel->data;
 506    push = screen->base.pushbuf;
 507    push->rsvd_kick = 16;
 508
 509    ret = nouveau_object_new(screen->base.channel, 0x00000000, NV01_NULL_CLASS,
 510                             NULL, 0, &screen->null);
 511    if (ret)
 512       FAIL_SCREEN_INIT("error allocating null object: %d\n", ret);
 513
 514    /* DMA_FENCE refuses to accept DMA objects with "adjust" filled in,
 515     * this means that the address pointed at by the DMA object must
 516     * be 4KiB aligned, which means this object needs to be the first
 517     * one allocated on the channel.
 518     */
 519    ret = nouveau_object_new(screen->base.channel, 0xbeef1e00,
 520                             NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
 521                             .length = 32 }, sizeof(struct nv04_notify),
 522                             &screen->fence);
 523    if (ret)
 524       FAIL_SCREEN_INIT("error allocating fence notifier: %d\n", ret);
 525
 526    /* DMA_NOTIFY object, we don't actually use this but M2MF fails without */
 527    ret = nouveau_object_new(screen->base.channel, 0xbeef0301,
 528                             NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
 529                             .length = 32 }, sizeof(struct nv04_notify),
 530                             &screen->ntfy);
 531    if (ret)
 532       FAIL_SCREEN_INIT("error allocating sync notifier: %d\n", ret);
 533
 534    /* DMA_QUERY, used to implement occlusion queries, we attempt to allocate
 535     * the remainder of the "notifier block" assigned by the kernel for
 536     * use as query objects
 537     */
 538    ret = nouveau_object_new(screen->base.channel, 0xbeef0351,
 539                             NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
 540                             .length = 4096 - 128 }, sizeof(struct nv04_notify),
 541                             &screen->query);
 542    if (ret)
 543       FAIL_SCREEN_INIT("error allocating query notifier: %d\n", ret);
 544
 545    ret = nouveau_heap_init(&screen->query_heap, 0, 4096 - 128);
 546    if (ret)
 547       FAIL_SCREEN_INIT("error creating query heap: %d\n", ret);
 548
 549    LIST_INITHEAD(&screen->queries);
 550
 551    /* Vertex program resources (code/data), currently 6 of the constant
 552     * slots are reserved to implement user clipping planes
 553     */
 554    if (oclass < NV40_3D_CLASS) {
 555       nouveau_heap_init(&screen->vp_exec_heap, 0, 256);
 556       nouveau_heap_init(&screen->vp_data_heap, 6, 256 - 6);
 557    } else {
 558       nouveau_heap_init(&screen->vp_exec_heap, 0, 512);
 559       nouveau_heap_init(&screen->vp_data_heap, 6, 468 - 6);
 560    }
 561
 562    ret = nouveau_bo_wrap(screen->base.device, fifo->notify, &screen->notify);
 563    if (ret == 0)
 564       ret = nouveau_bo_map(screen->notify, 0, screen->base.client);
 565    if (ret)
 566       FAIL_SCREEN_INIT("error mapping notifier memory: %d\n", ret);
 567
 568    ret = nouveau_object_new(screen->base.channel, 0xbeef3097, oclass,
 569                             NULL, 0, &screen->eng3d);
 570    if (ret)
 571       FAIL_SCREEN_INIT("error allocating 3d object: %d\n", ret);
 572
 573    BEGIN_NV04(push, NV01_SUBC(3D, OBJECT), 1);
 574    PUSH_DATA (push, screen->eng3d->handle);
 575    BEGIN_NV04(push, NV30_3D(DMA_NOTIFY), 13);
 576    PUSH_DATA (push, screen->ntfy->handle);
 577    PUSH_DATA (push, fifo->vram);     /* TEXTURE0 */
 578    PUSH_DATA (push, fifo->gart);     /* TEXTURE1 */
 579    PUSH_DATA (push, fifo->vram);     /* COLOR1 */
 580    PUSH_DATA (push, screen->null->handle);  /* UNK190 */
 581    PUSH_DATA (push, fifo->vram);     /* COLOR0 */
 582    PUSH_DATA (push, fifo->vram);     /* ZETA */
 583    PUSH_DATA (push, fifo->vram);     /* VTXBUF0 */
 584    PUSH_DATA (push, fifo->gart);     /* VTXBUF1 */
 585    PUSH_DATA (push, screen->fence->handle);  /* FENCE */
 586    PUSH_DATA (push, screen->query->handle);  /* QUERY - intr 0x80 if nullobj */
 587    PUSH_DATA (push, screen->null->handle);  /* UNK1AC */
 588    PUSH_DATA (push, screen->null->handle);  /* UNK1B0 */
 589    if (screen->eng3d->oclass < NV40_3D_CLASS) {
 590       BEGIN_NV04(push, SUBC_3D(0x03b0), 1);
 591       PUSH_DATA (push, 0x00100000);
 592       BEGIN_NV04(push, SUBC_3D(0x1d80), 1);
 593       PUSH_DATA (push, 3);
 594
 595       BEGIN_NV04(push, SUBC_3D(0x1e98), 1);
 596       PUSH_DATA (push, 0);
 597       BEGIN_NV04(push, SUBC_3D(0x17e0), 3);
 598       PUSH_DATA (push, fui(0.0));
 599       PUSH_DATA (push, fui(0.0));
 600       PUSH_DATA (push, fui(1.0));
 601       BEGIN_NV04(push, SUBC_3D(0x1f80), 16);
 602       for (i = 0; i < 16; i++)
 603          PUSH_DATA (push, (i == 8) ? 0x0000ffff : 0);
 604
 605       BEGIN_NV04(push, NV30_3D(RC_ENABLE), 1);
 606       PUSH_DATA (push, 0);
 607    } else {
 608       BEGIN_NV04(push, NV40_3D(DMA_COLOR2), 2);
 609       PUSH_DATA (push, fifo->vram);
 610       PUSH_DATA (push, fifo->vram);  /* COLOR3 */
 611
 612       BEGIN_NV04(push, SUBC_3D(0x1450), 1);
 613       PUSH_DATA (push, 0x00000004);
 614
 615       BEGIN_NV04(push, SUBC_3D(0x1ea4), 3); /* ZCULL */
 616       PUSH_DATA (push, 0x00000010);
 617       PUSH_DATA (push, 0x01000100);
 618       PUSH_DATA (push, 0xff800006);
 619
 620       /* vtxprog output routing */
 621       BEGIN_NV04(push, SUBC_3D(0x1fc4), 1);
 622       PUSH_DATA (push, 0x06144321);
 623       BEGIN_NV04(push, SUBC_3D(0x1fc8), 2);
 624       PUSH_DATA (push, 0xedcba987);
 625       PUSH_DATA (push, 0x0000006f);
 626       BEGIN_NV04(push, SUBC_3D(0x1fd0), 1);
 627       PUSH_DATA (push, 0x00171615);
 628       BEGIN_NV04(push, SUBC_3D(0x1fd4), 1);
 629       PUSH_DATA (push, 0x001b1a19);
 630
 631       BEGIN_NV04(push, SUBC_3D(0x1ef8), 1);
 632       PUSH_DATA (push, 0x0020ffff);
 633       BEGIN_NV04(push, SUBC_3D(0x1d64), 1);
 634       PUSH_DATA (push, 0x01d300d4);
 635
 636       BEGIN_NV04(push, NV40_3D(MIPMAP_ROUNDING), 1);
 637       PUSH_DATA (push, NV40_3D_MIPMAP_ROUNDING_MODE_DOWN);
 638    }
 639
 640    ret = nouveau_object_new(screen->base.channel, 0xbeef3901, NV03_M2MF_CLASS,
 641                             NULL, 0, &screen->m2mf);
 642    if (ret)
 643       FAIL_SCREEN_INIT("error allocating m2mf object: %d\n", ret);
 644
 645    BEGIN_NV04(push, NV01_SUBC(M2MF, OBJECT), 1);
 646    PUSH_DATA (push, screen->m2mf->handle);
 647    BEGIN_NV04(push, NV03_M2MF(DMA_NOTIFY), 1);
 648    PUSH_DATA (push, screen->ntfy->handle);
 649
 650    ret = nouveau_object_new(screen->base.channel, 0xbeef6201,
 651                             NV10_SURFACE_2D_CLASS, NULL, 0, &screen->surf2d);
 652    if (ret)
 653       FAIL_SCREEN_INIT("error allocating surf2d object: %d\n", ret);
 654
 655    BEGIN_NV04(push, NV01_SUBC(SF2D, OBJECT), 1);
 656    PUSH_DATA (push, screen->surf2d->handle);
 657    BEGIN_NV04(push, NV04_SF2D(DMA_NOTIFY), 1);
 658    PUSH_DATA (push, screen->ntfy->handle);
 659
 660    if (dev->chipset < 0x40)
 661       oclass = NV30_SURFACE_SWZ_CLASS;
 662    else
 663       oclass = NV40_SURFACE_SWZ_CLASS;
 664
 665    ret = nouveau_object_new(screen->base.channel, 0xbeef5201, oclass,
 666                             NULL, 0, &screen->swzsurf);
 667    if (ret)
 668       FAIL_SCREEN_INIT("error allocating swizzled surface object: %d\n", ret);
 669
 670    BEGIN_NV04(push, NV01_SUBC(SSWZ, OBJECT), 1);
 671    PUSH_DATA (push, screen->swzsurf->handle);
 672    BEGIN_NV04(push, NV04_SSWZ(DMA_NOTIFY), 1);
 673    PUSH_DATA (push, screen->ntfy->handle);
 674
 675    if (dev->chipset < 0x40)
 676       oclass = NV30_SIFM_CLASS;
 677    else
 678       oclass = NV40_SIFM_CLASS;
 679
 680    ret = nouveau_object_new(screen->base.channel, 0xbeef7701, oclass,
 681                             NULL, 0, &screen->sifm);
 682    if (ret)
 683       FAIL_SCREEN_INIT("error allocating scaled image object: %d\n", ret);
 684
 685    BEGIN_NV04(push, NV01_SUBC(SIFM, OBJECT), 1);
 686    PUSH_DATA (push, screen->sifm->handle);
 687    BEGIN_NV04(push, NV03_SIFM(DMA_NOTIFY), 1);
 688    PUSH_DATA (push, screen->ntfy->handle);
 689    BEGIN_NV04(push, NV05_SIFM(COLOR_CONVERSION), 1);
 690    PUSH_DATA (push, NV05_SIFM_COLOR_CONVERSION_TRUNCATE);
 691
 692    nouveau_pushbuf_kick(push, push->channel);
 693
 694    nouveau_fence_new(&screen->base, &screen->base.fence.current, false);
 695    return pscreen;
 696 }