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    case PIPE_CAP_CLEAR_TEXTURE:
 177       return 0;
 178
 179    case PIPE_CAP_VENDOR_ID:
 180       return 0x10de;
 181    case PIPE_CAP_DEVICE_ID: {
 182       uint64_t device_id;
 183       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
 184          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
 185          return -1;
 186       }
 187       return device_id;
 188    }
 189    case PIPE_CAP_ACCELERATED:
 190       return 1;
 191    case PIPE_CAP_VIDEO_MEMORY:
 192       return dev->vram_size >> 20;
 193    case PIPE_CAP_UMA:
 194       return 0;
 195    }
 196
 197    debug_printf("unknown param %d\n", param);
 198    return 0;
 199 }
 200
 201 static float
 202 nv30_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
 203 {
 204    struct nv30_screen *screen = nv30_screen(pscreen);
 205    struct nouveau_object *eng3d = screen->eng3d;
 206
 207    switch (param) {
 208    case PIPE_CAPF_MAX_LINE_WIDTH:
 209    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 210       return 10.0;
 211    case PIPE_CAPF_MAX_POINT_WIDTH:
 212    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 213       return 64.0;
 214    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 215       return (eng3d->oclass >= NV40_3D_CLASS) ? 16.0 : 8.0;
 216    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 217       return 15.0;
 218    default:
 219       debug_printf("unknown paramf %d\n", param);
 220       return 0;
 221    }
 222 }
 223
 224 static int
 225 nv30_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
 226                              enum pipe_shader_cap param)
 227 {
 228    struct nv30_screen *screen = nv30_screen(pscreen);
 229    struct nouveau_object *eng3d = screen->eng3d;
 230
 231    switch (shader) {
 232    case PIPE_SHADER_VERTEX:
 233       switch (param) {
 234       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 235       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 236          return (eng3d->oclass >= NV40_3D_CLASS) ? 512 : 256;
 237       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 238       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 239          return (eng3d->oclass >= NV40_3D_CLASS) ? 512 : 0;
 240       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 241          return 0;
 242       case PIPE_SHADER_CAP_MAX_INPUTS:
 243       case PIPE_SHADER_CAP_MAX_OUTPUTS:
 244          return 16;
 245       case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 246          return ((eng3d->oclass >= NV40_3D_CLASS) ? (468 - 6): (256 - 6)) * sizeof(float[4]);
 247       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 248          return 1;
 249       case PIPE_SHADER_CAP_MAX_TEMPS:
 250          return (eng3d->oclass >= NV40_3D_CLASS) ? 32 : 13;
 251       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 252       case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 253          return 0;
 254       case PIPE_SHADER_CAP_MAX_PREDS:
 255       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 256       case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 257       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 258       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 259       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 260       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 261       case PIPE_SHADER_CAP_SUBROUTINES:
 262       case PIPE_SHADER_CAP_INTEGERS:
 263       case PIPE_SHADER_CAP_DOUBLES:
 264       case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 265       case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 266       case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 267       case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 268          return 0;
 269       case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 270          return 32;
 271       default:
 272          debug_printf("unknown vertex shader param %d\n", param);
 273          return 0;
 274       }
 275       break;
 276    case PIPE_SHADER_FRAGMENT:
 277       switch (param) {
 278       case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 279       case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 280       case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 281       case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 282          return 4096;
 283       case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 284          return 0;
 285       case PIPE_SHADER_CAP_MAX_INPUTS:
 286          return 8; /* should be possible to do 10 with nv4x */
 287       case PIPE_SHADER_CAP_MAX_OUTPUTS:
 288          return 4;
 289       case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 290          return ((eng3d->oclass >= NV40_3D_CLASS) ? 224 : 32) * sizeof(float[4]);
 291       case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 292          return 1;
 293       case PIPE_SHADER_CAP_MAX_TEMPS:
 294          return 32;
 295       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 296       case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 297          return 16;
 298       case PIPE_SHADER_CAP_MAX_PREDS:
 299       case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 300       case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 301       case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 302       case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 303       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 304       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 305       case PIPE_SHADER_CAP_SUBROUTINES:
 306       case PIPE_SHADER_CAP_DOUBLES:
 307       case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 308       case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 309       case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 310       case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 311          return 0;
 312       case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 313          return 32;
 314       default:
 315          debug_printf("unknown fragment shader param %d\n", param);
 316          return 0;
 317       }
 318       break;
 319    default:
 320       return 0;
 321    }
 322 }
 323
 324 static boolean
 325 nv30_screen_is_format_supported(struct pipe_screen *pscreen,
 326                                 enum pipe_format format,
 327                                 enum pipe_texture_target target,
 328                                 unsigned sample_count,
 329                                 unsigned bindings)
 330 {
 331    if (sample_count > nv30_screen(pscreen)->max_sample_count)
 332       return false;
 333
 334    if (!(0x00000017 & (1 << sample_count)))
 335       return false;
 336
 337    if (!util_format_is_supported(format, bindings)) {
 338       return false;
 339    }
 340
 341    /* transfers & shared are always supported */
 342    bindings &= ~(PIPE_BIND_TRANSFER_READ |
 343                  PIPE_BIND_TRANSFER_WRITE |
 344                  PIPE_BIND_SHARED);
 345
 346    return (nv30_format_info(pscreen, format)->bindings & bindings) == bindings;
 347 }
 348
 349 static void
 350 nv30_screen_fence_emit(struct pipe_screen *pscreen, uint32_t *sequence)
 351 {
 352    struct nv30_screen *screen = nv30_screen(pscreen);
 353    struct nouveau_pushbuf *push = screen->base.pushbuf;
 354
 355    *sequence = ++screen->base.fence.sequence;
 356
 357    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 3);
 358    PUSH_DATA (push, NV30_3D_FENCE_OFFSET |
 359               (2 /* size */ << 18) | (7 /* subchan */ << 13));
 360    PUSH_DATA (push, 0);
 361    PUSH_DATA (push, *sequence);
 362 }
 363
 364 static uint32_t
 365 nv30_screen_fence_update(struct pipe_screen *pscreen)
 366 {
 367    struct nv30_screen *screen = nv30_screen(pscreen);
 368    struct nv04_notify *fence = screen->fence->data;
 369    return *(uint32_t *)((char *)screen->notify->map + fence->offset);
 370 }
 371
 372 static void
 373 nv30_screen_destroy(struct pipe_screen *pscreen)
 374 {
 375    struct nv30_screen *screen = nv30_screen(pscreen);
 376
 377    if (!nouveau_drm_screen_unref(&screen->base))
 378       return;
 379
 380    if (screen->base.fence.current) {
 381       struct nouveau_fence *current = NULL;
 382
 383       /* nouveau_fence_wait will create a new current fence, so wait on the
 384        * _current_ one, and remove both.
 385        */
 386       nouveau_fence_ref(screen->base.fence.current, &current);
 387       nouveau_fence_wait(current, NULL);
 388       nouveau_fence_ref(NULL, &current);
 389       nouveau_fence_ref(NULL, &screen->base.fence.current);
 390    }
 391
 392    nouveau_bo_ref(NULL, &screen->notify);
 393
 394    nouveau_heap_destroy(&screen->query_heap);
 395    nouveau_heap_destroy(&screen->vp_exec_heap);
 396    nouveau_heap_destroy(&screen->vp_data_heap);
 397
 398    nouveau_object_del(&screen->query);
 399    nouveau_object_del(&screen->fence);
 400    nouveau_object_del(&screen->ntfy);
 401
 402    nouveau_object_del(&screen->sifm);
 403    nouveau_object_del(&screen->swzsurf);
 404    nouveau_object_del(&screen->surf2d);
 405    nouveau_object_del(&screen->m2mf);
 406    nouveau_object_del(&screen->eng3d);
 407    nouveau_object_del(&screen->null);
 408
 409    nouveau_screen_fini(&screen->base);
 410    FREE(screen);
 411 }
 412
 413 #define FAIL_SCREEN_INIT(str, err)                    \
 414    do {                                               \
 415       NOUVEAU_ERR(str, err);                          \
 416       nv30_screen_destroy(pscreen);                   \
 417       return NULL;                                    \
 418    } while(0)
 419
 420 struct nouveau_screen *
 421 nv30_screen_create(struct nouveau_device *dev)
 422 {
 423    struct nv30_screen *screen = CALLOC_STRUCT(nv30_screen);
 424    struct pipe_screen *pscreen;
 425    struct nouveau_pushbuf *push;
 426    struct nv04_fifo *fifo;
 427    unsigned oclass = 0;
 428    int ret, i;
 429
 430    if (!screen)
 431       return NULL;
 432
 433    switch (dev->chipset & 0xf0) {
 434    case 0x30:
 435       if (RANKINE_0397_CHIPSET & (1 << (dev->chipset & 0x0f)))
 436          oclass = NV30_3D_CLASS;
 437       else
 438       if (RANKINE_0697_CHIPSET & (1 << (dev->chipset & 0x0f)))
 439          oclass = NV34_3D_CLASS;
 440       else
 441       if (RANKINE_0497_CHIPSET & (1 << (dev->chipset & 0x0f)))
 442          oclass = NV35_3D_CLASS;
 443       break;
 444    case 0x40:
 445       if (CURIE_4097_CHIPSET & (1 << (dev->chipset & 0x0f)))
 446          oclass = NV40_3D_CLASS;
 447       else
 448       if (CURIE_4497_CHIPSET & (1 << (dev->chipset & 0x0f)))
 449          oclass = NV44_3D_CLASS;
 450       break;
 451    case 0x60:
 452       if (CURIE_4497_CHIPSET6X & (1 << (dev->chipset & 0x0f)))
 453          oclass = NV44_3D_CLASS;
 454       break;
 455    default:
 456       break;
 457    }
 458
 459    if (!oclass) {
 460       NOUVEAU_ERR("unknown 3d class for 0x%02x\n", dev->chipset);
 461       FREE(screen);
 462       return NULL;
 463    }
 464
 465    /*
 466     * Some modern apps try to use msaa without keeping in mind the
 467     * restrictions on videomem of older cards. Resulting in dmesg saying:
 468     * [ 1197.850642] nouveau E[soffice.bin[3785]] fail ttm_validate
 469     * [ 1197.850648] nouveau E[soffice.bin[3785]] validating bo list
 470     * [ 1197.850654] nouveau E[soffice.bin[3785]] validate: -12
 471     *
 472     * Because we are running out of video memory, after which the program
 473     * using the msaa visual freezes, and eventually the entire system freezes.
 474     *
 475     * To work around this we do not allow msaa visauls by default and allow
 476     * the user to override this via NV30_MAX_MSAA.
 477     */
 478    screen->max_sample_count = debug_get_num_option("NV30_MAX_MSAA", 0);
 479    if (screen->max_sample_count > 4)
 480       screen->max_sample_count = 4;
 481
 482    pscreen = &screen->base.base;
 483    pscreen->destroy = nv30_screen_destroy;
 484    pscreen->get_param = nv30_screen_get_param;
 485    pscreen->get_paramf = nv30_screen_get_paramf;
 486    pscreen->get_shader_param = nv30_screen_get_shader_param;
 487    pscreen->context_create = nv30_context_create;
 488    pscreen->is_format_supported = nv30_screen_is_format_supported;
 489    nv30_resource_screen_init(pscreen);
 490    nouveau_screen_init_vdec(&screen->base);
 491
 492    screen->base.fence.emit = nv30_screen_fence_emit;
 493    screen->base.fence.update = nv30_screen_fence_update;
 494
 495    ret = nouveau_screen_init(&screen->base, dev);
 496    if (ret)
 497       FAIL_SCREEN_INIT("nv30_screen_init failed: %d\n", ret);
 498
 499    screen->base.vidmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
 500    screen->base.sysmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
 501    if (oclass == NV40_3D_CLASS) {
 502       screen->base.vidmem_bindings |= PIPE_BIND_INDEX_BUFFER;
 503       screen->base.sysmem_bindings |= PIPE_BIND_INDEX_BUFFER;
 504    }
 505
 506    fifo = screen->base.channel->data;
 507    push = screen->base.pushbuf;
 508    push->rsvd_kick = 16;
 509
 510    ret = nouveau_object_new(screen->base.channel, 0x00000000, NV01_NULL_CLASS,
 511                             NULL, 0, &screen->null);
 512    if (ret)
 513       FAIL_SCREEN_INIT("error allocating null object: %d\n", ret);
 514
 515    /* DMA_FENCE refuses to accept DMA objects with "adjust" filled in,
 516     * this means that the address pointed at by the DMA object must
 517     * be 4KiB aligned, which means this object needs to be the first
 518     * one allocated on the channel.
 519     */
 520    ret = nouveau_object_new(screen->base.channel, 0xbeef1e00,
 521                             NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
 522                             .length = 32 }, sizeof(struct nv04_notify),
 523                             &screen->fence);
 524    if (ret)
 525       FAIL_SCREEN_INIT("error allocating fence notifier: %d\n", ret);
 526
 527    /* DMA_NOTIFY object, we don't actually use this but M2MF fails without */
 528    ret = nouveau_object_new(screen->base.channel, 0xbeef0301,
 529                             NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
 530                             .length = 32 }, sizeof(struct nv04_notify),
 531                             &screen->ntfy);
 532    if (ret)
 533       FAIL_SCREEN_INIT("error allocating sync notifier: %d\n", ret);
 534
 535    /* DMA_QUERY, used to implement occlusion queries, we attempt to allocate
 536     * the remainder of the "notifier block" assigned by the kernel for
 537     * use as query objects
 538     */
 539    ret = nouveau_object_new(screen->base.channel, 0xbeef0351,
 540                             NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
 541                             .length = 4096 - 128 }, sizeof(struct nv04_notify),
 542                             &screen->query);
 543    if (ret)
 544       FAIL_SCREEN_INIT("error allocating query notifier: %d\n", ret);
 545
 546    ret = nouveau_heap_init(&screen->query_heap, 0, 4096 - 128);
 547    if (ret)
 548       FAIL_SCREEN_INIT("error creating query heap: %d\n", ret);
 549
 550    LIST_INITHEAD(&screen->queries);
 551
 552    /* Vertex program resources (code/data), currently 6 of the constant
 553     * slots are reserved to implement user clipping planes
 554     */
 555    if (oclass < NV40_3D_CLASS) {
 556       nouveau_heap_init(&screen->vp_exec_heap, 0, 256);
 557       nouveau_heap_init(&screen->vp_data_heap, 6, 256 - 6);
 558    } else {
 559       nouveau_heap_init(&screen->vp_exec_heap, 0, 512);
 560       nouveau_heap_init(&screen->vp_data_heap, 6, 468 - 6);
 561    }
 562
 563    ret = nouveau_bo_wrap(screen->base.device, fifo->notify, &screen->notify);
 564    if (ret == 0)
 565       ret = nouveau_bo_map(screen->notify, 0, screen->base.client);
 566    if (ret)
 567       FAIL_SCREEN_INIT("error mapping notifier memory: %d\n", ret);
 568
 569    ret = nouveau_object_new(screen->base.channel, 0xbeef3097, oclass,
 570                             NULL, 0, &screen->eng3d);
 571    if (ret)
 572       FAIL_SCREEN_INIT("error allocating 3d object: %d\n", ret);
 573
 574    BEGIN_NV04(push, NV01_SUBC(3D, OBJECT), 1);
 575    PUSH_DATA (push, screen->eng3d->handle);
 576    BEGIN_NV04(push, NV30_3D(DMA_NOTIFY), 13);
 577    PUSH_DATA (push, screen->ntfy->handle);
 578    PUSH_DATA (push, fifo->vram);     /* TEXTURE0 */
 579    PUSH_DATA (push, fifo->gart);     /* TEXTURE1 */
 580    PUSH_DATA (push, fifo->vram);     /* COLOR1 */
 581    PUSH_DATA (push, screen->null->handle);  /* UNK190 */
 582    PUSH_DATA (push, fifo->vram);     /* COLOR0 */
 583    PUSH_DATA (push, fifo->vram);     /* ZETA */
 584    PUSH_DATA (push, fifo->vram);     /* VTXBUF0 */
 585    PUSH_DATA (push, fifo->gart);     /* VTXBUF1 */
 586    PUSH_DATA (push, screen->fence->handle);  /* FENCE */
 587    PUSH_DATA (push, screen->query->handle);  /* QUERY - intr 0x80 if nullobj */
 588    PUSH_DATA (push, screen->null->handle);  /* UNK1AC */
 589    PUSH_DATA (push, screen->null->handle);  /* UNK1B0 */
 590    if (screen->eng3d->oclass < NV40_3D_CLASS) {
 591       BEGIN_NV04(push, SUBC_3D(0x03b0), 1);
 592       PUSH_DATA (push, 0x00100000);
 593       BEGIN_NV04(push, SUBC_3D(0x1d80), 1);
 594       PUSH_DATA (push, 3);
 595
 596       BEGIN_NV04(push, SUBC_3D(0x1e98), 1);
 597       PUSH_DATA (push, 0);
 598       BEGIN_NV04(push, SUBC_3D(0x17e0), 3);
 599       PUSH_DATA (push, fui(0.0));
 600       PUSH_DATA (push, fui(0.0));
 601       PUSH_DATA (push, fui(1.0));
 602       BEGIN_NV04(push, SUBC_3D(0x1f80), 16);
 603       for (i = 0; i < 16; i++)
 604          PUSH_DATA (push, (i == 8) ? 0x0000ffff : 0);
 605
 606       BEGIN_NV04(push, NV30_3D(RC_ENABLE), 1);
 607       PUSH_DATA (push, 0);
 608    } else {
 609       BEGIN_NV04(push, NV40_3D(DMA_COLOR2), 2);
 610       PUSH_DATA (push, fifo->vram);
 611       PUSH_DATA (push, fifo->vram);  /* COLOR3 */
 612
 613       BEGIN_NV04(push, SUBC_3D(0x1450), 1);
 614       PUSH_DATA (push, 0x00000004);
 615
 616       BEGIN_NV04(push, SUBC_3D(0x1ea4), 3); /* ZCULL */
 617       PUSH_DATA (push, 0x00000010);
 618       PUSH_DATA (push, 0x01000100);
 619       PUSH_DATA (push, 0xff800006);
 620
 621       /* vtxprog output routing */
 622       BEGIN_NV04(push, SUBC_3D(0x1fc4), 1);
 623       PUSH_DATA (push, 0x06144321);
 624       BEGIN_NV04(push, SUBC_3D(0x1fc8), 2);
 625       PUSH_DATA (push, 0xedcba987);
 626       PUSH_DATA (push, 0x0000006f);
 627       BEGIN_NV04(push, SUBC_3D(0x1fd0), 1);
 628       PUSH_DATA (push, 0x00171615);
 629       BEGIN_NV04(push, SUBC_3D(0x1fd4), 1);
 630       PUSH_DATA (push, 0x001b1a19);
 631
 632       BEGIN_NV04(push, SUBC_3D(0x1ef8), 1);
 633       PUSH_DATA (push, 0x0020ffff);
 634       BEGIN_NV04(push, SUBC_3D(0x1d64), 1);
 635       PUSH_DATA (push, 0x01d300d4);
 636
 637       BEGIN_NV04(push, NV40_3D(MIPMAP_ROUNDING), 1);
 638       PUSH_DATA (push, NV40_3D_MIPMAP_ROUNDING_MODE_DOWN);
 639    }
 640
 641    ret = nouveau_object_new(screen->base.channel, 0xbeef3901, NV03_M2MF_CLASS,
 642                             NULL, 0, &screen->m2mf);
 643    if (ret)
 644       FAIL_SCREEN_INIT("error allocating m2mf object: %d\n", ret);
 645
 646    BEGIN_NV04(push, NV01_SUBC(M2MF, OBJECT), 1);
 647    PUSH_DATA (push, screen->m2mf->handle);
 648    BEGIN_NV04(push, NV03_M2MF(DMA_NOTIFY), 1);
 649    PUSH_DATA (push, screen->ntfy->handle);
 650
 651    ret = nouveau_object_new(screen->base.channel, 0xbeef6201,
 652                             NV10_SURFACE_2D_CLASS, NULL, 0, &screen->surf2d);
 653    if (ret)
 654       FAIL_SCREEN_INIT("error allocating surf2d object: %d\n", ret);
 655
 656    BEGIN_NV04(push, NV01_SUBC(SF2D, OBJECT), 1);
 657    PUSH_DATA (push, screen->surf2d->handle);
 658    BEGIN_NV04(push, NV04_SF2D(DMA_NOTIFY), 1);
 659    PUSH_DATA (push, screen->ntfy->handle);
 660
 661    if (dev->chipset < 0x40)
 662       oclass = NV30_SURFACE_SWZ_CLASS;
 663    else
 664       oclass = NV40_SURFACE_SWZ_CLASS;
 665
 666    ret = nouveau_object_new(screen->base.channel, 0xbeef5201, oclass,
 667                             NULL, 0, &screen->swzsurf);
 668    if (ret)
 669       FAIL_SCREEN_INIT("error allocating swizzled surface object: %d\n", ret);
 670
 671    BEGIN_NV04(push, NV01_SUBC(SSWZ, OBJECT), 1);
 672    PUSH_DATA (push, screen->swzsurf->handle);
 673    BEGIN_NV04(push, NV04_SSWZ(DMA_NOTIFY), 1);
 674    PUSH_DATA (push, screen->ntfy->handle);
 675
 676    if (dev->chipset < 0x40)
 677       oclass = NV30_SIFM_CLASS;
 678    else
 679       oclass = NV40_SIFM_CLASS;
 680
 681    ret = nouveau_object_new(screen->base.channel, 0xbeef7701, oclass,
 682                             NULL, 0, &screen->sifm);
 683    if (ret)
 684       FAIL_SCREEN_INIT("error allocating scaled image object: %d\n", ret);
 685
 686    BEGIN_NV04(push, NV01_SUBC(SIFM, OBJECT), 1);
 687    PUSH_DATA (push, screen->sifm->handle);
 688    BEGIN_NV04(push, NV03_SIFM(DMA_NOTIFY), 1);
 689    PUSH_DATA (push, screen->ntfy->handle);
 690    BEGIN_NV04(push, NV05_SIFM(COLOR_CONVERSION), 1);
 691    PUSH_DATA (push, NV05_SIFM_COLOR_CONVERSION_TRUNCATE);
 692
 693    nouveau_pushbuf_kick(push, push->channel);
 694
 695    nouveau_fence_new(&screen->base, &screen->base.fence.current, false);
 696    return &screen->base;
 697 }