src/gallium/drivers/nouveau/nv50/nv50_screen.c

   1 /*
   2  * Copyright 2010 Christoph Bumiller
   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
  23 #include <errno.h>
  24 #include <xf86drm.h>
  25 #include <nouveau_drm.h>
  26 #include "util/u_format.h"
  27 #include "util/u_format_s3tc.h"
  28 #include "pipe/p_screen.h"
  29
  30 #include "nv50/nv50_context.h"
  31 #include "nv50/nv50_screen.h"
  32
  33 #include "nouveau_vp3_video.h"
  34
  35 #include "nv_object.xml.h"
  36
  37 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
  38 #define LOCAL_WARPS_ALLOC 32
  39 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
  40 #define STACK_WARPS_ALLOC 32
  41
  42 #define THREADS_IN_WARP 32
  43
  44 static boolean
  45 nv50_screen_is_format_supported(struct pipe_screen *pscreen,
  46                                 enum pipe_format format,
  47                                 enum pipe_texture_target target,
  48                                 unsigned sample_count,
  49                                 unsigned bindings)
  50 {
  51    if (sample_count > 8)
  52       return false;
  53    if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
  54       return false;
  55    if (sample_count == 8 && util_format_get_blocksizebits(format) >= 128)
  56       return false;
  57
  58    if (!util_format_is_supported(format, bindings))
  59       return false;
  60
  61    switch (format) {
  62    case PIPE_FORMAT_Z16_UNORM:
  63       if (nv50_screen(pscreen)->tesla->oclass < NVA0_3D_CLASS)
  64          return false;
  65       break;
  66    default:
  67       break;
  68    }
  69
  70    if (bindings & PIPE_BIND_LINEAR)
  71       if (util_format_is_depth_or_stencil(format) ||
  72           (target != PIPE_TEXTURE_1D &&
  73            target != PIPE_TEXTURE_2D &&
  74            target != PIPE_TEXTURE_RECT) ||
  75           sample_count > 1)
  76          return false;
  77
  78    /* transfers & shared are always supported */
  79    bindings &= ~(PIPE_BIND_TRANSFER_READ |
  80                  PIPE_BIND_TRANSFER_WRITE |
  81                  PIPE_BIND_LINEAR |
  82                  PIPE_BIND_SHARED);
  83
  84    return (( nv50_format_table[format].usage |
  85             nv50_vertex_format[format].usage) & bindings) == bindings;
  86 }
  87
  88 static int
  89 nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
  90 {
  91    const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
  92    struct nouveau_device *dev = nouveau_screen(pscreen)->device;
  93
  94    switch (param) {
  95    /* non-boolean caps */
  96    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
  97       return 14;
  98    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
  99       return 12;
 100    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 101       return 14;
 102    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 103       return 512;
 104    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 105    case PIPE_CAP_MIN_TEXEL_OFFSET:
 106       return -8;
 107    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 108    case PIPE_CAP_MAX_TEXEL_OFFSET:
 109       return 7;
 110    case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 111       return 128 * 1024 * 1024;
 112    case PIPE_CAP_GLSL_FEATURE_LEVEL:
 113       return 330;
 114    case PIPE_CAP_MAX_RENDER_TARGETS:
 115       return 8;
 116    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 117       return 1;
 118    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 119       return 4;
 120    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 121    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 122       return 64;
 123    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 124    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 125       return 1024;
 126    case PIPE_CAP_MAX_VERTEX_STREAMS:
 127       return 1;
 128    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 129       return 2048;
 130    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 131       return 256;
 132    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 133       return 16; /* 256 for binding as RT, but that's not possible in GL */
 134    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 135       return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
 136    case PIPE_CAP_MAX_VIEWPORTS:
 137       return NV50_MAX_VIEWPORTS;
 138    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 139       return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
 140    case PIPE_CAP_ENDIANNESS:
 141       return PIPE_ENDIAN_LITTLE;
 142    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 143       return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
 144
 145    /* supported caps */
 146    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 147    case PIPE_CAP_TEXTURE_SWIZZLE:
 148    case PIPE_CAP_TEXTURE_SHADOW_MAP:
 149    case PIPE_CAP_NPOT_TEXTURES:
 150    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 151    case PIPE_CAP_ANISOTROPIC_FILTER:
 152    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 153    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 154    case PIPE_CAP_TWO_SIDED_STENCIL:
 155    case PIPE_CAP_DEPTH_CLIP_DISABLE:
 156    case PIPE_CAP_POINT_SPRITE:
 157    case PIPE_CAP_SM3:
 158    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 159    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 160    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 161    case PIPE_CAP_QUERY_TIMESTAMP:
 162    case PIPE_CAP_QUERY_TIME_ELAPSED:
 163    case PIPE_CAP_OCCLUSION_QUERY:
 164    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 165    case PIPE_CAP_INDEP_BLEND_ENABLE:
 166    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 167    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 168    case PIPE_CAP_PRIMITIVE_RESTART:
 169    case PIPE_CAP_TGSI_INSTANCEID:
 170    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 171    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 172    case PIPE_CAP_CONDITIONAL_RENDER:
 173    case PIPE_CAP_TEXTURE_BARRIER:
 174    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 175    case PIPE_CAP_START_INSTANCE:
 176    case PIPE_CAP_USER_CONSTANT_BUFFERS:
 177    case PIPE_CAP_USER_INDEX_BUFFERS:
 178    case PIPE_CAP_USER_VERTEX_BUFFERS:
 179    case PIPE_CAP_TEXTURE_MULTISAMPLE:
 180    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 181    case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 182    case PIPE_CAP_SAMPLER_VIEW_TARGET:
 183    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 184    case PIPE_CAP_CLIP_HALFZ:
 185    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 186    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 187    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 188    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 189    case PIPE_CAP_DEPTH_BOUNDS_TEST:
 190    case PIPE_CAP_TGSI_TXQS:
 191    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 192    case PIPE_CAP_SHAREABLE_SHADERS:
 193    case PIPE_CAP_CLEAR_TEXTURE:
 194    case PIPE_CAP_COMPUTE:
 195    case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 196    case PIPE_CAP_INVALIDATE_BUFFER:
 197    case PIPE_CAP_STRING_MARKER:
 198    case PIPE_CAP_CULL_DISTANCE:
 199       return 1;
 200    case PIPE_CAP_SEAMLESS_CUBE_MAP:
 201       return 1; /* class_3d >= NVA0_3D_CLASS; */
 202    /* supported on nva0+ */
 203    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 204       return class_3d >= NVA0_3D_CLASS;
 205    /* supported on nva3+ */
 206    case PIPE_CAP_CUBE_MAP_ARRAY:
 207    case PIPE_CAP_INDEP_BLEND_FUNC:
 208    case PIPE_CAP_TEXTURE_QUERY_LOD:
 209    case PIPE_CAP_SAMPLE_SHADING:
 210    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 211       return class_3d >= NVA3_3D_CLASS;
 212
 213    /* unsupported caps */
 214    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 215    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 216    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 217    case PIPE_CAP_SHADER_STENCIL_EXPORT:
 218    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 219    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 220    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 221    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 222    case PIPE_CAP_TGSI_TEXCOORD:
 223    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 224    case PIPE_CAP_TEXTURE_GATHER_SM5:
 225    case PIPE_CAP_FAKE_SW_MSAA:
 226    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 227    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 228    case PIPE_CAP_DRAW_INDIRECT:
 229    case PIPE_CAP_MULTI_DRAW_INDIRECT:
 230    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 231    case PIPE_CAP_VERTEXID_NOBASE:
 232    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
 233    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 234    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 235    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 236    case PIPE_CAP_DRAW_PARAMETERS:
 237    case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 238    case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 239    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 240    case PIPE_CAP_GENERATE_MIPMAP:
 241    case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 242    case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 243    case PIPE_CAP_QUERY_BUFFER_OBJECT:
 244    case PIPE_CAP_QUERY_MEMORY_INFO:
 245    case PIPE_CAP_PCI_GROUP:
 246    case PIPE_CAP_PCI_BUS:
 247    case PIPE_CAP_PCI_DEVICE:
 248    case PIPE_CAP_PCI_FUNCTION:
 249    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 250    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 251    case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 252    case PIPE_CAP_TGSI_VOTE:
 253    case PIPE_CAP_MAX_WINDOW_RECTANGLES:
 254       return 0;
 255
 256    case PIPE_CAP_VENDOR_ID:
 257       return 0x10de;
 258    case PIPE_CAP_DEVICE_ID: {
 259       uint64_t device_id;
 260       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
 261          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
 262          return -1;
 263       }
 264       return device_id;
 265    }
 266    case PIPE_CAP_ACCELERATED:
 267       return 1;
 268    case PIPE_CAP_VIDEO_MEMORY:
 269       return dev->vram_size >> 20;
 270    case PIPE_CAP_UMA:
 271       return 0;
 272    }
 273
 274    NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
 275    return 0;
 276 }
 277
 278 static int
 279 nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
 280                              enum pipe_shader_cap param)
 281 {
 282    switch (shader) {
 283    case PIPE_SHADER_VERTEX:
 284    case PIPE_SHADER_GEOMETRY:
 285    case PIPE_SHADER_FRAGMENT:
 286       break;
 287    case PIPE_SHADER_COMPUTE:
 288    default:
 289       return 0;
 290    }
 291
 292    switch (param) {
 293    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 294    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 295    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 296    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 297       return 16384;
 298    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 299       return 4;
 300    case PIPE_SHADER_CAP_MAX_INPUTS:
 301       if (shader == PIPE_SHADER_VERTEX)
 302          return 32;
 303       return 15;
 304    case PIPE_SHADER_CAP_MAX_OUTPUTS:
 305       return 16;
 306    case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 307       return 65536;
 308    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 309       return NV50_MAX_PIPE_CONSTBUFS;
 310    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 311       return shader != PIPE_SHADER_FRAGMENT;
 312    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 313    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 314    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 315       return 1;
 316    case PIPE_SHADER_CAP_MAX_PREDS:
 317       return 0;
 318    case PIPE_SHADER_CAP_MAX_TEMPS:
 319       return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
 320    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 321       return 1;
 322    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 323       return 1;
 324    case PIPE_SHADER_CAP_SUBROUTINES:
 325       return 0; /* please inline, or provide function declarations */
 326    case PIPE_SHADER_CAP_INTEGERS:
 327       return 1;
 328    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 329       /* The chip could handle more sampler views than samplers */
 330    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 331       return MIN2(16, PIPE_MAX_SAMPLERS);
 332    case PIPE_SHADER_CAP_PREFERRED_IR:
 333       return PIPE_SHADER_IR_TGSI;
 334    case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 335       return 32;
 336    case PIPE_SHADER_CAP_DOUBLES:
 337    case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 338    case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 339    case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 340    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 341    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 342    case PIPE_SHADER_CAP_SUPPORTED_IRS:
 343    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 344       return 0;
 345    default:
 346       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
 347       return 0;
 348    }
 349 }
 350
 351 static float
 352 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
 353 {
 354    switch (param) {
 355    case PIPE_CAPF_MAX_LINE_WIDTH:
 356    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 357       return 10.0f;
 358    case PIPE_CAPF_MAX_POINT_WIDTH:
 359    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 360       return 64.0f;
 361    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 362       return 16.0f;
 363    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 364       return 4.0f;
 365    case PIPE_CAPF_GUARD_BAND_LEFT:
 366    case PIPE_CAPF_GUARD_BAND_TOP:
 367       return 0.0f;
 368    case PIPE_CAPF_GUARD_BAND_RIGHT:
 369    case PIPE_CAPF_GUARD_BAND_BOTTOM:
 370       return 0.0f; /* that or infinity */
 371    }
 372
 373    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
 374    return 0.0f;
 375 }
 376
 377 static int
 378 nv50_screen_get_compute_param(struct pipe_screen *pscreen,
 379                               enum pipe_shader_ir ir_type,
 380                               enum pipe_compute_cap param, void *data)
 381 {
 382    struct nv50_screen *screen = nv50_screen(pscreen);
 383
 384 #define RET(x) do {                  \
 385    if (data)                         \
 386       memcpy(data, x, sizeof(x));    \
 387    return sizeof(x);                 \
 388 } while (0)
 389
 390    switch (param) {
 391    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
 392       RET((uint64_t []) { 2 });
 393    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
 394       RET(((uint64_t []) { 65535, 65535 }));
 395    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
 396       RET(((uint64_t []) { 512, 512, 64 }));
 397    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
 398       RET((uint64_t []) { 512 });
 399    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g0-15[] */
 400       RET((uint64_t []) { 1ULL << 32 });
 401    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
 402       RET((uint64_t []) { 16 << 10 });
 403    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
 404       RET((uint64_t []) { 16 << 10 });
 405    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
 406       RET((uint64_t []) { 4096 });
 407    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
 408       RET((uint32_t []) { 32 });
 409    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
 410       RET((uint64_t []) { 1ULL << 40 });
 411    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
 412       RET((uint32_t []) { 0 });
 413    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
 414       RET((uint32_t []) { screen->mp_count });
 415    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
 416       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
 417    default:
 418       return 0;
 419    }
 420
 421 #undef RET
 422 }
 423
 424 static void
 425 nv50_screen_destroy(struct pipe_screen *pscreen)
 426 {
 427    struct nv50_screen *screen = nv50_screen(pscreen);
 428
 429    if (!nouveau_drm_screen_unref(&screen->base))
 430       return;
 431
 432    if (screen->base.fence.current) {
 433       struct nouveau_fence *current = NULL;
 434
 435       /* nouveau_fence_wait will create a new current fence, so wait on the
 436        * _current_ one, and remove both.
 437        */
 438       nouveau_fence_ref(screen->base.fence.current, &current);
 439       nouveau_fence_wait(current, NULL);
 440       nouveau_fence_ref(NULL, &current);
 441       nouveau_fence_ref(NULL, &screen->base.fence.current);
 442    }
 443    if (screen->base.pushbuf)
 444       screen->base.pushbuf->user_priv = NULL;
 445
 446    if (screen->blitter)
 447       nv50_blitter_destroy(screen);
 448    if (screen->pm.prog) {
 449       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
 450       nv50_program_destroy(NULL, screen->pm.prog);
 451       FREE(screen->pm.prog);
 452    }
 453
 454    nouveau_bo_ref(NULL, &screen->code);
 455    nouveau_bo_ref(NULL, &screen->tls_bo);
 456    nouveau_bo_ref(NULL, &screen->stack_bo);
 457    nouveau_bo_ref(NULL, &screen->txc);
 458    nouveau_bo_ref(NULL, &screen->uniforms);
 459    nouveau_bo_ref(NULL, &screen->fence.bo);
 460
 461    nouveau_heap_destroy(&screen->vp_code_heap);
 462    nouveau_heap_destroy(&screen->gp_code_heap);
 463    nouveau_heap_destroy(&screen->fp_code_heap);
 464
 465    FREE(screen->tic.entries);
 466
 467    nouveau_object_del(&screen->tesla);
 468    nouveau_object_del(&screen->eng2d);
 469    nouveau_object_del(&screen->m2mf);
 470    nouveau_object_del(&screen->compute);
 471    nouveau_object_del(&screen->sync);
 472
 473    nouveau_screen_fini(&screen->base);
 474
 475    FREE(screen);
 476 }
 477
 478 static void
 479 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
 480 {
 481    struct nv50_screen *screen = nv50_screen(pscreen);
 482    struct nouveau_pushbuf *push = screen->base.pushbuf;
 483
 484    /* we need to do it after possible flush in MARK_RING */
 485    *sequence = ++screen->base.fence.sequence;
 486
 487    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
 488    PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
 489    PUSH_DATAh(push, screen->fence.bo->offset);
 490    PUSH_DATA (push, screen->fence.bo->offset);
 491    PUSH_DATA (push, *sequence);
 492    PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
 493                     NV50_3D_QUERY_GET_UNK4 |
 494                     NV50_3D_QUERY_GET_UNIT_CROP |
 495                     NV50_3D_QUERY_GET_TYPE_QUERY |
 496                     NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
 497                     NV50_3D_QUERY_GET_SHORT);
 498 }
 499
 500 static u32
 501 nv50_screen_fence_update(struct pipe_screen *pscreen)
 502 {
 503    return nv50_screen(pscreen)->fence.map[0];
 504 }
 505
 506 static void
 507 nv50_screen_init_hwctx(struct nv50_screen *screen)
 508 {
 509    struct nouveau_pushbuf *push = screen->base.pushbuf;
 510    struct nv04_fifo *fifo;
 511    unsigned i;
 512
 513    fifo = (struct nv04_fifo *)screen->base.channel->data;
 514
 515    BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
 516    PUSH_DATA (push, screen->m2mf->handle);
 517    BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
 518    PUSH_DATA (push, screen->sync->handle);
 519    PUSH_DATA (push, fifo->vram);
 520    PUSH_DATA (push, fifo->vram);
 521
 522    BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
 523    PUSH_DATA (push, screen->eng2d->handle);
 524    BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
 525    PUSH_DATA (push, screen->sync->handle);
 526    PUSH_DATA (push, fifo->vram);
 527    PUSH_DATA (push, fifo->vram);
 528    PUSH_DATA (push, fifo->vram);
 529    BEGIN_NV04(push, NV50_2D(OPERATION), 1);
 530    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
 531    BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
 532    PUSH_DATA (push, 0);
 533    BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
 534    PUSH_DATA (push, 0);
 535    BEGIN_NV04(push, SUBC_2D(0x0888), 1);
 536    PUSH_DATA (push, 1);
 537    BEGIN_NV04(push, NV50_2D(COND_MODE), 1);
 538    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
 539
 540    BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
 541    PUSH_DATA (push, screen->tesla->handle);
 542
 543    BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
 544    PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
 545
 546    BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
 547    PUSH_DATA (push, screen->sync->handle);
 548    BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
 549    for (i = 0; i < 11; ++i)
 550       PUSH_DATA(push, fifo->vram);
 551    BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
 552    for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
 553       PUSH_DATA(push, fifo->vram);
 554
 555    BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
 556    PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
 557    BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
 558    PUSH_DATA (push, 0xf);
 559
 560    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
 561       BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
 562       PUSH_DATA (push, 0x18);
 563    }
 564
 565    BEGIN_NV04(push, NV50_3D(ZETA_COMP_ENABLE), 1);
 566    PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 567
 568    BEGIN_NV04(push, NV50_3D(RT_COMP_ENABLE(0)), 8);
 569    for (i = 0; i < 8; ++i)
 570       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 571
 572    BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
 573    PUSH_DATA (push, 1);
 574
 575    BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
 576    PUSH_DATA (push, 0);
 577    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
 578    PUSH_DATA (push, 0);
 579    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
 580    PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
 581    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
 582    PUSH_DATA (push, 0);
 583    BEGIN_NV04(push, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
 584    PUSH_DATA (push, 1);
 585    BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
 586    PUSH_DATA (push, 1);
 587
 588    if (screen->tesla->oclass >= NVA0_3D_CLASS) {
 589       BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
 590       PUSH_DATA (push, 0);
 591    }
 592
 593    BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
 594    PUSH_DATA (push, 0);
 595    BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
 596    PUSH_DATA (push, 0);
 597    PUSH_DATA (push, 0);
 598    BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
 599    PUSH_DATA (push, 0x3f);
 600
 601    BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
 602    PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 603    PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 604
 605    BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
 606    PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 607    PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 608
 609    BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
 610    PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 611    PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 612
 613    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 614    PUSH_DATAh(push, screen->tls_bo->offset);
 615    PUSH_DATA (push, screen->tls_bo->offset);
 616    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 617
 618    BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
 619    PUSH_DATAh(push, screen->stack_bo->offset);
 620    PUSH_DATA (push, screen->stack_bo->offset);
 621    PUSH_DATA (push, 4);
 622
 623    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 624    PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
 625    PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
 626    PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
 627
 628    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 629    PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
 630    PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
 631    PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
 632
 633    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 634    PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
 635    PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
 636    PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
 637
 638    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 639    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
 640    PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
 641    PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
 642
 643    BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
 644    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
 645    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
 646    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
 647
 648    /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
 649    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
 650    PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
 651    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
 652    PUSH_DATAf(push, 0.0f);
 653    PUSH_DATAf(push, 0.0f);
 654    PUSH_DATAf(push, 0.0f);
 655    PUSH_DATAf(push, 0.0f);
 656    BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
 657    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 658    PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 659
 660    nv50_upload_ms_info(push);
 661
 662    /* max TIC (bits 4:8) & TSC bindings, per program type */
 663    for (i = 0; i < 3; ++i) {
 664       BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
 665       PUSH_DATA (push, 0x54);
 666    }
 667
 668    BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
 669    PUSH_DATAh(push, screen->txc->offset);
 670    PUSH_DATA (push, screen->txc->offset);
 671    PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
 672
 673    BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
 674    PUSH_DATAh(push, screen->txc->offset + 65536);
 675    PUSH_DATA (push, screen->txc->offset + 65536);
 676    PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
 677
 678    BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
 679    PUSH_DATA (push, 0);
 680
 681    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
 682    PUSH_DATA (push, 0);
 683    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
 684    PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
 685    BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
 686    for (i = 0; i < 8 * 2; ++i)
 687       PUSH_DATA(push, 0);
 688    BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
 689    PUSH_DATA (push, 0);
 690
 691    BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
 692    PUSH_DATA (push, 1);
 693    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 694       BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(i)), 2);
 695       PUSH_DATAf(push, 0.0f);
 696       PUSH_DATAf(push, 1.0f);
 697       BEGIN_NV04(push, NV50_3D(VIEWPORT_HORIZ(i)), 2);
 698       PUSH_DATA (push, 8192 << 16);
 699       PUSH_DATA (push, 8192 << 16);
 700    }
 701
 702    BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
 703 #ifdef NV50_SCISSORS_CLIPPING
 704    PUSH_DATA (push, 0x0000);
 705 #else
 706    PUSH_DATA (push, 0x1080);
 707 #endif
 708
 709    BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
 710    PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
 711
 712    /* We use scissors instead of exact view volume clipping,
 713     * so they're always enabled.
 714     */
 715    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 716       BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(i)), 3);
 717       PUSH_DATA (push, 1);
 718       PUSH_DATA (push, 8192 << 16);
 719       PUSH_DATA (push, 8192 << 16);
 720    }
 721
 722    BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
 723    PUSH_DATA (push, 1);
 724    BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
 725    PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
 726    BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
 727    PUSH_DATA (push, 0x11111111);
 728    BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
 729    PUSH_DATA (push, 1);
 730
 731    BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
 732    PUSH_DATA (push, 0);
 733    if (screen->base.class_3d >= NV84_3D_CLASS) {
 734       BEGIN_NV04(push, NV84_3D(VERTEX_ID_BASE), 1);
 735       PUSH_DATA (push, 0);
 736    }
 737
 738    PUSH_KICK (push);
 739 }
 740
 741 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
 742       uint64_t *tls_size)
 743 {
 744    struct nouveau_device *dev = screen->base.device;
 745    int ret;
 746
 747    screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
 748          ONE_TEMP_SIZE;
 749    if (nouveau_mesa_debug)
 750       debug_printf("allocating space for %u temps\n",
 751             util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
 752    *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
 753          screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
 754
 755    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 756                         *tls_size, NULL, &screen->tls_bo);
 757    if (ret) {
 758       NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
 759       return ret;
 760    }
 761
 762    return 0;
 763 }
 764
 765 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
 766 {
 767    struct nouveau_pushbuf *push = screen->base.pushbuf;
 768    int ret;
 769    uint64_t tls_size;
 770
 771    if (tls_space < screen->cur_tls_space)
 772       return 0;
 773    if (tls_space > screen->max_tls_space) {
 774       /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
 775        * LOCAL_WARPS_NO_CLAMP) */
 776       NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
 777             (unsigned)(tls_space / ONE_TEMP_SIZE),
 778             (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
 779       return -ENOMEM;
 780    }
 781
 782    nouveau_bo_ref(NULL, &screen->tls_bo);
 783    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
 784    if (ret)
 785       return ret;
 786
 787    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 788    PUSH_DATAh(push, screen->tls_bo->offset);
 789    PUSH_DATA (push, screen->tls_bo->offset);
 790    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 791
 792    return 1;
 793 }
 794
 795 struct nouveau_screen *
 796 nv50_screen_create(struct nouveau_device *dev)
 797 {
 798    struct nv50_screen *screen;
 799    struct pipe_screen *pscreen;
 800    struct nouveau_object *chan;
 801    uint64_t value;
 802    uint32_t tesla_class;
 803    unsigned stack_size;
 804    int ret;
 805
 806    screen = CALLOC_STRUCT(nv50_screen);
 807    if (!screen)
 808       return NULL;
 809    pscreen = &screen->base.base;
 810    pscreen->destroy = nv50_screen_destroy;
 811
 812    ret = nouveau_screen_init(&screen->base, dev);
 813    if (ret) {
 814       NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
 815       goto fail;
 816    }
 817
 818    /* TODO: Prevent FIFO prefetch before transfer of index buffers and
 819     *  admit them to VRAM.
 820     */
 821    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
 822       PIPE_BIND_VERTEX_BUFFER;
 823    screen->base.sysmem_bindings |=
 824       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
 825
 826    screen->base.pushbuf->user_priv = screen;
 827    screen->base.pushbuf->rsvd_kick = 5;
 828
 829    chan = screen->base.channel;
 830
 831    pscreen->context_create = nv50_create;
 832    pscreen->is_format_supported = nv50_screen_is_format_supported;
 833    pscreen->get_param = nv50_screen_get_param;
 834    pscreen->get_shader_param = nv50_screen_get_shader_param;
 835    pscreen->get_paramf = nv50_screen_get_paramf;
 836    pscreen->get_compute_param = nv50_screen_get_compute_param;
 837    pscreen->get_driver_query_info = nv50_screen_get_driver_query_info;
 838    pscreen->get_driver_query_group_info = nv50_screen_get_driver_query_group_info;
 839
 840    nv50_screen_init_resource_functions(pscreen);
 841
 842    if (screen->base.device->chipset < 0x84 ||
 843        debug_get_bool_option("NOUVEAU_PMPEG", false)) {
 844       /* PMPEG */
 845       nouveau_screen_init_vdec(&screen->base);
 846    } else if (screen->base.device->chipset < 0x98 ||
 847               screen->base.device->chipset == 0xa0) {
 848       /* VP2 */
 849       screen->base.base.get_video_param = nv84_screen_get_video_param;
 850       screen->base.base.is_video_format_supported = nv84_screen_video_supported;
 851    } else {
 852       /* VP3/4 */
 853       screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
 854       screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
 855    }
 856
 857    ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
 858                         NULL, &screen->fence.bo);
 859    if (ret) {
 860       NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
 861       goto fail;
 862    }
 863
 864    nouveau_bo_map(screen->fence.bo, 0, NULL);
 865    screen->fence.map = screen->fence.bo->map;
 866    screen->base.fence.emit = nv50_screen_fence_emit;
 867    screen->base.fence.update = nv50_screen_fence_update;
 868
 869    ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
 870                             &(struct nv04_notify){ .length = 32 },
 871                             sizeof(struct nv04_notify), &screen->sync);
 872    if (ret) {
 873       NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
 874       goto fail;
 875    }
 876
 877    ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
 878                             NULL, 0, &screen->m2mf);
 879    if (ret) {
 880       NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
 881       goto fail;
 882    }
 883
 884    ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
 885                             NULL, 0, &screen->eng2d);
 886    if (ret) {
 887       NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
 888       goto fail;
 889    }
 890
 891    switch (dev->chipset & 0xf0) {
 892    case 0x50:
 893       tesla_class = NV50_3D_CLASS;
 894       break;
 895    case 0x80:
 896    case 0x90:
 897       tesla_class = NV84_3D_CLASS;
 898       break;
 899    case 0xa0:
 900       switch (dev->chipset) {
 901       case 0xa0:
 902       case 0xaa:
 903       case 0xac:
 904          tesla_class = NVA0_3D_CLASS;
 905          break;
 906       case 0xaf:
 907          tesla_class = NVAF_3D_CLASS;
 908          break;
 909       default:
 910          tesla_class = NVA3_3D_CLASS;
 911          break;
 912       }
 913       break;
 914    default:
 915       NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
 916       goto fail;
 917    }
 918    screen->base.class_3d = tesla_class;
 919
 920    ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
 921                             NULL, 0, &screen->tesla);
 922    if (ret) {
 923       NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
 924       goto fail;
 925    }
 926
 927    /* This over-allocates by a page. The GP, which would execute at the end of
 928     * the last page, would trigger faults. The going theory is that it
 929     * prefetches up to a certain amount.
 930     */
 931    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 932                         (3 << NV50_CODE_BO_SIZE_LOG2) + 0x1000,
 933                         NULL, &screen->code);
 934    if (ret) {
 935       NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
 936       goto fail;
 937    }
 938
 939    nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 940    nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 941    nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 942
 943    nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
 944
 945    screen->TPs = util_bitcount(value & 0xffff);
 946    screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
 947
 948    screen->mp_count = screen->TPs * screen->MPsInTP;
 949
 950    stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
 951          STACK_WARPS_ALLOC * 64 * 8;
 952
 953    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
 954                         &screen->stack_bo);
 955    if (ret) {
 956       NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
 957       goto fail;
 958    }
 959
 960    uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
 961          screen->MPsInTP * LOCAL_WARPS_ALLOC *  THREADS_IN_WARP *
 962          ONE_TEMP_SIZE;
 963    screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
 964    screen->max_tls_space /= 2; /* half of vram */
 965
 966    /* hw can address max 64 KiB */
 967    screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
 968
 969    uint64_t tls_size;
 970    unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
 971    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
 972    if (ret)
 973       goto fail;
 974
 975    if (nouveau_mesa_debug)
 976       debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
 977             screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
 978
 979    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
 980                         &screen->uniforms);
 981    if (ret) {
 982       NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
 983       goto fail;
 984    }
 985
 986    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
 987                         &screen->txc);
 988    if (ret) {
 989       NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
 990       goto fail;
 991    }
 992
 993    screen->tic.entries = CALLOC(4096, sizeof(void *));
 994    screen->tsc.entries = screen->tic.entries + 2048;
 995
 996    if (!nv50_blitter_create(screen))
 997       goto fail;
 998
 999    nv50_screen_init_hwctx(screen);
1000
1001    ret = nv50_screen_compute_setup(screen, screen->base.pushbuf);
1002    if (ret) {
1003       NOUVEAU_ERR("Failed to init compute context: %d\n", ret);
1004       goto fail;
1005    }
1006
1007    nouveau_fence_new(&screen->base, &screen->base.fence.current, false);
1008
1009    return &screen->base;
1010
1011 fail:
1012    screen->base.base.context_create = NULL;
1013    return &screen->base;
1014 }
1015
1016 int
1017 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
1018 {
1019    int i = screen->tic.next;
1020
1021    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1022       i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1023
1024    screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1025
1026    if (screen->tic.entries[i])
1027       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1028
1029    screen->tic.entries[i] = entry;
1030    return i;
1031 }
1032
1033 int
1034 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
1035 {
1036    int i = screen->tsc.next;
1037
1038    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1039       i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1040
1041    screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1042
1043    if (screen->tsc.entries[i])
1044       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1045
1046    screen->tsc.entries[i] = entry;
1047    return i;
1048 }