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    /* shared is always supported */
  79    bindings &= ~(PIPE_BIND_LINEAR |
  80                  PIPE_BIND_SHARED);
  81
  82    return (( nv50_format_table[format].usage |
  83             nv50_vertex_format[format].usage) & bindings) == bindings;
  84 }
  85
  86 static int
  87 nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
  88 {
  89    const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
  90    struct nouveau_device *dev = nouveau_screen(pscreen)->device;
  91
  92    switch (param) {
  93    /* non-boolean caps */
  94    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
  95       return 14;
  96    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
  97       return 12;
  98    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
  99       return 14;
 100    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 101       return 512;
 102    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 103    case PIPE_CAP_MIN_TEXEL_OFFSET:
 104       return -8;
 105    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 106    case PIPE_CAP_MAX_TEXEL_OFFSET:
 107       return 7;
 108    case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 109       return 128 * 1024 * 1024;
 110    case PIPE_CAP_GLSL_FEATURE_LEVEL:
 111       return 330;
 112    case PIPE_CAP_MAX_RENDER_TARGETS:
 113       return 8;
 114    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 115       return 1;
 116    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 117       return 4;
 118    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 119    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 120       return 64;
 121    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 122    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 123       return 1024;
 124    case PIPE_CAP_MAX_VERTEX_STREAMS:
 125       return 1;
 126    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 127       return 2048;
 128    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 129       return 256;
 130    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 131       return 16; /* 256 for binding as RT, but that's not possible in GL */
 132    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 133       return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
 134    case PIPE_CAP_MAX_VIEWPORTS:
 135       return NV50_MAX_VIEWPORTS;
 136    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 137       return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
 138    case PIPE_CAP_ENDIANNESS:
 139       return PIPE_ENDIAN_LITTLE;
 140    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 141       return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
 142    case PIPE_CAP_MAX_WINDOW_RECTANGLES:
 143       return NV50_MAX_WINDOW_RECTANGLES;
 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_MIXED_COLOR_DEPTH_BITS:
 152    case PIPE_CAP_ANISOTROPIC_FILTER:
 153    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 154    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 155    case PIPE_CAP_TWO_SIDED_STENCIL:
 156    case PIPE_CAP_DEPTH_CLIP_DISABLE:
 157    case PIPE_CAP_POINT_SPRITE:
 158    case PIPE_CAP_SM3:
 159    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 160    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 161    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 162    case PIPE_CAP_QUERY_TIMESTAMP:
 163    case PIPE_CAP_QUERY_TIME_ELAPSED:
 164    case PIPE_CAP_OCCLUSION_QUERY:
 165    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 166    case PIPE_CAP_INDEP_BLEND_ENABLE:
 167    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 168    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 169    case PIPE_CAP_PRIMITIVE_RESTART:
 170    case PIPE_CAP_TGSI_INSTANCEID:
 171    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 172    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 173    case PIPE_CAP_CONDITIONAL_RENDER:
 174    case PIPE_CAP_TEXTURE_BARRIER:
 175    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 176    case PIPE_CAP_START_INSTANCE:
 177    case PIPE_CAP_USER_CONSTANT_BUFFERS:
 178    case PIPE_CAP_USER_INDEX_BUFFERS:
 179    case PIPE_CAP_USER_VERTEX_BUFFERS:
 180    case PIPE_CAP_TEXTURE_MULTISAMPLE:
 181    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 182    case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 183    case PIPE_CAP_SAMPLER_VIEW_TARGET:
 184    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 185    case PIPE_CAP_CLIP_HALFZ:
 186    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 187    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 188    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 189    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 190    case PIPE_CAP_DEPTH_BOUNDS_TEST:
 191    case PIPE_CAP_TGSI_TXQS:
 192    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 193    case PIPE_CAP_SHAREABLE_SHADERS:
 194    case PIPE_CAP_CLEAR_TEXTURE:
 195    case PIPE_CAP_COMPUTE:
 196    case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 197    case PIPE_CAP_INVALIDATE_BUFFER:
 198    case PIPE_CAP_STRING_MARKER:
 199    case PIPE_CAP_CULL_DISTANCE:
 200    case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 201    case PIPE_CAP_TGSI_MUL_ZERO_WINS:
 202       return 1;
 203    case PIPE_CAP_SEAMLESS_CUBE_MAP:
 204       return 1; /* class_3d >= NVA0_3D_CLASS; */
 205    /* supported on nva0+ */
 206    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 207       return class_3d >= NVA0_3D_CLASS;
 208    /* supported on nva3+ */
 209    case PIPE_CAP_CUBE_MAP_ARRAY:
 210    case PIPE_CAP_INDEP_BLEND_FUNC:
 211    case PIPE_CAP_TEXTURE_QUERY_LOD:
 212    case PIPE_CAP_SAMPLE_SHADING:
 213    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 214       return class_3d >= NVA3_3D_CLASS;
 215
 216    /* unsupported caps */
 217    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 218    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 219    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 220    case PIPE_CAP_SHADER_STENCIL_EXPORT:
 221    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 222    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 223    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 224    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 225    case PIPE_CAP_TGSI_TEXCOORD:
 226    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 227    case PIPE_CAP_TEXTURE_GATHER_SM5:
 228    case PIPE_CAP_FAKE_SW_MSAA:
 229    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 230    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 231    case PIPE_CAP_DRAW_INDIRECT:
 232    case PIPE_CAP_MULTI_DRAW_INDIRECT:
 233    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 234    case PIPE_CAP_VERTEXID_NOBASE:
 235    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
 236    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 237    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 238    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 239    case PIPE_CAP_DRAW_PARAMETERS:
 240    case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 241    case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 242    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 243    case PIPE_CAP_GENERATE_MIPMAP:
 244    case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 245    case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 246    case PIPE_CAP_QUERY_BUFFER_OBJECT:
 247    case PIPE_CAP_QUERY_MEMORY_INFO:
 248    case PIPE_CAP_PCI_GROUP:
 249    case PIPE_CAP_PCI_BUS:
 250    case PIPE_CAP_PCI_DEVICE:
 251    case PIPE_CAP_PCI_FUNCTION:
 252    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 253    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 254    case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 255    case PIPE_CAP_TGSI_VOTE:
 256    case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 257    case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 258    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 259    case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
 260    case PIPE_CAP_NATIVE_FENCE_FD:
 261    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 262    case PIPE_CAP_TGSI_FS_FBFETCH:
 263    case PIPE_CAP_DOUBLES:
 264    case PIPE_CAP_INT64:
 265    case PIPE_CAP_INT64_DIVMOD:
 266       return 0;
 267
 268    case PIPE_CAP_VENDOR_ID:
 269       return 0x10de;
 270    case PIPE_CAP_DEVICE_ID: {
 271       uint64_t device_id;
 272       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
 273          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
 274          return -1;
 275       }
 276       return device_id;
 277    }
 278    case PIPE_CAP_ACCELERATED:
 279       return 1;
 280    case PIPE_CAP_VIDEO_MEMORY:
 281       return dev->vram_size >> 20;
 282    case PIPE_CAP_UMA:
 283       return 0;
 284    }
 285
 286    NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
 287    return 0;
 288 }
 289
 290 static int
 291 nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
 292                              enum pipe_shader_cap param)
 293 {
 294    switch (shader) {
 295    case PIPE_SHADER_VERTEX:
 296    case PIPE_SHADER_GEOMETRY:
 297    case PIPE_SHADER_FRAGMENT:
 298       break;
 299    case PIPE_SHADER_COMPUTE:
 300    default:
 301       return 0;
 302    }
 303
 304    switch (param) {
 305    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 306    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 307    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 308    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 309       return 16384;
 310    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 311       return 4;
 312    case PIPE_SHADER_CAP_MAX_INPUTS:
 313       if (shader == PIPE_SHADER_VERTEX)
 314          return 32;
 315       return 15;
 316    case PIPE_SHADER_CAP_MAX_OUTPUTS:
 317       return 16;
 318    case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 319       return 65536;
 320    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 321       return NV50_MAX_PIPE_CONSTBUFS;
 322    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 323       return shader != PIPE_SHADER_FRAGMENT;
 324    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 325    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 326    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 327       return 1;
 328    case PIPE_SHADER_CAP_MAX_PREDS:
 329       return 0;
 330    case PIPE_SHADER_CAP_MAX_TEMPS:
 331       return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
 332    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 333       return 1;
 334    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 335       return 1;
 336    case PIPE_SHADER_CAP_SUBROUTINES:
 337       return 0; /* please inline, or provide function declarations */
 338    case PIPE_SHADER_CAP_INTEGERS:
 339       return 1;
 340    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 341       /* The chip could handle more sampler views than samplers */
 342    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 343       return MIN2(16, PIPE_MAX_SAMPLERS);
 344    case PIPE_SHADER_CAP_PREFERRED_IR:
 345       return PIPE_SHADER_IR_TGSI;
 346    case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 347       return 32;
 348    case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 349    case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 350    case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 351    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 352    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 353    case PIPE_SHADER_CAP_SUPPORTED_IRS:
 354    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 355    case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
 356       return 0;
 357    default:
 358       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
 359       return 0;
 360    }
 361 }
 362
 363 static float
 364 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
 365 {
 366    switch (param) {
 367    case PIPE_CAPF_MAX_LINE_WIDTH:
 368    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 369       return 10.0f;
 370    case PIPE_CAPF_MAX_POINT_WIDTH:
 371    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 372       return 64.0f;
 373    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 374       return 16.0f;
 375    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 376       return 4.0f;
 377    case PIPE_CAPF_GUARD_BAND_LEFT:
 378    case PIPE_CAPF_GUARD_BAND_TOP:
 379       return 0.0f;
 380    case PIPE_CAPF_GUARD_BAND_RIGHT:
 381    case PIPE_CAPF_GUARD_BAND_BOTTOM:
 382       return 0.0f; /* that or infinity */
 383    }
 384
 385    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
 386    return 0.0f;
 387 }
 388
 389 static int
 390 nv50_screen_get_compute_param(struct pipe_screen *pscreen,
 391                               enum pipe_shader_ir ir_type,
 392                               enum pipe_compute_cap param, void *data)
 393 {
 394    struct nv50_screen *screen = nv50_screen(pscreen);
 395
 396 #define RET(x) do {                  \
 397    if (data)                         \
 398       memcpy(data, x, sizeof(x));    \
 399    return sizeof(x);                 \
 400 } while (0)
 401
 402    switch (param) {
 403    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
 404       RET((uint64_t []) { 2 });
 405    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
 406       RET(((uint64_t []) { 65535, 65535 }));
 407    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
 408       RET(((uint64_t []) { 512, 512, 64 }));
 409    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
 410       RET((uint64_t []) { 512 });
 411    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g0-15[] */
 412       RET((uint64_t []) { 1ULL << 32 });
 413    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
 414       RET((uint64_t []) { 16 << 10 });
 415    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
 416       RET((uint64_t []) { 16 << 10 });
 417    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
 418       RET((uint64_t []) { 4096 });
 419    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
 420       RET((uint32_t []) { 32 });
 421    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
 422       RET((uint64_t []) { 1ULL << 40 });
 423    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
 424       RET((uint32_t []) { 0 });
 425    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
 426       RET((uint32_t []) { screen->mp_count });
 427    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
 428       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
 429    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
 430       RET((uint32_t []) { 32 });
 431    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
 432       RET((uint64_t []) { 0 });
 433    default:
 434       return 0;
 435    }
 436
 437 #undef RET
 438 }
 439
 440 static void
 441 nv50_screen_destroy(struct pipe_screen *pscreen)
 442 {
 443    struct nv50_screen *screen = nv50_screen(pscreen);
 444
 445    if (!nouveau_drm_screen_unref(&screen->base))
 446       return;
 447
 448    if (screen->base.fence.current) {
 449       struct nouveau_fence *current = NULL;
 450
 451       /* nouveau_fence_wait will create a new current fence, so wait on the
 452        * _current_ one, and remove both.
 453        */
 454       nouveau_fence_ref(screen->base.fence.current, &current);
 455       nouveau_fence_wait(current, NULL);
 456       nouveau_fence_ref(NULL, &current);
 457       nouveau_fence_ref(NULL, &screen->base.fence.current);
 458    }
 459    if (screen->base.pushbuf)
 460       screen->base.pushbuf->user_priv = NULL;
 461
 462    if (screen->blitter)
 463       nv50_blitter_destroy(screen);
 464    if (screen->pm.prog) {
 465       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
 466       nv50_program_destroy(NULL, screen->pm.prog);
 467       FREE(screen->pm.prog);
 468    }
 469
 470    nouveau_bo_ref(NULL, &screen->code);
 471    nouveau_bo_ref(NULL, &screen->tls_bo);
 472    nouveau_bo_ref(NULL, &screen->stack_bo);
 473    nouveau_bo_ref(NULL, &screen->txc);
 474    nouveau_bo_ref(NULL, &screen->uniforms);
 475    nouveau_bo_ref(NULL, &screen->fence.bo);
 476
 477    nouveau_heap_destroy(&screen->vp_code_heap);
 478    nouveau_heap_destroy(&screen->gp_code_heap);
 479    nouveau_heap_destroy(&screen->fp_code_heap);
 480
 481    FREE(screen->tic.entries);
 482
 483    nouveau_object_del(&screen->tesla);
 484    nouveau_object_del(&screen->eng2d);
 485    nouveau_object_del(&screen->m2mf);
 486    nouveau_object_del(&screen->compute);
 487    nouveau_object_del(&screen->sync);
 488
 489    nouveau_screen_fini(&screen->base);
 490
 491    FREE(screen);
 492 }
 493
 494 static void
 495 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
 496 {
 497    struct nv50_screen *screen = nv50_screen(pscreen);
 498    struct nouveau_pushbuf *push = screen->base.pushbuf;
 499
 500    /* we need to do it after possible flush in MARK_RING */
 501    *sequence = ++screen->base.fence.sequence;
 502
 503    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
 504    PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
 505    PUSH_DATAh(push, screen->fence.bo->offset);
 506    PUSH_DATA (push, screen->fence.bo->offset);
 507    PUSH_DATA (push, *sequence);
 508    PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
 509                     NV50_3D_QUERY_GET_UNK4 |
 510                     NV50_3D_QUERY_GET_UNIT_CROP |
 511                     NV50_3D_QUERY_GET_TYPE_QUERY |
 512                     NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
 513                     NV50_3D_QUERY_GET_SHORT);
 514 }
 515
 516 static u32
 517 nv50_screen_fence_update(struct pipe_screen *pscreen)
 518 {
 519    return nv50_screen(pscreen)->fence.map[0];
 520 }
 521
 522 static void
 523 nv50_screen_init_hwctx(struct nv50_screen *screen)
 524 {
 525    struct nouveau_pushbuf *push = screen->base.pushbuf;
 526    struct nv04_fifo *fifo;
 527    unsigned i;
 528
 529    fifo = (struct nv04_fifo *)screen->base.channel->data;
 530
 531    BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
 532    PUSH_DATA (push, screen->m2mf->handle);
 533    BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
 534    PUSH_DATA (push, screen->sync->handle);
 535    PUSH_DATA (push, fifo->vram);
 536    PUSH_DATA (push, fifo->vram);
 537
 538    BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
 539    PUSH_DATA (push, screen->eng2d->handle);
 540    BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
 541    PUSH_DATA (push, screen->sync->handle);
 542    PUSH_DATA (push, fifo->vram);
 543    PUSH_DATA (push, fifo->vram);
 544    PUSH_DATA (push, fifo->vram);
 545    BEGIN_NV04(push, NV50_2D(OPERATION), 1);
 546    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
 547    BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
 548    PUSH_DATA (push, 0);
 549    BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
 550    PUSH_DATA (push, 0);
 551    BEGIN_NV04(push, SUBC_2D(0x0888), 1);
 552    PUSH_DATA (push, 1);
 553    BEGIN_NV04(push, NV50_2D(COND_MODE), 1);
 554    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
 555
 556    BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
 557    PUSH_DATA (push, screen->tesla->handle);
 558
 559    BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
 560    PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
 561
 562    BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
 563    PUSH_DATA (push, screen->sync->handle);
 564    BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
 565    for (i = 0; i < 11; ++i)
 566       PUSH_DATA(push, fifo->vram);
 567    BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
 568    for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
 569       PUSH_DATA(push, fifo->vram);
 570
 571    BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
 572    PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
 573    BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
 574    PUSH_DATA (push, 0xf);
 575
 576    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
 577       BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
 578       PUSH_DATA (push, 0x18);
 579    }
 580
 581    BEGIN_NV04(push, NV50_3D(ZETA_COMP_ENABLE), 1);
 582    PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 583
 584    BEGIN_NV04(push, NV50_3D(RT_COMP_ENABLE(0)), 8);
 585    for (i = 0; i < 8; ++i)
 586       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 587
 588    BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
 589    PUSH_DATA (push, 1);
 590
 591    BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
 592    PUSH_DATA (push, 0);
 593    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
 594    PUSH_DATA (push, 0);
 595    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
 596    PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
 597    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
 598    PUSH_DATA (push, 0);
 599    BEGIN_NV04(push, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
 600    PUSH_DATA (push, 1);
 601    BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
 602    PUSH_DATA (push, 1);
 603
 604    if (screen->tesla->oclass >= NVA0_3D_CLASS) {
 605       BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
 606       PUSH_DATA (push, 0);
 607    }
 608
 609    BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
 610    PUSH_DATA (push, 0);
 611    BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
 612    PUSH_DATA (push, 0);
 613    PUSH_DATA (push, 0);
 614    BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
 615    PUSH_DATA (push, 0x3f);
 616
 617    BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
 618    PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 619    PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 620
 621    BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
 622    PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 623    PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 624
 625    BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
 626    PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 627    PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 628
 629    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 630    PUSH_DATAh(push, screen->tls_bo->offset);
 631    PUSH_DATA (push, screen->tls_bo->offset);
 632    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 633
 634    BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
 635    PUSH_DATAh(push, screen->stack_bo->offset);
 636    PUSH_DATA (push, screen->stack_bo->offset);
 637    PUSH_DATA (push, 4);
 638
 639    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 640    PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
 641    PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
 642    PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
 643
 644    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 645    PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
 646    PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
 647    PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
 648
 649    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 650    PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
 651    PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
 652    PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
 653
 654    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 655    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
 656    PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
 657    PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
 658
 659    BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
 660    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
 661    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
 662    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
 663
 664    /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
 665    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
 666    PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
 667    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
 668    PUSH_DATAf(push, 0.0f);
 669    PUSH_DATAf(push, 0.0f);
 670    PUSH_DATAf(push, 0.0f);
 671    PUSH_DATAf(push, 0.0f);
 672    BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
 673    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 674    PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 675
 676    nv50_upload_ms_info(push);
 677
 678    /* max TIC (bits 4:8) & TSC bindings, per program type */
 679    for (i = 0; i < 3; ++i) {
 680       BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
 681       PUSH_DATA (push, 0x54);
 682    }
 683
 684    BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
 685    PUSH_DATAh(push, screen->txc->offset);
 686    PUSH_DATA (push, screen->txc->offset);
 687    PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
 688
 689    BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
 690    PUSH_DATAh(push, screen->txc->offset + 65536);
 691    PUSH_DATA (push, screen->txc->offset + 65536);
 692    PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
 693
 694    BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
 695    PUSH_DATA (push, 0);
 696
 697    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
 698    PUSH_DATA (push, 0);
 699    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
 700    PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
 701    BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
 702    for (i = 0; i < 8 * 2; ++i)
 703       PUSH_DATA(push, 0);
 704    BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
 705    PUSH_DATA (push, 0);
 706
 707    BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
 708    PUSH_DATA (push, 1);
 709    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 710       BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(i)), 2);
 711       PUSH_DATAf(push, 0.0f);
 712       PUSH_DATAf(push, 1.0f);
 713       BEGIN_NV04(push, NV50_3D(VIEWPORT_HORIZ(i)), 2);
 714       PUSH_DATA (push, 8192 << 16);
 715       PUSH_DATA (push, 8192 << 16);
 716    }
 717
 718    BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
 719 #ifdef NV50_SCISSORS_CLIPPING
 720    PUSH_DATA (push, 0x0000);
 721 #else
 722    PUSH_DATA (push, 0x1080);
 723 #endif
 724
 725    BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
 726    PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
 727
 728    /* We use scissors instead of exact view volume clipping,
 729     * so they're always enabled.
 730     */
 731    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 732       BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(i)), 3);
 733       PUSH_DATA (push, 1);
 734       PUSH_DATA (push, 8192 << 16);
 735       PUSH_DATA (push, 8192 << 16);
 736    }
 737
 738    BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
 739    PUSH_DATA (push, 1);
 740    BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
 741    PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
 742    BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
 743    PUSH_DATA (push, 0x11111111);
 744    BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
 745    PUSH_DATA (push, 1);
 746
 747    BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
 748    PUSH_DATA (push, 0);
 749    if (screen->base.class_3d >= NV84_3D_CLASS) {
 750       BEGIN_NV04(push, NV84_3D(VERTEX_ID_BASE), 1);
 751       PUSH_DATA (push, 0);
 752    }
 753
 754    PUSH_KICK (push);
 755 }
 756
 757 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
 758       uint64_t *tls_size)
 759 {
 760    struct nouveau_device *dev = screen->base.device;
 761    int ret;
 762
 763    screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
 764          ONE_TEMP_SIZE;
 765    if (nouveau_mesa_debug)
 766       debug_printf("allocating space for %u temps\n",
 767             util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
 768    *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
 769          screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
 770
 771    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 772                         *tls_size, NULL, &screen->tls_bo);
 773    if (ret) {
 774       NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
 775       return ret;
 776    }
 777
 778    return 0;
 779 }
 780
 781 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
 782 {
 783    struct nouveau_pushbuf *push = screen->base.pushbuf;
 784    int ret;
 785    uint64_t tls_size;
 786
 787    if (tls_space < screen->cur_tls_space)
 788       return 0;
 789    if (tls_space > screen->max_tls_space) {
 790       /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
 791        * LOCAL_WARPS_NO_CLAMP) */
 792       NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
 793             (unsigned)(tls_space / ONE_TEMP_SIZE),
 794             (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
 795       return -ENOMEM;
 796    }
 797
 798    nouveau_bo_ref(NULL, &screen->tls_bo);
 799    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
 800    if (ret)
 801       return ret;
 802
 803    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 804    PUSH_DATAh(push, screen->tls_bo->offset);
 805    PUSH_DATA (push, screen->tls_bo->offset);
 806    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 807
 808    return 1;
 809 }
 810
 811 struct nouveau_screen *
 812 nv50_screen_create(struct nouveau_device *dev)
 813 {
 814    struct nv50_screen *screen;
 815    struct pipe_screen *pscreen;
 816    struct nouveau_object *chan;
 817    uint64_t value;
 818    uint32_t tesla_class;
 819    unsigned stack_size;
 820    int ret;
 821
 822    screen = CALLOC_STRUCT(nv50_screen);
 823    if (!screen)
 824       return NULL;
 825    pscreen = &screen->base.base;
 826    pscreen->destroy = nv50_screen_destroy;
 827
 828    ret = nouveau_screen_init(&screen->base, dev);
 829    if (ret) {
 830       NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
 831       goto fail;
 832    }
 833
 834    /* TODO: Prevent FIFO prefetch before transfer of index buffers and
 835     *  admit them to VRAM.
 836     */
 837    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
 838       PIPE_BIND_VERTEX_BUFFER;
 839    screen->base.sysmem_bindings |=
 840       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
 841
 842    screen->base.pushbuf->user_priv = screen;
 843    screen->base.pushbuf->rsvd_kick = 5;
 844
 845    chan = screen->base.channel;
 846
 847    pscreen->context_create = nv50_create;
 848    pscreen->is_format_supported = nv50_screen_is_format_supported;
 849    pscreen->get_param = nv50_screen_get_param;
 850    pscreen->get_shader_param = nv50_screen_get_shader_param;
 851    pscreen->get_paramf = nv50_screen_get_paramf;
 852    pscreen->get_compute_param = nv50_screen_get_compute_param;
 853    pscreen->get_driver_query_info = nv50_screen_get_driver_query_info;
 854    pscreen->get_driver_query_group_info = nv50_screen_get_driver_query_group_info;
 855
 856    nv50_screen_init_resource_functions(pscreen);
 857
 858    if (screen->base.device->chipset < 0x84 ||
 859        debug_get_bool_option("NOUVEAU_PMPEG", false)) {
 860       /* PMPEG */
 861       nouveau_screen_init_vdec(&screen->base);
 862    } else if (screen->base.device->chipset < 0x98 ||
 863               screen->base.device->chipset == 0xa0) {
 864       /* VP2 */
 865       screen->base.base.get_video_param = nv84_screen_get_video_param;
 866       screen->base.base.is_video_format_supported = nv84_screen_video_supported;
 867    } else {
 868       /* VP3/4 */
 869       screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
 870       screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
 871    }
 872
 873    ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
 874                         NULL, &screen->fence.bo);
 875    if (ret) {
 876       NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
 877       goto fail;
 878    }
 879
 880    nouveau_bo_map(screen->fence.bo, 0, NULL);
 881    screen->fence.map = screen->fence.bo->map;
 882    screen->base.fence.emit = nv50_screen_fence_emit;
 883    screen->base.fence.update = nv50_screen_fence_update;
 884
 885    ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
 886                             &(struct nv04_notify){ .length = 32 },
 887                             sizeof(struct nv04_notify), &screen->sync);
 888    if (ret) {
 889       NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
 890       goto fail;
 891    }
 892
 893    ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
 894                             NULL, 0, &screen->m2mf);
 895    if (ret) {
 896       NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
 897       goto fail;
 898    }
 899
 900    ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
 901                             NULL, 0, &screen->eng2d);
 902    if (ret) {
 903       NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
 904       goto fail;
 905    }
 906
 907    switch (dev->chipset & 0xf0) {
 908    case 0x50:
 909       tesla_class = NV50_3D_CLASS;
 910       break;
 911    case 0x80:
 912    case 0x90:
 913       tesla_class = NV84_3D_CLASS;
 914       break;
 915    case 0xa0:
 916       switch (dev->chipset) {
 917       case 0xa0:
 918       case 0xaa:
 919       case 0xac:
 920          tesla_class = NVA0_3D_CLASS;
 921          break;
 922       case 0xaf:
 923          tesla_class = NVAF_3D_CLASS;
 924          break;
 925       default:
 926          tesla_class = NVA3_3D_CLASS;
 927          break;
 928       }
 929       break;
 930    default:
 931       NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
 932       goto fail;
 933    }
 934    screen->base.class_3d = tesla_class;
 935
 936    ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
 937                             NULL, 0, &screen->tesla);
 938    if (ret) {
 939       NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
 940       goto fail;
 941    }
 942
 943    /* This over-allocates by a page. The GP, which would execute at the end of
 944     * the last page, would trigger faults. The going theory is that it
 945     * prefetches up to a certain amount.
 946     */
 947    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 948                         (3 << NV50_CODE_BO_SIZE_LOG2) + 0x1000,
 949                         NULL, &screen->code);
 950    if (ret) {
 951       NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
 952       goto fail;
 953    }
 954
 955    nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 956    nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 957    nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 958
 959    nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
 960
 961    screen->TPs = util_bitcount(value & 0xffff);
 962    screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
 963
 964    screen->mp_count = screen->TPs * screen->MPsInTP;
 965
 966    stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
 967          STACK_WARPS_ALLOC * 64 * 8;
 968
 969    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
 970                         &screen->stack_bo);
 971    if (ret) {
 972       NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
 973       goto fail;
 974    }
 975
 976    uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
 977          screen->MPsInTP * LOCAL_WARPS_ALLOC *  THREADS_IN_WARP *
 978          ONE_TEMP_SIZE;
 979    screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
 980    screen->max_tls_space /= 2; /* half of vram */
 981
 982    /* hw can address max 64 KiB */
 983    screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
 984
 985    uint64_t tls_size;
 986    unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
 987    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
 988    if (ret)
 989       goto fail;
 990
 991    if (nouveau_mesa_debug)
 992       debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
 993             screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
 994
 995    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
 996                         &screen->uniforms);
 997    if (ret) {
 998       NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
 999       goto fail;
1000    }
1001
1002    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
1003                         &screen->txc);
1004    if (ret) {
1005       NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
1006       goto fail;
1007    }
1008
1009    screen->tic.entries = CALLOC(4096, sizeof(void *));
1010    screen->tsc.entries = screen->tic.entries + 2048;
1011
1012    if (!nv50_blitter_create(screen))
1013       goto fail;
1014
1015    nv50_screen_init_hwctx(screen);
1016
1017    ret = nv50_screen_compute_setup(screen, screen->base.pushbuf);
1018    if (ret) {
1019       NOUVEAU_ERR("Failed to init compute context: %d\n", ret);
1020       goto fail;
1021    }
1022
1023    nouveau_fence_new(&screen->base, &screen->base.fence.current);
1024
1025    return &screen->base;
1026
1027 fail:
1028    screen->base.base.context_create = NULL;
1029    return &screen->base;
1030 }
1031
1032 int
1033 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
1034 {
1035    int i = screen->tic.next;
1036
1037    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1038       i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1039
1040    screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1041
1042    if (screen->tic.entries[i])
1043       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1044
1045    screen->tic.entries[i] = entry;
1046    return i;
1047 }
1048
1049 int
1050 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
1051 {
1052    int i = screen->tsc.next;
1053
1054    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1055       i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1056
1057    screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1058
1059    if (screen->tsc.entries[i])
1060       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1061
1062    screen->tsc.entries[i] = entry;
1063    return i;
1064 }