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_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    case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 200    case PIPE_CAP_TGSI_MUL_ZERO_WINS:
 201    case PIPE_CAP_TGSI_TEX_TXF_LZ:
 202    case PIPE_CAP_TGSI_CLOCK:
 203    case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
 204    case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 205       return 1;
 206    case PIPE_CAP_SEAMLESS_CUBE_MAP:
 207       return 1; /* class_3d >= NVA0_3D_CLASS; */
 208    /* supported on nva0+ */
 209    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 210       return class_3d >= NVA0_3D_CLASS;
 211    /* supported on nva3+ */
 212    case PIPE_CAP_CUBE_MAP_ARRAY:
 213    case PIPE_CAP_INDEP_BLEND_FUNC:
 214    case PIPE_CAP_TEXTURE_QUERY_LOD:
 215    case PIPE_CAP_SAMPLE_SHADING:
 216    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 217       return class_3d >= NVA3_3D_CLASS;
 218
 219    /* unsupported caps */
 220    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 221    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 222    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 223    case PIPE_CAP_SHADER_STENCIL_EXPORT:
 224    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 225    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 226    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 227    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 228    case PIPE_CAP_TGSI_TEXCOORD:
 229    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 230    case PIPE_CAP_TEXTURE_GATHER_SM5:
 231    case PIPE_CAP_FAKE_SW_MSAA:
 232    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 233    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 234    case PIPE_CAP_DRAW_INDIRECT:
 235    case PIPE_CAP_MULTI_DRAW_INDIRECT:
 236    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 237    case PIPE_CAP_VERTEXID_NOBASE:
 238    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
 239    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 240    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 241    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 242    case PIPE_CAP_DRAW_PARAMETERS:
 243    case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 244    case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 245    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 246    case PIPE_CAP_GENERATE_MIPMAP:
 247    case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 248    case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 249    case PIPE_CAP_QUERY_BUFFER_OBJECT:
 250    case PIPE_CAP_QUERY_MEMORY_INFO:
 251    case PIPE_CAP_PCI_GROUP:
 252    case PIPE_CAP_PCI_BUS:
 253    case PIPE_CAP_PCI_DEVICE:
 254    case PIPE_CAP_PCI_FUNCTION:
 255    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 256    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 257    case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 258    case PIPE_CAP_TGSI_VOTE:
 259    case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 260    case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 261    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 262    case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
 263    case PIPE_CAP_NATIVE_FENCE_FD:
 264    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 265    case PIPE_CAP_TGSI_FS_FBFETCH:
 266    case PIPE_CAP_DOUBLES:
 267    case PIPE_CAP_INT64:
 268    case PIPE_CAP_INT64_DIVMOD:
 269    case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
 270    case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
 271    case PIPE_CAP_TGSI_BALLOT:
 272    case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
 273    case PIPE_CAP_POST_DEPTH_COVERAGE:
 274    case PIPE_CAP_BINDLESS_TEXTURE:
 275    case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
 276    case PIPE_CAP_QUERY_SO_OVERFLOW:
 277    case PIPE_CAP_MEMOBJ:
 278    case PIPE_CAP_LOAD_CONSTBUF:
 279    case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
 280       return 0;
 281
 282    case PIPE_CAP_VENDOR_ID:
 283       return 0x10de;
 284    case PIPE_CAP_DEVICE_ID: {
 285       uint64_t device_id;
 286       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
 287          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
 288          return -1;
 289       }
 290       return device_id;
 291    }
 292    case PIPE_CAP_ACCELERATED:
 293       return 1;
 294    case PIPE_CAP_VIDEO_MEMORY:
 295       return dev->vram_size >> 20;
 296    case PIPE_CAP_UMA:
 297       return 0;
 298    }
 299
 300    NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
 301    return 0;
 302 }
 303
 304 static int
 305 nv50_screen_get_shader_param(struct pipe_screen *pscreen,
 306                              enum pipe_shader_type shader,
 307                              enum pipe_shader_cap param)
 308 {
 309    switch (shader) {
 310    case PIPE_SHADER_VERTEX:
 311    case PIPE_SHADER_GEOMETRY:
 312    case PIPE_SHADER_FRAGMENT:
 313       break;
 314    case PIPE_SHADER_COMPUTE:
 315    default:
 316       return 0;
 317    }
 318
 319    switch (param) {
 320    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 321    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 322    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 323    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 324       return 16384;
 325    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 326       return 4;
 327    case PIPE_SHADER_CAP_MAX_INPUTS:
 328       if (shader == PIPE_SHADER_VERTEX)
 329          return 32;
 330       return 15;
 331    case PIPE_SHADER_CAP_MAX_OUTPUTS:
 332       return 16;
 333    case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 334       return 65536;
 335    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 336       return NV50_MAX_PIPE_CONSTBUFS;
 337    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 338       return shader != PIPE_SHADER_FRAGMENT;
 339    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 340    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 341    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 342       return 1;
 343    case PIPE_SHADER_CAP_MAX_TEMPS:
 344       return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
 345    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 346       return 1;
 347    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 348       return 1;
 349    case PIPE_SHADER_CAP_INT64_ATOMICS:
 350    case PIPE_SHADER_CAP_FP16:
 351    case PIPE_SHADER_CAP_SUBROUTINES:
 352       return 0; /* please inline, or provide function declarations */
 353    case PIPE_SHADER_CAP_INTEGERS:
 354       return 1;
 355    case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
 356       return 1;
 357    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 358       /* The chip could handle more sampler views than samplers */
 359    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 360       return MIN2(16, PIPE_MAX_SAMPLERS);
 361    case PIPE_SHADER_CAP_PREFERRED_IR:
 362       return PIPE_SHADER_IR_TGSI;
 363    case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 364       return 32;
 365    case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 366    case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 367    case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
 368    case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 369    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 370    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 371    case PIPE_SHADER_CAP_SUPPORTED_IRS:
 372    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 373    case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
 374       return 0;
 375    default:
 376       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
 377       return 0;
 378    }
 379 }
 380
 381 static float
 382 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
 383 {
 384    switch (param) {
 385    case PIPE_CAPF_MAX_LINE_WIDTH:
 386    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 387       return 10.0f;
 388    case PIPE_CAPF_MAX_POINT_WIDTH:
 389    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 390       return 64.0f;
 391    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 392       return 16.0f;
 393    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 394       return 4.0f;
 395    case PIPE_CAPF_GUARD_BAND_LEFT:
 396    case PIPE_CAPF_GUARD_BAND_TOP:
 397       return 0.0f;
 398    case PIPE_CAPF_GUARD_BAND_RIGHT:
 399    case PIPE_CAPF_GUARD_BAND_BOTTOM:
 400       return 0.0f; /* that or infinity */
 401    }
 402
 403    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
 404    return 0.0f;
 405 }
 406
 407 static int
 408 nv50_screen_get_compute_param(struct pipe_screen *pscreen,
 409                               enum pipe_shader_ir ir_type,
 410                               enum pipe_compute_cap param, void *data)
 411 {
 412    struct nv50_screen *screen = nv50_screen(pscreen);
 413
 414 #define RET(x) do {                  \
 415    if (data)                         \
 416       memcpy(data, x, sizeof(x));    \
 417    return sizeof(x);                 \
 418 } while (0)
 419
 420    switch (param) {
 421    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
 422       RET((uint64_t []) { 2 });
 423    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
 424       RET(((uint64_t []) { 65535, 65535 }));
 425    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
 426       RET(((uint64_t []) { 512, 512, 64 }));
 427    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
 428       RET((uint64_t []) { 512 });
 429    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g0-15[] */
 430       RET((uint64_t []) { 1ULL << 32 });
 431    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
 432       RET((uint64_t []) { 16 << 10 });
 433    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
 434       RET((uint64_t []) { 16 << 10 });
 435    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
 436       RET((uint64_t []) { 4096 });
 437    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
 438       RET((uint32_t []) { 32 });
 439    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
 440       RET((uint64_t []) { 1ULL << 40 });
 441    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
 442       RET((uint32_t []) { 0 });
 443    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
 444       RET((uint32_t []) { screen->mp_count });
 445    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
 446       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
 447    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
 448       RET((uint32_t []) { 32 });
 449    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
 450       RET((uint64_t []) { 0 });
 451    default:
 452       return 0;
 453    }
 454
 455 #undef RET
 456 }
 457
 458 static void
 459 nv50_screen_destroy(struct pipe_screen *pscreen)
 460 {
 461    struct nv50_screen *screen = nv50_screen(pscreen);
 462
 463    if (!nouveau_drm_screen_unref(&screen->base))
 464       return;
 465
 466    if (screen->base.fence.current) {
 467       struct nouveau_fence *current = NULL;
 468
 469       /* nouveau_fence_wait will create a new current fence, so wait on the
 470        * _current_ one, and remove both.
 471        */
 472       nouveau_fence_ref(screen->base.fence.current, &current);
 473       nouveau_fence_wait(current, NULL);
 474       nouveau_fence_ref(NULL, &current);
 475       nouveau_fence_ref(NULL, &screen->base.fence.current);
 476    }
 477    if (screen->base.pushbuf)
 478       screen->base.pushbuf->user_priv = NULL;
 479
 480    if (screen->blitter)
 481       nv50_blitter_destroy(screen);
 482    if (screen->pm.prog) {
 483       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
 484       nv50_program_destroy(NULL, screen->pm.prog);
 485       FREE(screen->pm.prog);
 486    }
 487
 488    nouveau_bo_ref(NULL, &screen->code);
 489    nouveau_bo_ref(NULL, &screen->tls_bo);
 490    nouveau_bo_ref(NULL, &screen->stack_bo);
 491    nouveau_bo_ref(NULL, &screen->txc);
 492    nouveau_bo_ref(NULL, &screen->uniforms);
 493    nouveau_bo_ref(NULL, &screen->fence.bo);
 494
 495    nouveau_heap_destroy(&screen->vp_code_heap);
 496    nouveau_heap_destroy(&screen->gp_code_heap);
 497    nouveau_heap_destroy(&screen->fp_code_heap);
 498
 499    FREE(screen->tic.entries);
 500
 501    nouveau_object_del(&screen->tesla);
 502    nouveau_object_del(&screen->eng2d);
 503    nouveau_object_del(&screen->m2mf);
 504    nouveau_object_del(&screen->compute);
 505    nouveau_object_del(&screen->sync);
 506
 507    nouveau_screen_fini(&screen->base);
 508
 509    FREE(screen);
 510 }
 511
 512 static void
 513 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
 514 {
 515    struct nv50_screen *screen = nv50_screen(pscreen);
 516    struct nouveau_pushbuf *push = screen->base.pushbuf;
 517
 518    /* we need to do it after possible flush in MARK_RING */
 519    *sequence = ++screen->base.fence.sequence;
 520
 521    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
 522    PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
 523    PUSH_DATAh(push, screen->fence.bo->offset);
 524    PUSH_DATA (push, screen->fence.bo->offset);
 525    PUSH_DATA (push, *sequence);
 526    PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
 527                     NV50_3D_QUERY_GET_UNK4 |
 528                     NV50_3D_QUERY_GET_UNIT_CROP |
 529                     NV50_3D_QUERY_GET_TYPE_QUERY |
 530                     NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
 531                     NV50_3D_QUERY_GET_SHORT);
 532 }
 533
 534 static u32
 535 nv50_screen_fence_update(struct pipe_screen *pscreen)
 536 {
 537    return nv50_screen(pscreen)->fence.map[0];
 538 }
 539
 540 static void
 541 nv50_screen_init_hwctx(struct nv50_screen *screen)
 542 {
 543    struct nouveau_pushbuf *push = screen->base.pushbuf;
 544    struct nv04_fifo *fifo;
 545    unsigned i;
 546
 547    fifo = (struct nv04_fifo *)screen->base.channel->data;
 548
 549    BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
 550    PUSH_DATA (push, screen->m2mf->handle);
 551    BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
 552    PUSH_DATA (push, screen->sync->handle);
 553    PUSH_DATA (push, fifo->vram);
 554    PUSH_DATA (push, fifo->vram);
 555
 556    BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
 557    PUSH_DATA (push, screen->eng2d->handle);
 558    BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
 559    PUSH_DATA (push, screen->sync->handle);
 560    PUSH_DATA (push, fifo->vram);
 561    PUSH_DATA (push, fifo->vram);
 562    PUSH_DATA (push, fifo->vram);
 563    BEGIN_NV04(push, NV50_2D(OPERATION), 1);
 564    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
 565    BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
 566    PUSH_DATA (push, 0);
 567    BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
 568    PUSH_DATA (push, 0);
 569    BEGIN_NV04(push, SUBC_2D(0x0888), 1);
 570    PUSH_DATA (push, 1);
 571    BEGIN_NV04(push, NV50_2D(COND_MODE), 1);
 572    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
 573
 574    BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
 575    PUSH_DATA (push, screen->tesla->handle);
 576
 577    BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
 578    PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
 579
 580    BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
 581    PUSH_DATA (push, screen->sync->handle);
 582    BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
 583    for (i = 0; i < 11; ++i)
 584       PUSH_DATA(push, fifo->vram);
 585    BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
 586    for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
 587       PUSH_DATA(push, fifo->vram);
 588
 589    BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
 590    PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
 591    BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
 592    PUSH_DATA (push, 0xf);
 593
 594    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
 595       BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
 596       PUSH_DATA (push, 0x18);
 597    }
 598
 599    BEGIN_NV04(push, NV50_3D(ZETA_COMP_ENABLE), 1);
 600    PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 601
 602    BEGIN_NV04(push, NV50_3D(RT_COMP_ENABLE(0)), 8);
 603    for (i = 0; i < 8; ++i)
 604       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 605
 606    BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
 607    PUSH_DATA (push, 1);
 608
 609    BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
 610    PUSH_DATA (push, 0);
 611    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
 612    PUSH_DATA (push, 0);
 613    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
 614    PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
 615    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
 616    PUSH_DATA (push, 0);
 617    BEGIN_NV04(push, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
 618    PUSH_DATA (push, 1);
 619    BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
 620    PUSH_DATA (push, 1);
 621
 622    if (screen->tesla->oclass >= NVA0_3D_CLASS) {
 623       BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
 624       PUSH_DATA (push, 0);
 625    }
 626
 627    BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
 628    PUSH_DATA (push, 0);
 629    BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
 630    PUSH_DATA (push, 0);
 631    PUSH_DATA (push, 0);
 632    BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
 633    PUSH_DATA (push, 0x3f);
 634
 635    BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
 636    PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 637    PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 638
 639    BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
 640    PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 641    PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 642
 643    BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
 644    PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 645    PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 646
 647    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 648    PUSH_DATAh(push, screen->tls_bo->offset);
 649    PUSH_DATA (push, screen->tls_bo->offset);
 650    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 651
 652    BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
 653    PUSH_DATAh(push, screen->stack_bo->offset);
 654    PUSH_DATA (push, screen->stack_bo->offset);
 655    PUSH_DATA (push, 4);
 656
 657    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 658    PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
 659    PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
 660    PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
 661
 662    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 663    PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
 664    PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
 665    PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
 666
 667    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 668    PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
 669    PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
 670    PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
 671
 672    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 673    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
 674    PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
 675    PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
 676
 677    BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
 678    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
 679    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
 680    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
 681
 682    /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
 683    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
 684    PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
 685    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
 686    PUSH_DATAf(push, 0.0f);
 687    PUSH_DATAf(push, 0.0f);
 688    PUSH_DATAf(push, 0.0f);
 689    PUSH_DATAf(push, 0.0f);
 690    BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
 691    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 692    PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 693
 694    nv50_upload_ms_info(push);
 695
 696    /* max TIC (bits 4:8) & TSC bindings, per program type */
 697    for (i = 0; i < 3; ++i) {
 698       BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
 699       PUSH_DATA (push, 0x54);
 700    }
 701
 702    BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
 703    PUSH_DATAh(push, screen->txc->offset);
 704    PUSH_DATA (push, screen->txc->offset);
 705    PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
 706
 707    BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
 708    PUSH_DATAh(push, screen->txc->offset + 65536);
 709    PUSH_DATA (push, screen->txc->offset + 65536);
 710    PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
 711
 712    BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
 713    PUSH_DATA (push, 0);
 714
 715    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
 716    PUSH_DATA (push, 0);
 717    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
 718    PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
 719    BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
 720    for (i = 0; i < 8 * 2; ++i)
 721       PUSH_DATA(push, 0);
 722    BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
 723    PUSH_DATA (push, 0);
 724
 725    BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
 726    PUSH_DATA (push, 1);
 727    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 728       BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(i)), 2);
 729       PUSH_DATAf(push, 0.0f);
 730       PUSH_DATAf(push, 1.0f);
 731       BEGIN_NV04(push, NV50_3D(VIEWPORT_HORIZ(i)), 2);
 732       PUSH_DATA (push, 8192 << 16);
 733       PUSH_DATA (push, 8192 << 16);
 734    }
 735
 736    BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
 737 #ifdef NV50_SCISSORS_CLIPPING
 738    PUSH_DATA (push, 0x0000);
 739 #else
 740    PUSH_DATA (push, 0x1080);
 741 #endif
 742
 743    BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
 744    PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
 745
 746    /* We use scissors instead of exact view volume clipping,
 747     * so they're always enabled.
 748     */
 749    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 750       BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(i)), 3);
 751       PUSH_DATA (push, 1);
 752       PUSH_DATA (push, 8192 << 16);
 753       PUSH_DATA (push, 8192 << 16);
 754    }
 755
 756    BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
 757    PUSH_DATA (push, 1);
 758    BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
 759    PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
 760    BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
 761    PUSH_DATA (push, 0x11111111);
 762    BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
 763    PUSH_DATA (push, 1);
 764
 765    BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
 766    PUSH_DATA (push, 0);
 767    if (screen->base.class_3d >= NV84_3D_CLASS) {
 768       BEGIN_NV04(push, NV84_3D(VERTEX_ID_BASE), 1);
 769       PUSH_DATA (push, 0);
 770    }
 771
 772    PUSH_KICK (push);
 773 }
 774
 775 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
 776       uint64_t *tls_size)
 777 {
 778    struct nouveau_device *dev = screen->base.device;
 779    int ret;
 780
 781    screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
 782          ONE_TEMP_SIZE;
 783    if (nouveau_mesa_debug)
 784       debug_printf("allocating space for %u temps\n",
 785             util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
 786    *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
 787          screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
 788
 789    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 790                         *tls_size, NULL, &screen->tls_bo);
 791    if (ret) {
 792       NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
 793       return ret;
 794    }
 795
 796    return 0;
 797 }
 798
 799 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
 800 {
 801    struct nouveau_pushbuf *push = screen->base.pushbuf;
 802    int ret;
 803    uint64_t tls_size;
 804
 805    if (tls_space < screen->cur_tls_space)
 806       return 0;
 807    if (tls_space > screen->max_tls_space) {
 808       /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
 809        * LOCAL_WARPS_NO_CLAMP) */
 810       NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
 811             (unsigned)(tls_space / ONE_TEMP_SIZE),
 812             (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
 813       return -ENOMEM;
 814    }
 815
 816    nouveau_bo_ref(NULL, &screen->tls_bo);
 817    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
 818    if (ret)
 819       return ret;
 820
 821    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 822    PUSH_DATAh(push, screen->tls_bo->offset);
 823    PUSH_DATA (push, screen->tls_bo->offset);
 824    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 825
 826    return 1;
 827 }
 828
 829 struct nouveau_screen *
 830 nv50_screen_create(struct nouveau_device *dev)
 831 {
 832    struct nv50_screen *screen;
 833    struct pipe_screen *pscreen;
 834    struct nouveau_object *chan;
 835    uint64_t value;
 836    uint32_t tesla_class;
 837    unsigned stack_size;
 838    int ret;
 839
 840    screen = CALLOC_STRUCT(nv50_screen);
 841    if (!screen)
 842       return NULL;
 843    pscreen = &screen->base.base;
 844    pscreen->destroy = nv50_screen_destroy;
 845
 846    ret = nouveau_screen_init(&screen->base, dev);
 847    if (ret) {
 848       NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
 849       goto fail;
 850    }
 851
 852    /* TODO: Prevent FIFO prefetch before transfer of index buffers and
 853     *  admit them to VRAM.
 854     */
 855    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
 856       PIPE_BIND_VERTEX_BUFFER;
 857    screen->base.sysmem_bindings |=
 858       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
 859
 860    screen->base.pushbuf->user_priv = screen;
 861    screen->base.pushbuf->rsvd_kick = 5;
 862
 863    chan = screen->base.channel;
 864
 865    pscreen->context_create = nv50_create;
 866    pscreen->is_format_supported = nv50_screen_is_format_supported;
 867    pscreen->get_param = nv50_screen_get_param;
 868    pscreen->get_shader_param = nv50_screen_get_shader_param;
 869    pscreen->get_paramf = nv50_screen_get_paramf;
 870    pscreen->get_compute_param = nv50_screen_get_compute_param;
 871    pscreen->get_driver_query_info = nv50_screen_get_driver_query_info;
 872    pscreen->get_driver_query_group_info = nv50_screen_get_driver_query_group_info;
 873
 874    nv50_screen_init_resource_functions(pscreen);
 875
 876    if (screen->base.device->chipset < 0x84 ||
 877        debug_get_bool_option("NOUVEAU_PMPEG", false)) {
 878       /* PMPEG */
 879       nouveau_screen_init_vdec(&screen->base);
 880    } else if (screen->base.device->chipset < 0x98 ||
 881               screen->base.device->chipset == 0xa0) {
 882       /* VP2 */
 883       screen->base.base.get_video_param = nv84_screen_get_video_param;
 884       screen->base.base.is_video_format_supported = nv84_screen_video_supported;
 885    } else {
 886       /* VP3/4 */
 887       screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
 888       screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
 889    }
 890
 891    ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
 892                         NULL, &screen->fence.bo);
 893    if (ret) {
 894       NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
 895       goto fail;
 896    }
 897
 898    nouveau_bo_map(screen->fence.bo, 0, NULL);
 899    screen->fence.map = screen->fence.bo->map;
 900    screen->base.fence.emit = nv50_screen_fence_emit;
 901    screen->base.fence.update = nv50_screen_fence_update;
 902
 903    ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
 904                             &(struct nv04_notify){ .length = 32 },
 905                             sizeof(struct nv04_notify), &screen->sync);
 906    if (ret) {
 907       NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
 908       goto fail;
 909    }
 910
 911    ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
 912                             NULL, 0, &screen->m2mf);
 913    if (ret) {
 914       NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
 915       goto fail;
 916    }
 917
 918    ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
 919                             NULL, 0, &screen->eng2d);
 920    if (ret) {
 921       NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
 922       goto fail;
 923    }
 924
 925    switch (dev->chipset & 0xf0) {
 926    case 0x50:
 927       tesla_class = NV50_3D_CLASS;
 928       break;
 929    case 0x80:
 930    case 0x90:
 931       tesla_class = NV84_3D_CLASS;
 932       break;
 933    case 0xa0:
 934       switch (dev->chipset) {
 935       case 0xa0:
 936       case 0xaa:
 937       case 0xac:
 938          tesla_class = NVA0_3D_CLASS;
 939          break;
 940       case 0xaf:
 941          tesla_class = NVAF_3D_CLASS;
 942          break;
 943       default:
 944          tesla_class = NVA3_3D_CLASS;
 945          break;
 946       }
 947       break;
 948    default:
 949       NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
 950       goto fail;
 951    }
 952    screen->base.class_3d = tesla_class;
 953
 954    ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
 955                             NULL, 0, &screen->tesla);
 956    if (ret) {
 957       NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
 958       goto fail;
 959    }
 960
 961    /* This over-allocates by a page. The GP, which would execute at the end of
 962     * the last page, would trigger faults. The going theory is that it
 963     * prefetches up to a certain amount.
 964     */
 965    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 966                         (3 << NV50_CODE_BO_SIZE_LOG2) + 0x1000,
 967                         NULL, &screen->code);
 968    if (ret) {
 969       NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
 970       goto fail;
 971    }
 972
 973    nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 974    nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 975    nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
 976
 977    nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
 978
 979    screen->TPs = util_bitcount(value & 0xffff);
 980    screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
 981
 982    screen->mp_count = screen->TPs * screen->MPsInTP;
 983
 984    stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
 985          STACK_WARPS_ALLOC * 64 * 8;
 986
 987    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
 988                         &screen->stack_bo);
 989    if (ret) {
 990       NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
 991       goto fail;
 992    }
 993
 994    uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
 995          screen->MPsInTP * LOCAL_WARPS_ALLOC *  THREADS_IN_WARP *
 996          ONE_TEMP_SIZE;
 997    screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
 998    screen->max_tls_space /= 2; /* half of vram */
 999
1000    /* hw can address max 64 KiB */
1001    screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
1002
1003    uint64_t tls_size;
1004    unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
1005    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
1006    if (ret)
1007       goto fail;
1008
1009    if (nouveau_mesa_debug)
1010       debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
1011             screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
1012
1013    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
1014                         &screen->uniforms);
1015    if (ret) {
1016       NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
1017       goto fail;
1018    }
1019
1020    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
1021                         &screen->txc);
1022    if (ret) {
1023       NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
1024       goto fail;
1025    }
1026
1027    screen->tic.entries = CALLOC(4096, sizeof(void *));
1028    screen->tsc.entries = screen->tic.entries + 2048;
1029
1030    if (!nv50_blitter_create(screen))
1031       goto fail;
1032
1033    nv50_screen_init_hwctx(screen);
1034
1035    ret = nv50_screen_compute_setup(screen, screen->base.pushbuf);
1036    if (ret) {
1037       NOUVEAU_ERR("Failed to init compute context: %d\n", ret);
1038       goto fail;
1039    }
1040
1041    nouveau_fence_new(&screen->base, &screen->base.fence.current);
1042
1043    return &screen->base;
1044
1045 fail:
1046    screen->base.base.context_create = NULL;
1047    return &screen->base;
1048 }
1049
1050 int
1051 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
1052 {
1053    int i = screen->tic.next;
1054
1055    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1056       i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1057
1058    screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1059
1060    if (screen->tic.entries[i])
1061       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1062
1063    screen->tic.entries[i] = entry;
1064    return i;
1065 }
1066
1067 int
1068 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
1069 {
1070    int i = screen->tsc.next;
1071
1072    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1073       i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1074
1075    screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1076
1077    if (screen->tsc.entries[i])
1078       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1079
1080    screen->tsc.entries[i] = entry;
1081    return i;
1082 }