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 storage_sample_count,
  50                                 unsigned bindings)
  51 {
  52    if (sample_count > 8)
  53       return false;
  54    if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
  55       return false;
  56    if (sample_count == 8 && util_format_get_blocksizebits(format) >= 128)
  57       return false;
  58
  59    if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
  60       return false;
  61
  62    switch (format) {
  63    case PIPE_FORMAT_Z16_UNORM:
  64       if (nv50_screen(pscreen)->tesla->oclass < NVA0_3D_CLASS)
  65          return false;
  66       break;
  67    default:
  68       break;
  69    }
  70
  71    if (bindings & PIPE_BIND_LINEAR)
  72       if (util_format_is_depth_or_stencil(format) ||
  73           (target != PIPE_TEXTURE_1D &&
  74            target != PIPE_TEXTURE_2D &&
  75            target != PIPE_TEXTURE_RECT) ||
  76           sample_count > 1)
  77          return false;
  78
  79    /* shared is always supported */
  80    bindings &= ~(PIPE_BIND_LINEAR |
  81                  PIPE_BIND_SHARED);
  82
  83    return (( nv50_format_table[format].usage |
  84             nv50_vertex_format[format].usage) & bindings) == bindings;
  85 }
  86
  87 static int
  88 nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
  89 {
  90    const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
  91    struct nouveau_device *dev = nouveau_screen(pscreen)->device;
  92
  93    switch (param) {
  94    /* non-boolean caps */
  95    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
  96       return 14;
  97    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
  98       return 12;
  99    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 100       return 14;
 101    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 102       return 512;
 103    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 104    case PIPE_CAP_MIN_TEXEL_OFFSET:
 105       return -8;
 106    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 107    case PIPE_CAP_MAX_TEXEL_OFFSET:
 108       return 7;
 109    case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 110       return 128 * 1024 * 1024;
 111    case PIPE_CAP_GLSL_FEATURE_LEVEL:
 112       return 330;
 113    case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
 114       return 330;
 115    case PIPE_CAP_MAX_RENDER_TARGETS:
 116       return 8;
 117    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 118       return 1;
 119    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 120       return 4;
 121    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 122    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 123       return 64;
 124    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 125    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 126       return 1024;
 127    case PIPE_CAP_MAX_VERTEX_STREAMS:
 128       return 1;
 129    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 130       return 2048;
 131    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 132       return 256;
 133    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 134       return 16; /* 256 for binding as RT, but that's not possible in GL */
 135    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 136       return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
 137    case PIPE_CAP_MAX_VIEWPORTS:
 138       return NV50_MAX_VIEWPORTS;
 139    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 140       return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
 141    case PIPE_CAP_ENDIANNESS:
 142       return PIPE_ENDIAN_LITTLE;
 143    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 144       return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
 145    case PIPE_CAP_MAX_WINDOW_RECTANGLES:
 146       return NV50_MAX_WINDOW_RECTANGLES;
 147
 148    /* supported caps */
 149    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 150    case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
 151    case PIPE_CAP_TEXTURE_SWIZZLE:
 152    case PIPE_CAP_NPOT_TEXTURES:
 153    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 154    case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
 155    case PIPE_CAP_ANISOTROPIC_FILTER:
 156    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 157    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 158    case PIPE_CAP_DEPTH_CLIP_DISABLE:
 159    case PIPE_CAP_POINT_SPRITE:
 160    case PIPE_CAP_SM3:
 161    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 162    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 163    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 164    case PIPE_CAP_QUERY_TIMESTAMP:
 165    case PIPE_CAP_QUERY_TIME_ELAPSED:
 166    case PIPE_CAP_OCCLUSION_QUERY:
 167    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 168    case PIPE_CAP_INDEP_BLEND_ENABLE:
 169    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 170    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 171    case PIPE_CAP_PRIMITIVE_RESTART:
 172    case PIPE_CAP_TGSI_INSTANCEID:
 173    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 174    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 175    case PIPE_CAP_CONDITIONAL_RENDER:
 176    case PIPE_CAP_TEXTURE_BARRIER:
 177    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 178    case PIPE_CAP_START_INSTANCE:
 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    case PIPE_CAP_TGSI_TEX_TXF_LZ:
 203    case PIPE_CAP_TGSI_CLOCK:
 204    case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
 205    case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 206       return 1;
 207    case PIPE_CAP_SEAMLESS_CUBE_MAP:
 208       return 1; /* class_3d >= NVA0_3D_CLASS; */
 209    /* supported on nva0+ */
 210    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 211       return class_3d >= NVA0_3D_CLASS;
 212    /* supported on nva3+ */
 213    case PIPE_CAP_CUBE_MAP_ARRAY:
 214    case PIPE_CAP_INDEP_BLEND_FUNC:
 215    case PIPE_CAP_TEXTURE_QUERY_LOD:
 216    case PIPE_CAP_SAMPLE_SHADING:
 217    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 218       return class_3d >= NVA3_3D_CLASS;
 219
 220    /* unsupported caps */
 221    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 222    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 223    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 224    case PIPE_CAP_SHADER_STENCIL_EXPORT:
 225    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 226    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 227    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 228    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 229    case PIPE_CAP_TGSI_TEXCOORD:
 230    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 231    case PIPE_CAP_TEXTURE_GATHER_SM5:
 232    case PIPE_CAP_FAKE_SW_MSAA:
 233    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 234    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 235    case PIPE_CAP_DRAW_INDIRECT:
 236    case PIPE_CAP_MULTI_DRAW_INDIRECT:
 237    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 238    case PIPE_CAP_VERTEXID_NOBASE:
 239    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
 240    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 241    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 242    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 243    case PIPE_CAP_DRAW_PARAMETERS:
 244    case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 245    case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 246    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 247    case PIPE_CAP_GENERATE_MIPMAP:
 248    case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 249    case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 250    case PIPE_CAP_QUERY_BUFFER_OBJECT:
 251    case PIPE_CAP_QUERY_MEMORY_INFO:
 252    case PIPE_CAP_PCI_GROUP:
 253    case PIPE_CAP_PCI_BUS:
 254    case PIPE_CAP_PCI_DEVICE:
 255    case PIPE_CAP_PCI_FUNCTION:
 256    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 257    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 258    case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 259    case PIPE_CAP_TGSI_VOTE:
 260    case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 261    case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 262    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 263    case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
 264    case PIPE_CAP_NATIVE_FENCE_FD:
 265    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 266    case PIPE_CAP_TGSI_FS_FBFETCH:
 267    case PIPE_CAP_DOUBLES:
 268    case PIPE_CAP_INT64:
 269    case PIPE_CAP_INT64_DIVMOD:
 270    case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
 271    case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
 272    case PIPE_CAP_TGSI_BALLOT:
 273    case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
 274    case PIPE_CAP_POST_DEPTH_COVERAGE:
 275    case PIPE_CAP_BINDLESS_TEXTURE:
 276    case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
 277    case PIPE_CAP_QUERY_SO_OVERFLOW:
 278    case PIPE_CAP_MEMOBJ:
 279    case PIPE_CAP_LOAD_CONSTBUF:
 280    case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
 281    case PIPE_CAP_TILE_RASTER_ORDER:
 282    case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
 283    case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS:
 284    case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
 285    case PIPE_CAP_CONTEXT_PRIORITY_MASK:
 286    case PIPE_CAP_FENCE_SIGNAL:
 287    case PIPE_CAP_CONSTBUF0_FLAGS:
 288    case PIPE_CAP_PACKED_UNIFORMS:
 289    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
 290    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
 291    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
 292    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
 293    case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
 294    case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
 295    case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
 296       return 0;
 297
 298    case PIPE_CAP_MAX_GS_INVOCATIONS:
 299       return 32;
 300    case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
 301       return 1 << 27;
 302    case PIPE_CAP_VENDOR_ID:
 303       return 0x10de;
 304    case PIPE_CAP_DEVICE_ID: {
 305       uint64_t device_id;
 306       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
 307          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
 308          return -1;
 309       }
 310       return device_id;
 311    }
 312    case PIPE_CAP_ACCELERATED:
 313       return 1;
 314    case PIPE_CAP_VIDEO_MEMORY:
 315       return dev->vram_size >> 20;
 316    case PIPE_CAP_UMA:
 317       return 0;
 318    }
 319
 320    NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
 321    return 0;
 322 }
 323
 324 static int
 325 nv50_screen_get_shader_param(struct pipe_screen *pscreen,
 326                              enum pipe_shader_type shader,
 327                              enum pipe_shader_cap param)
 328 {
 329    switch (shader) {
 330    case PIPE_SHADER_VERTEX:
 331    case PIPE_SHADER_GEOMETRY:
 332    case PIPE_SHADER_FRAGMENT:
 333       break;
 334    case PIPE_SHADER_COMPUTE:
 335    default:
 336       return 0;
 337    }
 338
 339    switch (param) {
 340    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 341    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 342    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 343    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 344       return 16384;
 345    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 346       return 4;
 347    case PIPE_SHADER_CAP_MAX_INPUTS:
 348       if (shader == PIPE_SHADER_VERTEX)
 349          return 32;
 350       return 15;
 351    case PIPE_SHADER_CAP_MAX_OUTPUTS:
 352       return 16;
 353    case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 354       return 65536;
 355    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 356       return NV50_MAX_PIPE_CONSTBUFS;
 357    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 358       return shader != PIPE_SHADER_FRAGMENT;
 359    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 360    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 361    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 362       return 1;
 363    case PIPE_SHADER_CAP_MAX_TEMPS:
 364       return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
 365    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 366       return 1;
 367    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 368       return 1;
 369    case PIPE_SHADER_CAP_INT64_ATOMICS:
 370    case PIPE_SHADER_CAP_FP16:
 371    case PIPE_SHADER_CAP_SUBROUTINES:
 372       return 0; /* please inline, or provide function declarations */
 373    case PIPE_SHADER_CAP_INTEGERS:
 374       return 1;
 375    case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
 376       return 1;
 377    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 378       /* The chip could handle more sampler views than samplers */
 379    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 380       return MIN2(16, PIPE_MAX_SAMPLERS);
 381    case PIPE_SHADER_CAP_PREFERRED_IR:
 382       return PIPE_SHADER_IR_TGSI;
 383    case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 384       return 32;
 385    case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 386    case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 387    case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
 388    case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 389    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 390    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 391    case PIPE_SHADER_CAP_SUPPORTED_IRS:
 392    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 393    case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
 394    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
 395    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
 396       return 0;
 397    case PIPE_SHADER_CAP_SCALAR_ISA:
 398       return 1;
 399    default:
 400       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
 401       return 0;
 402    }
 403 }
 404
 405 static float
 406 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
 407 {
 408    switch (param) {
 409    case PIPE_CAPF_MAX_LINE_WIDTH:
 410    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 411       return 10.0f;
 412    case PIPE_CAPF_MAX_POINT_WIDTH:
 413    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 414       return 64.0f;
 415    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 416       return 16.0f;
 417    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 418       return 4.0f;
 419    case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
 420    case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
 421    case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
 422       return 0.0f;
 423    }
 424
 425    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
 426    return 0.0f;
 427 }
 428
 429 static int
 430 nv50_screen_get_compute_param(struct pipe_screen *pscreen,
 431                               enum pipe_shader_ir ir_type,
 432                               enum pipe_compute_cap param, void *data)
 433 {
 434    struct nv50_screen *screen = nv50_screen(pscreen);
 435
 436 #define RET(x) do {                  \
 437    if (data)                         \
 438       memcpy(data, x, sizeof(x));    \
 439    return sizeof(x);                 \
 440 } while (0)
 441
 442    switch (param) {
 443    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
 444       RET((uint64_t []) { 2 });
 445    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
 446       RET(((uint64_t []) { 65535, 65535 }));
 447    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
 448       RET(((uint64_t []) { 512, 512, 64 }));
 449    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
 450       RET((uint64_t []) { 512 });
 451    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g0-15[] */
 452       RET((uint64_t []) { 1ULL << 32 });
 453    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
 454       RET((uint64_t []) { 16 << 10 });
 455    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
 456       RET((uint64_t []) { 16 << 10 });
 457    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
 458       RET((uint64_t []) { 4096 });
 459    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
 460       RET((uint32_t []) { 32 });
 461    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
 462       RET((uint64_t []) { 1ULL << 40 });
 463    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
 464       RET((uint32_t []) { 0 });
 465    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
 466       RET((uint32_t []) { screen->mp_count });
 467    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
 468       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
 469    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
 470       RET((uint32_t []) { 32 });
 471    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
 472       RET((uint64_t []) { 0 });
 473    default:
 474       return 0;
 475    }
 476
 477 #undef RET
 478 }
 479
 480 static void
 481 nv50_screen_destroy(struct pipe_screen *pscreen)
 482 {
 483    struct nv50_screen *screen = nv50_screen(pscreen);
 484
 485    if (!nouveau_drm_screen_unref(&screen->base))
 486       return;
 487
 488    if (screen->base.fence.current) {
 489       struct nouveau_fence *current = NULL;
 490
 491       /* nouveau_fence_wait will create a new current fence, so wait on the
 492        * _current_ one, and remove both.
 493        */
 494       nouveau_fence_ref(screen->base.fence.current, &current);
 495       nouveau_fence_wait(current, NULL);
 496       nouveau_fence_ref(NULL, &current);
 497       nouveau_fence_ref(NULL, &screen->base.fence.current);
 498    }
 499    if (screen->base.pushbuf)
 500       screen->base.pushbuf->user_priv = NULL;
 501
 502    if (screen->blitter)
 503       nv50_blitter_destroy(screen);
 504    if (screen->pm.prog) {
 505       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
 506       nv50_program_destroy(NULL, screen->pm.prog);
 507       FREE(screen->pm.prog);
 508    }
 509
 510    nouveau_bo_ref(NULL, &screen->code);
 511    nouveau_bo_ref(NULL, &screen->tls_bo);
 512    nouveau_bo_ref(NULL, &screen->stack_bo);
 513    nouveau_bo_ref(NULL, &screen->txc);
 514    nouveau_bo_ref(NULL, &screen->uniforms);
 515    nouveau_bo_ref(NULL, &screen->fence.bo);
 516
 517    nouveau_heap_destroy(&screen->vp_code_heap);
 518    nouveau_heap_destroy(&screen->gp_code_heap);
 519    nouveau_heap_destroy(&screen->fp_code_heap);
 520
 521    FREE(screen->tic.entries);
 522
 523    nouveau_object_del(&screen->tesla);
 524    nouveau_object_del(&screen->eng2d);
 525    nouveau_object_del(&screen->m2mf);
 526    nouveau_object_del(&screen->compute);
 527    nouveau_object_del(&screen->sync);
 528
 529    nouveau_screen_fini(&screen->base);
 530
 531    FREE(screen);
 532 }
 533
 534 static void
 535 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
 536 {
 537    struct nv50_screen *screen = nv50_screen(pscreen);
 538    struct nouveau_pushbuf *push = screen->base.pushbuf;
 539
 540    /* we need to do it after possible flush in MARK_RING */
 541    *sequence = ++screen->base.fence.sequence;
 542
 543    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
 544    PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
 545    PUSH_DATAh(push, screen->fence.bo->offset);
 546    PUSH_DATA (push, screen->fence.bo->offset);
 547    PUSH_DATA (push, *sequence);
 548    PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
 549                     NV50_3D_QUERY_GET_UNK4 |
 550                     NV50_3D_QUERY_GET_UNIT_CROP |
 551                     NV50_3D_QUERY_GET_TYPE_QUERY |
 552                     NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
 553                     NV50_3D_QUERY_GET_SHORT);
 554 }
 555
 556 static u32
 557 nv50_screen_fence_update(struct pipe_screen *pscreen)
 558 {
 559    return nv50_screen(pscreen)->fence.map[0];
 560 }
 561
 562 static void
 563 nv50_screen_init_hwctx(struct nv50_screen *screen)
 564 {
 565    struct nouveau_pushbuf *push = screen->base.pushbuf;
 566    struct nv04_fifo *fifo;
 567    unsigned i;
 568
 569    fifo = (struct nv04_fifo *)screen->base.channel->data;
 570
 571    BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
 572    PUSH_DATA (push, screen->m2mf->handle);
 573    BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
 574    PUSH_DATA (push, screen->sync->handle);
 575    PUSH_DATA (push, fifo->vram);
 576    PUSH_DATA (push, fifo->vram);
 577
 578    BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
 579    PUSH_DATA (push, screen->eng2d->handle);
 580    BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
 581    PUSH_DATA (push, screen->sync->handle);
 582    PUSH_DATA (push, fifo->vram);
 583    PUSH_DATA (push, fifo->vram);
 584    PUSH_DATA (push, fifo->vram);
 585    BEGIN_NV04(push, NV50_2D(OPERATION), 1);
 586    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
 587    BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
 588    PUSH_DATA (push, 0);
 589    BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
 590    PUSH_DATA (push, 0);
 591    BEGIN_NV04(push, SUBC_2D(0x0888), 1);
 592    PUSH_DATA (push, 1);
 593    BEGIN_NV04(push, NV50_2D(COND_MODE), 1);
 594    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
 595
 596    BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
 597    PUSH_DATA (push, screen->tesla->handle);
 598
 599    BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
 600    PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
 601
 602    BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
 603    PUSH_DATA (push, screen->sync->handle);
 604    BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
 605    for (i = 0; i < 11; ++i)
 606       PUSH_DATA(push, fifo->vram);
 607    BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
 608    for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
 609       PUSH_DATA(push, fifo->vram);
 610
 611    BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
 612    PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
 613    BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
 614    PUSH_DATA (push, 0xf);
 615
 616    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
 617       BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
 618       PUSH_DATA (push, 0x18);
 619    }
 620
 621    BEGIN_NV04(push, NV50_3D(ZETA_COMP_ENABLE), 1);
 622    PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 623
 624    BEGIN_NV04(push, NV50_3D(RT_COMP_ENABLE(0)), 8);
 625    for (i = 0; i < 8; ++i)
 626       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
 627
 628    BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
 629    PUSH_DATA (push, 1);
 630
 631    BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
 632    PUSH_DATA (push, 0);
 633    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
 634    PUSH_DATA (push, 0);
 635    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
 636    PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
 637    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
 638    PUSH_DATA (push, 0);
 639    BEGIN_NV04(push, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
 640    PUSH_DATA (push, 1);
 641    BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
 642    PUSH_DATA (push, 1);
 643
 644    if (screen->tesla->oclass >= NVA0_3D_CLASS) {
 645       BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
 646       PUSH_DATA (push, 0);
 647    }
 648
 649    BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
 650    PUSH_DATA (push, 0);
 651    BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
 652    PUSH_DATA (push, 0);
 653    PUSH_DATA (push, 0);
 654    BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
 655    PUSH_DATA (push, 0x3f);
 656
 657    BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
 658    PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 659    PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
 660
 661    BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
 662    PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 663    PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
 664
 665    BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
 666    PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 667    PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
 668
 669    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 670    PUSH_DATAh(push, screen->tls_bo->offset);
 671    PUSH_DATA (push, screen->tls_bo->offset);
 672    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 673
 674    BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
 675    PUSH_DATAh(push, screen->stack_bo->offset);
 676    PUSH_DATA (push, screen->stack_bo->offset);
 677    PUSH_DATA (push, 4);
 678
 679    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 680    PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
 681    PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
 682    PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
 683
 684    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 685    PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
 686    PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
 687    PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
 688
 689    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 690    PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
 691    PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
 692    PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
 693
 694    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
 695    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
 696    PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
 697    PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
 698
 699    BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
 700    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
 701    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
 702    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
 703
 704    /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
 705    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
 706    PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
 707    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
 708    PUSH_DATAf(push, 0.0f);
 709    PUSH_DATAf(push, 0.0f);
 710    PUSH_DATAf(push, 0.0f);
 711    PUSH_DATAf(push, 0.0f);
 712    BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
 713    PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 714    PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
 715
 716    nv50_upload_ms_info(push);
 717
 718    /* max TIC (bits 4:8) & TSC bindings, per program type */
 719    for (i = 0; i < 3; ++i) {
 720       BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
 721       PUSH_DATA (push, 0x54);
 722    }
 723
 724    BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
 725    PUSH_DATAh(push, screen->txc->offset);
 726    PUSH_DATA (push, screen->txc->offset);
 727    PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
 728
 729    BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
 730    PUSH_DATAh(push, screen->txc->offset + 65536);
 731    PUSH_DATA (push, screen->txc->offset + 65536);
 732    PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
 733
 734    BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
 735    PUSH_DATA (push, 0);
 736
 737    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
 738    PUSH_DATA (push, 0);
 739    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
 740    PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
 741    BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
 742    for (i = 0; i < 8 * 2; ++i)
 743       PUSH_DATA(push, 0);
 744    BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
 745    PUSH_DATA (push, 0);
 746
 747    BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
 748    PUSH_DATA (push, 1);
 749    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 750       BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(i)), 2);
 751       PUSH_DATAf(push, 0.0f);
 752       PUSH_DATAf(push, 1.0f);
 753       BEGIN_NV04(push, NV50_3D(VIEWPORT_HORIZ(i)), 2);
 754       PUSH_DATA (push, 8192 << 16);
 755       PUSH_DATA (push, 8192 << 16);
 756    }
 757
 758    BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
 759 #ifdef NV50_SCISSORS_CLIPPING
 760    PUSH_DATA (push, 0x0000);
 761 #else
 762    PUSH_DATA (push, 0x1080);
 763 #endif
 764
 765    BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
 766    PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
 767
 768    /* We use scissors instead of exact view volume clipping,
 769     * so they're always enabled.
 770     */
 771    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
 772       BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(i)), 3);
 773       PUSH_DATA (push, 1);
 774       PUSH_DATA (push, 8192 << 16);
 775       PUSH_DATA (push, 8192 << 16);
 776    }
 777
 778    BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
 779    PUSH_DATA (push, 1);
 780    BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
 781    PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
 782    BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
 783    PUSH_DATA (push, 0x11111111);
 784    BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
 785    PUSH_DATA (push, 1);
 786
 787    BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
 788    PUSH_DATA (push, 0);
 789    if (screen->base.class_3d >= NV84_3D_CLASS) {
 790       BEGIN_NV04(push, NV84_3D(VERTEX_ID_BASE), 1);
 791       PUSH_DATA (push, 0);
 792    }
 793
 794    BEGIN_NV04(push, NV50_3D(UNK0FDC), 1);
 795    PUSH_DATA (push, 1);
 796    BEGIN_NV04(push, NV50_3D(UNK19C0), 1);
 797    PUSH_DATA (push, 1);
 798
 799    PUSH_KICK (push);
 800 }
 801
 802 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
 803       uint64_t *tls_size)
 804 {
 805    struct nouveau_device *dev = screen->base.device;
 806    int ret;
 807
 808    screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
 809          ONE_TEMP_SIZE;
 810    if (nouveau_mesa_debug)
 811       debug_printf("allocating space for %u temps\n",
 812             util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
 813    *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
 814          screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
 815
 816    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 817                         *tls_size, NULL, &screen->tls_bo);
 818    if (ret) {
 819       NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
 820       return ret;
 821    }
 822
 823    return 0;
 824 }
 825
 826 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
 827 {
 828    struct nouveau_pushbuf *push = screen->base.pushbuf;
 829    int ret;
 830    uint64_t tls_size;
 831
 832    if (tls_space < screen->cur_tls_space)
 833       return 0;
 834    if (tls_space > screen->max_tls_space) {
 835       /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
 836        * LOCAL_WARPS_NO_CLAMP) */
 837       NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
 838             (unsigned)(tls_space / ONE_TEMP_SIZE),
 839             (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
 840       return -ENOMEM;
 841    }
 842
 843    nouveau_bo_ref(NULL, &screen->tls_bo);
 844    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
 845    if (ret)
 846       return ret;
 847
 848    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
 849    PUSH_DATAh(push, screen->tls_bo->offset);
 850    PUSH_DATA (push, screen->tls_bo->offset);
 851    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
 852
 853    return 1;
 854 }
 855
 856 struct nouveau_screen *
 857 nv50_screen_create(struct nouveau_device *dev)
 858 {
 859    struct nv50_screen *screen;
 860    struct pipe_screen *pscreen;
 861    struct nouveau_object *chan;
 862    uint64_t value;
 863    uint32_t tesla_class;
 864    unsigned stack_size;
 865    int ret;
 866
 867    screen = CALLOC_STRUCT(nv50_screen);
 868    if (!screen)
 869       return NULL;
 870    pscreen = &screen->base.base;
 871    pscreen->destroy = nv50_screen_destroy;
 872
 873    ret = nouveau_screen_init(&screen->base, dev);
 874    if (ret) {
 875       NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
 876       goto fail;
 877    }
 878
 879    /* TODO: Prevent FIFO prefetch before transfer of index buffers and
 880     *  admit them to VRAM.
 881     */
 882    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
 883       PIPE_BIND_VERTEX_BUFFER;
 884    screen->base.sysmem_bindings |=
 885       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
 886
 887    screen->base.pushbuf->user_priv = screen;
 888    screen->base.pushbuf->rsvd_kick = 5;
 889
 890    chan = screen->base.channel;
 891
 892    pscreen->context_create = nv50_create;
 893    pscreen->is_format_supported = nv50_screen_is_format_supported;
 894    pscreen->get_param = nv50_screen_get_param;
 895    pscreen->get_shader_param = nv50_screen_get_shader_param;
 896    pscreen->get_paramf = nv50_screen_get_paramf;
 897    pscreen->get_compute_param = nv50_screen_get_compute_param;
 898    pscreen->get_driver_query_info = nv50_screen_get_driver_query_info;
 899    pscreen->get_driver_query_group_info = nv50_screen_get_driver_query_group_info;
 900
 901    nv50_screen_init_resource_functions(pscreen);
 902
 903    if (screen->base.device->chipset < 0x84 ||
 904        debug_get_bool_option("NOUVEAU_PMPEG", false)) {
 905       /* PMPEG */
 906       nouveau_screen_init_vdec(&screen->base);
 907    } else if (screen->base.device->chipset < 0x98 ||
 908               screen->base.device->chipset == 0xa0) {
 909       /* VP2 */
 910       screen->base.base.get_video_param = nv84_screen_get_video_param;
 911       screen->base.base.is_video_format_supported = nv84_screen_video_supported;
 912    } else {
 913       /* VP3/4 */
 914       screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
 915       screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
 916    }
 917
 918    ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
 919                         NULL, &screen->fence.bo);
 920    if (ret) {
 921       NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
 922       goto fail;
 923    }
 924
 925    nouveau_bo_map(screen->fence.bo, 0, NULL);
 926    screen->fence.map = screen->fence.bo->map;
 927    screen->base.fence.emit = nv50_screen_fence_emit;
 928    screen->base.fence.update = nv50_screen_fence_update;
 929
 930    ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
 931                             &(struct nv04_notify){ .length = 32 },
 932                             sizeof(struct nv04_notify), &screen->sync);
 933    if (ret) {
 934       NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
 935       goto fail;
 936    }
 937
 938    ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
 939                             NULL, 0, &screen->m2mf);
 940    if (ret) {
 941       NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
 942       goto fail;
 943    }
 944
 945    ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
 946                             NULL, 0, &screen->eng2d);
 947    if (ret) {
 948       NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
 949       goto fail;
 950    }
 951
 952    switch (dev->chipset & 0xf0) {
 953    case 0x50:
 954       tesla_class = NV50_3D_CLASS;
 955       break;
 956    case 0x80:
 957    case 0x90:
 958       tesla_class = NV84_3D_CLASS;
 959       break;
 960    case 0xa0:
 961       switch (dev->chipset) {
 962       case 0xa0:
 963       case 0xaa:
 964       case 0xac:
 965          tesla_class = NVA0_3D_CLASS;
 966          break;
 967       case 0xaf:
 968          tesla_class = NVAF_3D_CLASS;
 969          break;
 970       default:
 971          tesla_class = NVA3_3D_CLASS;
 972          break;
 973       }
 974       break;
 975    default:
 976       NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
 977       goto fail;
 978    }
 979    screen->base.class_3d = tesla_class;
 980
 981    ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
 982                             NULL, 0, &screen->tesla);
 983    if (ret) {
 984       NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
 985       goto fail;
 986    }
 987
 988    /* This over-allocates by a page. The GP, which would execute at the end of
 989     * the last page, would trigger faults. The going theory is that it
 990     * prefetches up to a certain amount.
 991     */
 992    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
 993                         (3 << NV50_CODE_BO_SIZE_LOG2) + 0x1000,
 994                         NULL, &screen->code);
 995    if (ret) {
 996       NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
 997       goto fail;
 998    }
 999
1000    nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1001    nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1002    nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1003
1004    nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1005
1006    screen->TPs = util_bitcount(value & 0xffff);
1007    screen->MPsInTP = util_bitcount(value & 0x0f000000);
1008
1009    screen->mp_count = screen->TPs * screen->MPsInTP;
1010
1011    stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
1012          STACK_WARPS_ALLOC * 64 * 8;
1013
1014    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
1015                         &screen->stack_bo);
1016    if (ret) {
1017       NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
1018       goto fail;
1019    }
1020
1021    uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
1022          screen->MPsInTP * LOCAL_WARPS_ALLOC *  THREADS_IN_WARP *
1023          ONE_TEMP_SIZE;
1024    screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
1025    screen->max_tls_space /= 2; /* half of vram */
1026
1027    /* hw can address max 64 KiB */
1028    screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
1029
1030    uint64_t tls_size;
1031    unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
1032    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
1033    if (ret)
1034       goto fail;
1035
1036    if (nouveau_mesa_debug)
1037       debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
1038             screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
1039
1040    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
1041                         &screen->uniforms);
1042    if (ret) {
1043       NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
1044       goto fail;
1045    }
1046
1047    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
1048                         &screen->txc);
1049    if (ret) {
1050       NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
1051       goto fail;
1052    }
1053
1054    screen->tic.entries = CALLOC(4096, sizeof(void *));
1055    screen->tsc.entries = screen->tic.entries + 2048;
1056
1057    if (!nv50_blitter_create(screen))
1058       goto fail;
1059
1060    nv50_screen_init_hwctx(screen);
1061
1062    ret = nv50_screen_compute_setup(screen, screen->base.pushbuf);
1063    if (ret) {
1064       NOUVEAU_ERR("Failed to init compute context: %d\n", ret);
1065       goto fail;
1066    }
1067
1068    nouveau_fence_new(&screen->base, &screen->base.fence.current);
1069
1070    return &screen->base;
1071
1072 fail:
1073    screen->base.base.context_create = NULL;
1074    return &screen->base;
1075 }
1076
1077 int
1078 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
1079 {
1080    int i = screen->tic.next;
1081
1082    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1083       i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1084
1085    screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1086
1087    if (screen->tic.entries[i])
1088       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1089
1090    screen->tic.entries[i] = entry;
1091    return i;
1092 }
1093
1094 int
1095 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
1096 {
1097    int i = screen->tsc.next;
1098
1099    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1100       i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1101
1102    screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1103
1104    if (screen->tsc.entries[i])
1105       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1106
1107    screen->tsc.entries[i] = entry;
1108    return i;
1109 }