src/gallium/drivers/llvmpipe/lp_screen.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2008 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28
  29 #include "util/u_memory.h"
  30 #include "util/u_math.h"
  31 #include "util/u_cpu_detect.h"
  32 #include "util/format/u_format.h"
  33 #include "util/u_screen.h"
  34 #include "util/u_string.h"
  35 #include "util/format/u_format_s3tc.h"
  36 #include "pipe/p_defines.h"
  37 #include "pipe/p_screen.h"
  38 #include "draw/draw_context.h"
  39 #include "gallivm/lp_bld_type.h"
  40 #include "gallivm/lp_bld_nir.h"
  41
  42 #include "util/os_misc.h"
  43 #include "util/os_time.h"
  44 #include "lp_texture.h"
  45 #include "lp_fence.h"
  46 #include "lp_jit.h"
  47 #include "lp_screen.h"
  48 #include "lp_context.h"
  49 #include "lp_debug.h"
  50 #include "lp_public.h"
  51 #include "lp_limits.h"
  52 #include "lp_rast.h"
  53 #include "lp_cs_tpool.h"
  54
  55 #include "state_tracker/sw_winsys.h"
  56
  57 #include "nir.h"
  58
  59 #ifdef DEBUG
  60 int LP_DEBUG = 0;
  61
  62 static const struct debug_named_value lp_debug_flags[] = {
  63    { "pipe",   DEBUG_PIPE, NULL },
  64    { "tgsi",   DEBUG_TGSI, NULL },
  65    { "tex",    DEBUG_TEX, NULL },
  66    { "setup",  DEBUG_SETUP, NULL },
  67    { "rast",   DEBUG_RAST, NULL },
  68    { "query",  DEBUG_QUERY, NULL },
  69    { "screen", DEBUG_SCREEN, NULL },
  70    { "counters", DEBUG_COUNTERS, NULL },
  71    { "scene", DEBUG_SCENE, NULL },
  72    { "fence", DEBUG_FENCE, NULL },
  73    { "mem", DEBUG_MEM, NULL },
  74    { "fs", DEBUG_FS, NULL },
  75    { "cs", DEBUG_CS, NULL },
  76    { "tgsi_ir", DEBUG_TGSI_IR, NULL },
  77    { "cl", DEBUG_CL, NULL },
  78    DEBUG_NAMED_VALUE_END
  79 };
  80 #endif
  81
  82 int LP_PERF = 0;
  83 static const struct debug_named_value lp_perf_flags[] = {
  84    { "texmem",         PERF_TEX_MEM, NULL },
  85    { "no_mipmap",      PERF_NO_MIPMAPS, NULL },
  86    { "no_linear",      PERF_NO_LINEAR, NULL },
  87    { "no_mip_linear",  PERF_NO_MIP_LINEAR, NULL },
  88    { "no_tex",         PERF_NO_TEX, NULL },
  89    { "no_blend",       PERF_NO_BLEND, NULL },
  90    { "no_depth",       PERF_NO_DEPTH, NULL },
  91    { "no_alphatest",   PERF_NO_ALPHATEST, NULL },
  92    DEBUG_NAMED_VALUE_END
  93 };
  94
  95
  96 static const char *
  97 llvmpipe_get_vendor(struct pipe_screen *screen)
  98 {
  99    return "VMware, Inc.";
 100 }
 101
 102
 103 static const char *
 104 llvmpipe_get_name(struct pipe_screen *screen)
 105 {
 106    static char buf[100];
 107    snprintf(buf, sizeof(buf), "llvmpipe (LLVM " MESA_LLVM_VERSION_STRING ", %u bits)",
 108             lp_native_vector_width );
 109    return buf;
 110 }
 111
 112
 113 static int
 114 llvmpipe_get_param(struct pipe_screen *screen, enum pipe_cap param)
 115 {
 116    switch (param) {
 117    case PIPE_CAP_NPOT_TEXTURES:
 118    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 119    case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
 120       return 1;
 121    case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD:
 122    case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES:
 123    case PIPE_CAP_VERTEX_SHADER_SATURATE:
 124       return 1;
 125    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 126       return 1;
 127    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 128       return PIPE_MAX_SO_BUFFERS;
 129    case PIPE_CAP_ANISOTROPIC_FILTER:
 130       return 0;
 131    case PIPE_CAP_POINT_SPRITE:
 132       return 1;
 133    case PIPE_CAP_MAX_RENDER_TARGETS:
 134       return PIPE_MAX_COLOR_BUFS;
 135    case PIPE_CAP_OCCLUSION_QUERY:
 136       return 1;
 137    case PIPE_CAP_QUERY_TIME_ELAPSED:
 138       return 0;
 139    case PIPE_CAP_QUERY_TIMESTAMP:
 140       return 1;
 141    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 142       return 1;
 143    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 144    case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
 145       return 1;
 146    case PIPE_CAP_TEXTURE_SWIZZLE:
 147       return 1;
 148    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 149    case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
 150       return 0;
 151    case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
 152       return 1 << (LP_MAX_TEXTURE_2D_LEVELS - 1);
 153    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 154       return LP_MAX_TEXTURE_3D_LEVELS;
 155    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 156       return LP_MAX_TEXTURE_CUBE_LEVELS;
 157    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 158       return LP_MAX_TEXTURE_ARRAY_LAYERS;
 159    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 160       return 1;
 161    case PIPE_CAP_INDEP_BLEND_ENABLE:
 162       return 1;
 163    case PIPE_CAP_INDEP_BLEND_FUNC:
 164       return 1;
 165    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 166    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 167    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 168       return 1;
 169    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 170    case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
 171       return 0;
 172    case PIPE_CAP_PRIMITIVE_RESTART:
 173       return 1;
 174    case PIPE_CAP_DEPTH_CLIP_DISABLE:
 175       return 1;
 176    case PIPE_CAP_SHADER_STENCIL_EXPORT:
 177       return 1;
 178    case PIPE_CAP_TGSI_INSTANCEID:
 179    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 180    case PIPE_CAP_START_INSTANCE:
 181       return 1;
 182    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 183       return 0;
 184    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 185       return 1;
 186    case PIPE_CAP_SEAMLESS_CUBE_MAP:
 187    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 188       return 1;
 189    /* this is a lie could support arbitrary large offsets */
 190    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 191    case PIPE_CAP_MIN_TEXEL_OFFSET:
 192       return -32;
 193    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 194    case PIPE_CAP_MAX_TEXEL_OFFSET:
 195       return 31;
 196    case PIPE_CAP_CONDITIONAL_RENDER:
 197    case PIPE_CAP_TEXTURE_BARRIER:
 198       return 1;
 199    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 200    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 201       return 16*4;
 202    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 203    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 204       return 1024;
 205    case PIPE_CAP_MAX_VERTEX_STREAMS: {
 206       struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 207       return lscreen->use_tgsi ? 1 : 4;
 208    }
 209    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 210       return 2048;
 211    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 212    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 213       return 1;
 214    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 215       return 0;
 216    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 217    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 218       return 1;
 219    case PIPE_CAP_GLSL_FEATURE_LEVEL:
 220       return 330;
 221    case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
 222       return 140;
 223    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 224       return 0;
 225    case PIPE_CAP_COMPUTE:
 226       return GALLIVM_HAVE_CORO;
 227    case PIPE_CAP_USER_VERTEX_BUFFERS:
 228       return 1;
 229    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 230    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 231    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 232       return 0;
 233    case PIPE_CAP_TGSI_TEXCOORD:
 234    case PIPE_CAP_DRAW_INDIRECT:
 235       return 1;
 236
 237    case PIPE_CAP_CUBE_MAP_ARRAY:
 238       return 1;
 239    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 240       return 16;
 241    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 242       return 64;
 243    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 244       return 1;
 245    case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 246       return 65536;
 247    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 248       return 16;
 249    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 250       return 0;
 251    case PIPE_CAP_MAX_VIEWPORTS:
 252       return PIPE_MAX_VIEWPORTS;
 253    case PIPE_CAP_ENDIANNESS:
 254       return PIPE_ENDIAN_NATIVE;
 255    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 256       return 1;
 257    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 258       return 1;
 259    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 260       return 4;
 261    case PIPE_CAP_TEXTURE_GATHER_SM5:
 262    case PIPE_CAP_SAMPLE_SHADING:
 263    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 264       return 0;
 265    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 266       return 1;
 267    case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 268       return 1;
 269    case PIPE_CAP_TGSI_TEX_TXF_LZ:
 270    case PIPE_CAP_SAMPLER_VIEW_TARGET:
 271       return 1;
 272    case PIPE_CAP_FAKE_SW_MSAA: {
 273       struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 274       return lscreen->use_tgsi ? 1 : 0;
 275    }
 276    case PIPE_CAP_TEXTURE_QUERY_LOD:
 277    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 278    case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 279    case PIPE_CAP_DOUBLES:
 280    case PIPE_CAP_INT64:
 281    case PIPE_CAP_INT64_DIVMOD:
 282    case PIPE_CAP_QUERY_SO_OVERFLOW:
 283    case PIPE_CAP_TGSI_DIV:
 284       return 1;
 285
 286    case PIPE_CAP_VENDOR_ID:
 287       return 0xFFFFFFFF;
 288    case PIPE_CAP_DEVICE_ID:
 289       return 0xFFFFFFFF;
 290    case PIPE_CAP_ACCELERATED:
 291       return 0;
 292    case PIPE_CAP_VIDEO_MEMORY: {
 293       /* XXX: Do we want to return the full amount fo system memory ? */
 294       uint64_t system_memory;
 295
 296       if (!os_get_total_physical_memory(&system_memory))
 297          return 0;
 298
 299       if (sizeof(void *) == 4)
 300          /* Cap to 2 GB on 32 bits system. We do this because llvmpipe does
 301           * eat application memory, which is quite limited on 32 bits. App
 302           * shouldn't expect too much available memory. */
 303          system_memory = MIN2(system_memory, 2048 << 20);
 304
 305       return (int)(system_memory >> 20);
 306    }
 307    case PIPE_CAP_UMA:
 308       return 0;
 309    case PIPE_CAP_CLIP_HALFZ:
 310       return 1;
 311    case PIPE_CAP_VERTEXID_NOBASE:
 312       return 0;
 313    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 314    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 315    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 316       return 1;
 317    case PIPE_CAP_CULL_DISTANCE:
 318       return 1;
 319    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 320       return 1;
 321    case PIPE_CAP_CLEAR_TEXTURE:
 322       return 1;
 323    case PIPE_CAP_MAX_VARYINGS:
 324       return 32;
 325    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 326       return 16;
 327    case PIPE_CAP_QUERY_BUFFER_OBJECT:
 328       return 1;
 329    case PIPE_CAP_DRAW_PARAMETERS:
 330       return 1;
 331    case PIPE_CAP_MULTI_DRAW_INDIRECT:
 332    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 333       return 1;
 334    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 335       return 32;
 336    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 337    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 338    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 339    case PIPE_CAP_DEPTH_BOUNDS_TEST:
 340    case PIPE_CAP_TGSI_TXQS:
 341    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 342    case PIPE_CAP_SHAREABLE_SHADERS:
 343    case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 344    case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 345    case PIPE_CAP_INVALIDATE_BUFFER:
 346    case PIPE_CAP_GENERATE_MIPMAP:
 347    case PIPE_CAP_STRING_MARKER:
 348    case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 349    case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 350    case PIPE_CAP_QUERY_MEMORY_INFO:
 351    case PIPE_CAP_PCI_GROUP:
 352    case PIPE_CAP_PCI_BUS:
 353    case PIPE_CAP_PCI_DEVICE:
 354    case PIPE_CAP_PCI_FUNCTION:
 355    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 356    case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 357    case PIPE_CAP_MAX_WINDOW_RECTANGLES:
 358    case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 359    case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 360    case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
 361    case PIPE_CAP_NATIVE_FENCE_FD:
 362    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 363    case PIPE_CAP_FBFETCH:
 364    case PIPE_CAP_TGSI_MUL_ZERO_WINS:
 365    case PIPE_CAP_TGSI_CLOCK:
 366    case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
 367    case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
 368    case PIPE_CAP_TGSI_BALLOT:
 369    case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
 370    case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
 371    case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 372    case PIPE_CAP_POST_DEPTH_COVERAGE:
 373    case PIPE_CAP_BINDLESS_TEXTURE:
 374    case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
 375    case PIPE_CAP_MEMOBJ:
 376    case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
 377    case PIPE_CAP_TILE_RASTER_ORDER:
 378    case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
 379    case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS:
 380    case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
 381    case PIPE_CAP_CONTEXT_PRIORITY_MASK:
 382    case PIPE_CAP_FENCE_SIGNAL:
 383    case PIPE_CAP_CONSTBUF0_FLAGS:
 384    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
 385    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
 386    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
 387    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
 388    case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
 389    case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
 390    case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
 391       return 0;
 392    case PIPE_CAP_MAX_GS_INVOCATIONS:
 393       return 32;
 394    case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
 395       return LP_MAX_TGSI_SHADER_BUFFER_SIZE;
 396    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 397    case PIPE_CAP_TGSI_TG4_COMPONENT_IN_SWIZZLE:
 398    case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 399       return 1;
 400    case PIPE_CAP_TGSI_VOTE:
 401    case PIPE_CAP_LOAD_CONSTBUF:
 402    case PIPE_CAP_TEXTURE_MULTISAMPLE:
 403    case PIPE_CAP_PACKED_UNIFORMS: {
 404       struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 405       return !lscreen->use_tgsi;
 406    }
 407    default:
 408       return u_pipe_screen_get_param_defaults(screen, param);
 409    }
 410 }
 411
 412 static int
 413 llvmpipe_get_shader_param(struct pipe_screen *screen,
 414                           enum pipe_shader_type shader,
 415                           enum pipe_shader_cap param)
 416 {
 417    struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 418    switch(shader)
 419    {
 420    case PIPE_SHADER_COMPUTE:
 421       if ((LP_DEBUG & DEBUG_CL) && param == PIPE_SHADER_CAP_SUPPORTED_IRS)
 422          return (1 << PIPE_SHADER_IR_TGSI) | (1 << PIPE_SHADER_IR_NIR) | (1 << PIPE_SHADER_IR_NIR_SERIALIZED);
 423    case PIPE_SHADER_FRAGMENT:
 424       if (param == PIPE_SHADER_CAP_PREFERRED_IR) {
 425          if (lscreen->use_tgsi)
 426             return PIPE_SHADER_IR_TGSI;
 427          else
 428             return PIPE_SHADER_IR_NIR;
 429       }
 430       switch (param) {
 431       default:
 432          return gallivm_get_shader_param(param);
 433       }
 434    case PIPE_SHADER_TESS_CTRL:
 435    case PIPE_SHADER_TESS_EVAL:
 436       if (lscreen->use_tgsi)
 437          return 0;
 438    case PIPE_SHADER_VERTEX:
 439    case PIPE_SHADER_GEOMETRY:
 440       if (param == PIPE_SHADER_CAP_PREFERRED_IR) {
 441          if (lscreen->use_tgsi)
 442             return PIPE_SHADER_IR_TGSI;
 443          else
 444             return PIPE_SHADER_IR_NIR;
 445       }
 446
 447       switch (param) {
 448       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 449          /* At this time, the draw module and llvmpipe driver only
 450           * support vertex shader texture lookups when LLVM is enabled in
 451           * the draw module.
 452           */
 453          if (debug_get_bool_option("DRAW_USE_LLVM", TRUE))
 454             return PIPE_MAX_SAMPLERS;
 455          else
 456             return 0;
 457       case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 458          if (debug_get_bool_option("DRAW_USE_LLVM", TRUE))
 459             return PIPE_MAX_SHADER_SAMPLER_VIEWS;
 460          else
 461             return 0;
 462       default:
 463          return draw_get_shader_param(shader, param);
 464       }
 465    default:
 466       return 0;
 467    }
 468 }
 469
 470 static float
 471 llvmpipe_get_paramf(struct pipe_screen *screen, enum pipe_capf param)
 472 {
 473    switch (param) {
 474    case PIPE_CAPF_MAX_LINE_WIDTH:
 475       /* fall-through */
 476    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 477       return 255.0; /* arbitrary */
 478    case PIPE_CAPF_MAX_POINT_WIDTH:
 479       /* fall-through */
 480    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 481       return LP_MAX_POINT_WIDTH; /* arbitrary */
 482    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 483       return 16.0; /* not actually signficant at this time */
 484    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 485       return 16.0; /* arbitrary */
 486    case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
 487       return 0.0;
 488    case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
 489       return 0.0;
 490    case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
 491       return 0.0;
 492    }
 493    /* should only get here on unhandled cases */
 494    debug_printf("Unexpected PIPE_CAP %d query\n", param);
 495    return 0.0;
 496 }
 497
 498 static int
 499 llvmpipe_get_compute_param(struct pipe_screen *_screen,
 500                            enum pipe_shader_ir ir_type,
 501                            enum pipe_compute_cap param,
 502                            void *ret)
 503 {
 504    switch (param) {
 505    case PIPE_COMPUTE_CAP_IR_TARGET:
 506       return 0;
 507    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
 508       if (ret) {
 509          uint64_t *grid_size = ret;
 510          grid_size[0] = 65535;
 511          grid_size[1] = 65535;
 512          grid_size[2] = 65535;
 513       }
 514       return 3 * sizeof(uint64_t) ;
 515    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
 516       if (ret) {
 517          uint64_t *block_size = ret;
 518          block_size[0] = 1024;
 519          block_size[1] = 1024;
 520          block_size[2] = 1024;
 521       }
 522       return 3 * sizeof(uint64_t);
 523    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
 524       if (ret) {
 525          uint64_t *max_threads_per_block = ret;
 526          *max_threads_per_block = 1024;
 527       }
 528       return sizeof(uint64_t);
 529    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
 530       if (ret) {
 531          uint64_t *max_local_size = ret;
 532          *max_local_size = 32768;
 533       }
 534       return sizeof(uint64_t);
 535    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
 536       if (ret) {
 537          uint32_t *grid_dim = ret;
 538          *grid_dim = 3;
 539       }
 540       return sizeof(uint32_t);
 541    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
 542       if (ret) {
 543          uint64_t *max_global_size = ret;
 544          *max_global_size = (1ULL << 31);
 545       }
 546       return sizeof(uint64_t);
 547    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
 548       if (ret) {
 549          uint64_t *max_mem_alloc_size = ret;
 550          *max_mem_alloc_size = (1ULL << 31);
 551       }
 552       return sizeof(uint64_t);
 553    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE:
 554       if (ret) {
 555          uint64_t *max_private = ret;
 556          *max_private = (1UL << 31);
 557       }
 558       return sizeof(uint64_t);
 559    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
 560       if (ret) {
 561          uint64_t *max_input = ret;
 562          *max_input = 4096;
 563       }
 564       return sizeof(uint64_t);
 565    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
 566       if (ret) {
 567          uint32_t *images = ret;
 568          *images = 0;
 569       }
 570       return sizeof(uint32_t);
 571    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
 572       return 0;
 573    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
 574       if (ret) {
 575          uint32_t *subgroup_size = ret;
 576          *subgroup_size = 32;
 577       }
 578       return sizeof(uint32_t);
 579    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
 580       if (ret) {
 581          uint32_t *max_compute_units = ret;
 582          *max_compute_units = 8;
 583       }
 584       return sizeof(uint32_t);
 585    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
 586       if (ret) {
 587          uint32_t *max_clock_freq = ret;
 588          *max_clock_freq = 300;
 589       }
 590       return sizeof(uint32_t);
 591    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
 592       if (ret) {
 593          uint32_t *address_bits = ret;
 594          *address_bits = 64;
 595       }
 596       return sizeof(uint32_t);
 597    }
 598    return 0;
 599 }
 600
 601 static const struct nir_shader_compiler_options gallivm_nir_options = {
 602    .lower_scmp = true,
 603    .lower_flrp32 = true,
 604    .lower_flrp64 = true,
 605    .lower_fsat = true,
 606    .lower_bitfield_insert_to_shifts = true,
 607    .lower_bitfield_extract_to_shifts = true,
 608    .lower_sub = true,
 609    .lower_ffma = true,
 610    .lower_fmod = true,
 611    .lower_hadd = true,
 612    .lower_add_sat = true,
 613    .lower_pack_snorm_2x16 = true,
 614    .lower_pack_snorm_4x8 = true,
 615    .lower_pack_unorm_2x16 = true,
 616    .lower_pack_unorm_4x8 = true,
 617    .lower_unpack_snorm_2x16 = true,
 618    .lower_unpack_snorm_4x8 = true,
 619    .lower_unpack_unorm_2x16 = true,
 620    .lower_unpack_unorm_4x8 = true,
 621    .lower_extract_byte = true,
 622    .lower_extract_word = true,
 623    .lower_rotate = true,
 624    .lower_ifind_msb = true,
 625    .max_unroll_iterations = 32,
 626    .use_interpolated_input_intrinsics = true,
 627    .lower_to_scalar = true,
 628 };
 629
 630 static void
 631 llvmpipe_finalize_nir(struct pipe_screen *screen,
 632                       void *nirptr,
 633                       bool optimize)
 634 {
 635    struct nir_shader *nir = (struct nir_shader *)nirptr;
 636    lp_build_opt_nir(nir);
 637 }
 638
 639 static inline const void *
 640 llvmpipe_get_compiler_options(struct pipe_screen *screen,
 641                               enum pipe_shader_ir ir,
 642                               enum pipe_shader_type shader)
 643 {
 644    assert(ir == PIPE_SHADER_IR_NIR);
 645    return &gallivm_nir_options;
 646 }
 647
 648 /**
 649  * Query format support for creating a texture, drawing surface, etc.
 650  * \param format  the format to test
 651  * \param type  one of PIPE_TEXTURE, PIPE_SURFACE
 652  */
 653 static bool
 654 llvmpipe_is_format_supported( struct pipe_screen *_screen,
 655                               enum pipe_format format,
 656                               enum pipe_texture_target target,
 657                               unsigned sample_count,
 658                               unsigned storage_sample_count,
 659                               unsigned bind)
 660 {
 661    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
 662    struct sw_winsys *winsys = screen->winsys;
 663    const struct util_format_description *format_desc;
 664
 665    format_desc = util_format_description(format);
 666    if (!format_desc)
 667       return false;
 668
 669    assert(target == PIPE_BUFFER ||
 670           target == PIPE_TEXTURE_1D ||
 671           target == PIPE_TEXTURE_1D_ARRAY ||
 672           target == PIPE_TEXTURE_2D ||
 673           target == PIPE_TEXTURE_2D_ARRAY ||
 674           target == PIPE_TEXTURE_RECT ||
 675           target == PIPE_TEXTURE_3D ||
 676           target == PIPE_TEXTURE_CUBE ||
 677           target == PIPE_TEXTURE_CUBE_ARRAY);
 678
 679    if (sample_count != 0 && sample_count != 1 && sample_count != 4)
 680       return false;
 681
 682    if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
 683       return false;
 684
 685    if (bind & PIPE_BIND_RENDER_TARGET) {
 686       if (format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) {
 687          /* this is a lie actually other formats COULD exist where we would fail */
 688          if (format_desc->nr_channels < 3)
 689             return false;
 690       }
 691       else if (format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB)
 692          return false;
 693
 694       if (format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN &&
 695           format != PIPE_FORMAT_R11G11B10_FLOAT)
 696          return false;
 697
 698       assert(format_desc->block.width == 1);
 699       assert(format_desc->block.height == 1);
 700
 701       if (format_desc->is_mixed)
 702          return false;
 703
 704       if (!format_desc->is_array && !format_desc->is_bitmask &&
 705           format != PIPE_FORMAT_R11G11B10_FLOAT)
 706          return false;
 707    }
 708
 709    if ((bind & (PIPE_BIND_RENDER_TARGET | PIPE_BIND_SAMPLER_VIEW)) &&
 710        ((bind & PIPE_BIND_DISPLAY_TARGET) == 0)) {
 711       /* Disable all 3-channel formats, where channel size != 32 bits.
 712        * In some cases we run into crashes (in generate_unswizzled_blend()),
 713        * for 3-channel RGB16 variants, there was an apparent LLVM bug.
 714        * In any case, disabling the shallower 3-channel formats avoids a
 715        * number of issues with GL_ARB_copy_image support.
 716        */
 717       if (format_desc->is_array &&
 718           format_desc->nr_channels == 3 &&
 719           format_desc->block.bits != 96) {
 720          return false;
 721       }
 722    }
 723
 724    if (bind & PIPE_BIND_DISPLAY_TARGET) {
 725       if(!winsys->is_displaytarget_format_supported(winsys, bind, format))
 726          return false;
 727    }
 728
 729    if (bind & PIPE_BIND_DEPTH_STENCIL) {
 730       if (format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
 731          return false;
 732
 733       if (format_desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
 734          return false;
 735    }
 736
 737    if (format_desc->layout == UTIL_FORMAT_LAYOUT_ASTC ||
 738        format_desc->layout == UTIL_FORMAT_LAYOUT_ATC ||
 739        format_desc->layout == UTIL_FORMAT_LAYOUT_FXT1) {
 740       /* Software decoding is not hooked up. */
 741       return false;
 742    }
 743
 744    if (format_desc->layout == UTIL_FORMAT_LAYOUT_ETC &&
 745        format != PIPE_FORMAT_ETC1_RGB8)
 746       return false;
 747
 748    /*
 749     * Everything can be supported by u_format
 750     * (those without fetch_rgba_float might be not but shouldn't hit that)
 751     */
 752
 753    return true;
 754 }
 755
 756
 757
 758
 759 static void
 760 llvmpipe_flush_frontbuffer(struct pipe_screen *_screen,
 761                            struct pipe_resource *resource,
 762                            unsigned level, unsigned layer,
 763                            void *context_private,
 764                            struct pipe_box *sub_box)
 765 {
 766    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
 767    struct sw_winsys *winsys = screen->winsys;
 768    struct llvmpipe_resource *texture = llvmpipe_resource(resource);
 769
 770    assert(texture->dt);
 771    if (texture->dt)
 772       winsys->displaytarget_display(winsys, texture->dt, context_private, sub_box);
 773 }
 774
 775 static void
 776 llvmpipe_destroy_screen( struct pipe_screen *_screen )
 777 {
 778    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
 779    struct sw_winsys *winsys = screen->winsys;
 780
 781    if (screen->cs_tpool)
 782       lp_cs_tpool_destroy(screen->cs_tpool);
 783
 784    if (screen->rast)
 785       lp_rast_destroy(screen->rast);
 786
 787    lp_jit_screen_cleanup(screen);
 788
 789    if(winsys->destroy)
 790       winsys->destroy(winsys);
 791
 792    glsl_type_singleton_decref();
 793
 794    mtx_destroy(&screen->rast_mutex);
 795    mtx_destroy(&screen->cs_mutex);
 796    FREE(screen);
 797 }
 798
 799
 800
 801
 802 /**
 803  * Fence reference counting.
 804  */
 805 static void
 806 llvmpipe_fence_reference(struct pipe_screen *screen,
 807                          struct pipe_fence_handle **ptr,
 808                          struct pipe_fence_handle *fence)
 809 {
 810    struct lp_fence **old = (struct lp_fence **) ptr;
 811    struct lp_fence *f = (struct lp_fence *) fence;
 812
 813    lp_fence_reference(old, f);
 814 }
 815
 816
 817 /**
 818  * Wait for the fence to finish.
 819  */
 820 static bool
 821 llvmpipe_fence_finish(struct pipe_screen *screen,
 822                       struct pipe_context *ctx,
 823                       struct pipe_fence_handle *fence_handle,
 824                       uint64_t timeout)
 825 {
 826    struct lp_fence *f = (struct lp_fence *) fence_handle;
 827
 828    if (!timeout)
 829       return lp_fence_signalled(f);
 830
 831    if (!lp_fence_signalled(f)) {
 832       if (timeout != PIPE_TIMEOUT_INFINITE)
 833          return lp_fence_timedwait(f, timeout);
 834
 835       lp_fence_wait(f);
 836    }
 837    return true;
 838 }
 839
 840 static uint64_t
 841 llvmpipe_get_timestamp(struct pipe_screen *_screen)
 842 {
 843    return os_time_get_nano();
 844 }
 845
 846 /**
 847  * Create a new pipe_screen object
 848  * Note: we're not presently subclassing pipe_screen (no llvmpipe_screen).
 849  */
 850 struct pipe_screen *
 851 llvmpipe_create_screen(struct sw_winsys *winsys)
 852 {
 853    struct llvmpipe_screen *screen;
 854
 855    util_cpu_detect();
 856
 857    glsl_type_singleton_init_or_ref();
 858
 859 #ifdef DEBUG
 860    LP_DEBUG = debug_get_flags_option("LP_DEBUG", lp_debug_flags, 0 );
 861 #endif
 862
 863    LP_PERF = debug_get_flags_option("LP_PERF", lp_perf_flags, 0 );
 864
 865    screen = CALLOC_STRUCT(llvmpipe_screen);
 866    if (!screen)
 867       return NULL;
 868
 869    if (!lp_jit_screen_init(screen)) {
 870       FREE(screen);
 871       return NULL;
 872    }
 873
 874    screen->winsys = winsys;
 875
 876    screen->base.destroy = llvmpipe_destroy_screen;
 877
 878    screen->base.get_name = llvmpipe_get_name;
 879    screen->base.get_vendor = llvmpipe_get_vendor;
 880    screen->base.get_device_vendor = llvmpipe_get_vendor; // TODO should be the CPU vendor
 881    screen->base.get_param = llvmpipe_get_param;
 882    screen->base.get_shader_param = llvmpipe_get_shader_param;
 883    screen->base.get_compute_param = llvmpipe_get_compute_param;
 884    screen->base.get_paramf = llvmpipe_get_paramf;
 885    screen->base.get_compiler_options = llvmpipe_get_compiler_options;
 886    screen->base.is_format_supported = llvmpipe_is_format_supported;
 887
 888    screen->base.context_create = llvmpipe_create_context;
 889    screen->base.flush_frontbuffer = llvmpipe_flush_frontbuffer;
 890    screen->base.fence_reference = llvmpipe_fence_reference;
 891    screen->base.fence_finish = llvmpipe_fence_finish;
 892
 893    screen->base.get_timestamp = llvmpipe_get_timestamp;
 894
 895    screen->base.finalize_nir = llvmpipe_finalize_nir;
 896    llvmpipe_init_screen_resource_funcs(&screen->base);
 897
 898    screen->use_tgsi = (LP_DEBUG & DEBUG_TGSI_IR);
 899    screen->num_threads = util_cpu_caps.nr_cpus > 1 ? util_cpu_caps.nr_cpus : 0;
 900 #ifdef EMBEDDED_DEVICE
 901    screen->num_threads = 0;
 902 #endif
 903    screen->num_threads = debug_get_num_option("LP_NUM_THREADS", screen->num_threads);
 904    screen->num_threads = MIN2(screen->num_threads, LP_MAX_THREADS);
 905
 906    screen->rast = lp_rast_create(screen->num_threads);
 907    if (!screen->rast) {
 908       lp_jit_screen_cleanup(screen);
 909       FREE(screen);
 910       return NULL;
 911    }
 912    (void) mtx_init(&screen->rast_mutex, mtx_plain);
 913
 914    screen->cs_tpool = lp_cs_tpool_create(screen->num_threads);
 915    if (!screen->cs_tpool) {
 916       lp_rast_destroy(screen->rast);
 917       lp_jit_screen_cleanup(screen);
 918       FREE(screen);
 919       return NULL;
 920    }
 921    (void) mtx_init(&screen->cs_mutex, mtx_plain);
 922
 923    return &screen->base;
 924 }