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 #include "util/disk_cache.h"
  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 "frontend/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    { "cache_stats", DEBUG_CACHE_STATS, NULL },
  79    DEBUG_NAMED_VALUE_END
  80 };
  81 #endif
  82
  83 int LP_PERF = 0;
  84 static const struct debug_named_value lp_perf_flags[] = {
  85    { "texmem",         PERF_TEX_MEM, NULL },
  86    { "no_mipmap",      PERF_NO_MIPMAPS, NULL },
  87    { "no_linear",      PERF_NO_LINEAR, NULL },
  88    { "no_mip_linear",  PERF_NO_MIP_LINEAR, NULL },
  89    { "no_tex",         PERF_NO_TEX, NULL },
  90    { "no_blend",       PERF_NO_BLEND, NULL },
  91    { "no_depth",       PERF_NO_DEPTH, NULL },
  92    { "no_alphatest",   PERF_NO_ALPHATEST, NULL },
  93    DEBUG_NAMED_VALUE_END
  94 };
  95
  96
  97 static const char *
  98 llvmpipe_get_vendor(struct pipe_screen *screen)
  99 {
 100    return "Mesa/X.org";
 101 }
 102
 103
 104 static const char *
 105 llvmpipe_get_name(struct pipe_screen *screen)
 106 {
 107    static char buf[100];
 108    snprintf(buf, sizeof(buf), "llvmpipe (LLVM " MESA_LLVM_VERSION_STRING ", %u bits)",
 109             lp_native_vector_width );
 110    return buf;
 111 }
 112
 113
 114 static int
 115 llvmpipe_get_param(struct pipe_screen *screen, enum pipe_cap param)
 116 {
 117    switch (param) {
 118    case PIPE_CAP_NPOT_TEXTURES:
 119    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 120    case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
 121       return 1;
 122    case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD:
 123    case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES:
 124    case PIPE_CAP_VERTEX_SHADER_SATURATE:
 125       return 1;
 126    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 127       return 1;
 128    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 129       return PIPE_MAX_SO_BUFFERS;
 130    case PIPE_CAP_POINT_SPRITE:
 131       return 1;
 132    case PIPE_CAP_MAX_RENDER_TARGETS:
 133       return PIPE_MAX_COLOR_BUFS;
 134    case PIPE_CAP_OCCLUSION_QUERY:
 135    case PIPE_CAP_QUERY_TIMESTAMP:
 136       return 1;
 137    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 138       return 1;
 139    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 140    case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
 141       return 1;
 142    case PIPE_CAP_TEXTURE_SWIZZLE:
 143       return 1;
 144    case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
 145       return 1 << (LP_MAX_TEXTURE_2D_LEVELS - 1);
 146    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 147       return LP_MAX_TEXTURE_3D_LEVELS;
 148    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 149       return LP_MAX_TEXTURE_CUBE_LEVELS;
 150    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 151       return LP_MAX_TEXTURE_ARRAY_LAYERS;
 152    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 153       return 1;
 154    case PIPE_CAP_INDEP_BLEND_ENABLE:
 155       return 1;
 156    case PIPE_CAP_INDEP_BLEND_FUNC:
 157       return 1;
 158    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 159    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 160    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 161       return 1;
 162    case PIPE_CAP_PRIMITIVE_RESTART:
 163    case PIPE_CAP_PRIMITIVE_RESTART_FIXED_INDEX:
 164       return 1;
 165    case PIPE_CAP_DEPTH_CLIP_DISABLE:
 166       return 1;
 167    case PIPE_CAP_SHADER_STENCIL_EXPORT:
 168       return 1;
 169    case PIPE_CAP_TGSI_INSTANCEID:
 170    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 171    case PIPE_CAP_START_INSTANCE:
 172    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 173       return 1;
 174    case PIPE_CAP_SEAMLESS_CUBE_MAP:
 175    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 176       return 1;
 177    /* this is a lie could support arbitrary large offsets */
 178    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 179    case PIPE_CAP_MIN_TEXEL_OFFSET:
 180       return -32;
 181    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 182    case PIPE_CAP_MAX_TEXEL_OFFSET:
 183       return 31;
 184    case PIPE_CAP_CONDITIONAL_RENDER:
 185    case PIPE_CAP_TEXTURE_BARRIER:
 186       return 1;
 187    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 188    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 189       return 16*4;
 190    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 191    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 192       return 1024;
 193    case PIPE_CAP_MAX_VERTEX_STREAMS: {
 194       struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 195       return lscreen->use_tgsi ? 1 : 4;
 196    }
 197    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 198       return 2048;
 199    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 200    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 201    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 202    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 203       return 1;
 204    case PIPE_CAP_GLSL_FEATURE_LEVEL: {
 205       struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 206       return lscreen->use_tgsi ? 330 : 420;
 207    }
 208    case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
 209       return 140;
 210    case PIPE_CAP_COMPUTE:
 211       return GALLIVM_HAVE_CORO;
 212    case PIPE_CAP_USER_VERTEX_BUFFERS:
 213       return 1;
 214    case PIPE_CAP_TGSI_TEXCOORD:
 215    case PIPE_CAP_DRAW_INDIRECT:
 216       return 1;
 217
 218    case PIPE_CAP_CUBE_MAP_ARRAY:
 219       return 1;
 220    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 221       return 16;
 222    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 223       return 64;
 224    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 225       return 1;
 226    case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 227       return 65536;
 228    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 229       return 16;
 230    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 231       return 0;
 232    case PIPE_CAP_MAX_VIEWPORTS:
 233       return PIPE_MAX_VIEWPORTS;
 234    case PIPE_CAP_ENDIANNESS:
 235       return PIPE_ENDIAN_NATIVE;
 236    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 237       return 1;
 238    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 239       return 1;
 240    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 241       return 4;
 242    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 243       return 1;
 244    case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 245       return 1;
 246    case PIPE_CAP_TGSI_TEX_TXF_LZ:
 247    case PIPE_CAP_SAMPLER_VIEW_TARGET:
 248       return 1;
 249    case PIPE_CAP_FAKE_SW_MSAA: {
 250       struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 251       return lscreen->use_tgsi ? 1 : 0;
 252    }
 253    case PIPE_CAP_TEXTURE_QUERY_LOD:
 254    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 255    case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 256    case PIPE_CAP_DOUBLES:
 257    case PIPE_CAP_INT64:
 258    case PIPE_CAP_INT64_DIVMOD:
 259    case PIPE_CAP_QUERY_SO_OVERFLOW:
 260    case PIPE_CAP_TGSI_DIV:
 261       return 1;
 262
 263    case PIPE_CAP_VENDOR_ID:
 264       return 0xFFFFFFFF;
 265    case PIPE_CAP_DEVICE_ID:
 266       return 0xFFFFFFFF;
 267    case PIPE_CAP_ACCELERATED:
 268       return 0;
 269    case PIPE_CAP_VIDEO_MEMORY: {
 270       /* XXX: Do we want to return the full amount fo system memory ? */
 271       uint64_t system_memory;
 272
 273       if (!os_get_total_physical_memory(&system_memory))
 274          return 0;
 275
 276       if (sizeof(void *) == 4)
 277          /* Cap to 2 GB on 32 bits system. We do this because llvmpipe does
 278           * eat application memory, which is quite limited on 32 bits. App
 279           * shouldn't expect too much available memory. */
 280          system_memory = MIN2(system_memory, 2048 << 20);
 281
 282       return (int)(system_memory >> 20);
 283    }
 284    case PIPE_CAP_UMA:
 285       return 0;
 286    case PIPE_CAP_CLIP_HALFZ:
 287       return 1;
 288    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 289    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 290    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 291       return 1;
 292    case PIPE_CAP_CULL_DISTANCE:
 293       return 1;
 294    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 295       return 1;
 296    case PIPE_CAP_CLEAR_TEXTURE:
 297       return 1;
 298    case PIPE_CAP_MAX_VARYINGS:
 299       return 32;
 300    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 301       return 16;
 302    case PIPE_CAP_QUERY_BUFFER_OBJECT:
 303       return 1;
 304    case PIPE_CAP_DRAW_PARAMETERS:
 305       return 1;
 306    case PIPE_CAP_FBFETCH:
 307       return 8;
 308    case PIPE_CAP_FBFETCH_COHERENT:
 309       return 0;
 310    case PIPE_CAP_MULTI_DRAW_INDIRECT:
 311    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 312       return 1;
 313    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 314       return 32;
 315    case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
 316       return 8;
 317    case PIPE_CAP_PCI_GROUP:
 318    case PIPE_CAP_PCI_BUS:
 319    case PIPE_CAP_PCI_DEVICE:
 320    case PIPE_CAP_PCI_FUNCTION:
 321    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 322    case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 323       return 0;
 324    case PIPE_CAP_MAX_GS_INVOCATIONS:
 325       return 32;
 326    case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
 327       return LP_MAX_TGSI_SHADER_BUFFER_SIZE;
 328    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 329    case PIPE_CAP_TGSI_TG4_COMPONENT_IN_SWIZZLE:
 330    case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 331       return 1;
 332    case PIPE_CAP_TGSI_TXQS:
 333    case PIPE_CAP_TGSI_VOTE:
 334    case PIPE_CAP_LOAD_CONSTBUF:
 335    case PIPE_CAP_TEXTURE_MULTISAMPLE:
 336    case PIPE_CAP_SAMPLE_SHADING:
 337    case PIPE_CAP_POST_DEPTH_COVERAGE:
 338    case PIPE_CAP_PACKED_UNIFORMS: {
 339       struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 340       return !lscreen->use_tgsi;
 341    }
 342    default:
 343       return u_pipe_screen_get_param_defaults(screen, param);
 344    }
 345 }
 346
 347 static int
 348 llvmpipe_get_shader_param(struct pipe_screen *screen,
 349                           enum pipe_shader_type shader,
 350                           enum pipe_shader_cap param)
 351 {
 352    struct llvmpipe_screen *lscreen = llvmpipe_screen(screen);
 353    switch(shader)
 354    {
 355    case PIPE_SHADER_COMPUTE:
 356       if ((LP_DEBUG & DEBUG_CL) && param == PIPE_SHADER_CAP_SUPPORTED_IRS)
 357          return (1 << PIPE_SHADER_IR_TGSI) | (1 << PIPE_SHADER_IR_NIR) | (1 << PIPE_SHADER_IR_NIR_SERIALIZED);
 358       /* fallthrough */
 359    case PIPE_SHADER_FRAGMENT:
 360       if (param == PIPE_SHADER_CAP_PREFERRED_IR) {
 361          if (lscreen->use_tgsi)
 362             return PIPE_SHADER_IR_TGSI;
 363          else
 364             return PIPE_SHADER_IR_NIR;
 365       }
 366       switch (param) {
 367       default:
 368          return gallivm_get_shader_param(param);
 369       }
 370       /* fallthrough */
 371    case PIPE_SHADER_TESS_CTRL:
 372    case PIPE_SHADER_TESS_EVAL:
 373       /* Tessellation shader needs llvm coroutines support */
 374       if (!GALLIVM_HAVE_CORO || lscreen->use_tgsi)
 375          return 0;
 376       /* fallthrough */
 377    case PIPE_SHADER_VERTEX:
 378    case PIPE_SHADER_GEOMETRY:
 379       if (param == PIPE_SHADER_CAP_PREFERRED_IR) {
 380          if (lscreen->use_tgsi)
 381             return PIPE_SHADER_IR_TGSI;
 382          else
 383             return PIPE_SHADER_IR_NIR;
 384       }
 385
 386       switch (param) {
 387       case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 388          /* At this time, the draw module and llvmpipe driver only
 389           * support vertex shader texture lookups when LLVM is enabled in
 390           * the draw module.
 391           */
 392          if (debug_get_bool_option("DRAW_USE_LLVM", TRUE))
 393             return PIPE_MAX_SAMPLERS;
 394          else
 395             return 0;
 396       case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 397          if (debug_get_bool_option("DRAW_USE_LLVM", TRUE))
 398             return PIPE_MAX_SHADER_SAMPLER_VIEWS;
 399          else
 400             return 0;
 401       default:
 402          return draw_get_shader_param(shader, param);
 403       }
 404    default:
 405       return 0;
 406    }
 407 }
 408
 409 static float
 410 llvmpipe_get_paramf(struct pipe_screen *screen, enum pipe_capf param)
 411 {
 412    switch (param) {
 413    case PIPE_CAPF_MAX_LINE_WIDTH:
 414       /* fall-through */
 415    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 416       return 255.0; /* arbitrary */
 417    case PIPE_CAPF_MAX_POINT_WIDTH:
 418       /* fall-through */
 419    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 420       return LP_MAX_POINT_WIDTH; /* arbitrary */
 421    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 422       return 16.0; /* not actually signficant at this time */
 423    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 424       return 16.0; /* arbitrary */
 425    case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
 426       return 0.0;
 427    case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
 428       return 0.0;
 429    case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
 430       return 0.0;
 431    }
 432    /* should only get here on unhandled cases */
 433    debug_printf("Unexpected PIPE_CAP %d query\n", param);
 434    return 0.0;
 435 }
 436
 437 static int
 438 llvmpipe_get_compute_param(struct pipe_screen *_screen,
 439                            enum pipe_shader_ir ir_type,
 440                            enum pipe_compute_cap param,
 441                            void *ret)
 442 {
 443    switch (param) {
 444    case PIPE_COMPUTE_CAP_IR_TARGET:
 445       return 0;
 446    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
 447       if (ret) {
 448          uint64_t *grid_size = ret;
 449          grid_size[0] = 65535;
 450          grid_size[1] = 65535;
 451          grid_size[2] = 65535;
 452       }
 453       return 3 * sizeof(uint64_t) ;
 454    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
 455       if (ret) {
 456          uint64_t *block_size = ret;
 457          block_size[0] = 1024;
 458          block_size[1] = 1024;
 459          block_size[2] = 1024;
 460       }
 461       return 3 * sizeof(uint64_t);
 462    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
 463       if (ret) {
 464          uint64_t *max_threads_per_block = ret;
 465          *max_threads_per_block = 1024;
 466       }
 467       return sizeof(uint64_t);
 468    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
 469       if (ret) {
 470          uint64_t *max_local_size = ret;
 471          *max_local_size = 32768;
 472       }
 473       return sizeof(uint64_t);
 474    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
 475       if (ret) {
 476          uint32_t *grid_dim = ret;
 477          *grid_dim = 3;
 478       }
 479       return sizeof(uint32_t);
 480    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
 481       if (ret) {
 482          uint64_t *max_global_size = ret;
 483          *max_global_size = (1ULL << 31);
 484       }
 485       return sizeof(uint64_t);
 486    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
 487       if (ret) {
 488          uint64_t *max_mem_alloc_size = ret;
 489          *max_mem_alloc_size = (1ULL << 31);
 490       }
 491       return sizeof(uint64_t);
 492    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE:
 493       if (ret) {
 494          uint64_t *max_private = ret;
 495          *max_private = (1UL << 31);
 496       }
 497       return sizeof(uint64_t);
 498    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
 499       if (ret) {
 500          uint64_t *max_input = ret;
 501          *max_input = 4096;
 502       }
 503       return sizeof(uint64_t);
 504    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
 505       if (ret) {
 506          uint32_t *images = ret;
 507          *images = 0;
 508       }
 509       return sizeof(uint32_t);
 510    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
 511       return 0;
 512    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
 513       if (ret) {
 514          uint32_t *subgroup_size = ret;
 515          *subgroup_size = 32;
 516       }
 517       return sizeof(uint32_t);
 518    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
 519       if (ret) {
 520          uint32_t *max_compute_units = ret;
 521          *max_compute_units = 8;
 522       }
 523       return sizeof(uint32_t);
 524    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
 525       if (ret) {
 526          uint32_t *max_clock_freq = ret;
 527          *max_clock_freq = 300;
 528       }
 529       return sizeof(uint32_t);
 530    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
 531       if (ret) {
 532          uint32_t *address_bits = ret;
 533          *address_bits = 64;
 534       }
 535       return sizeof(uint32_t);
 536    }
 537    return 0;
 538 }
 539
 540 static const struct nir_shader_compiler_options gallivm_nir_options = {
 541    .lower_scmp = true,
 542    .lower_flrp32 = true,
 543    .lower_flrp64 = true,
 544    .lower_fsat = true,
 545    .lower_bitfield_insert_to_shifts = true,
 546    .lower_bitfield_extract_to_shifts = true,
 547    .lower_sub = true,
 548    .lower_ffma = true,
 549    .lower_fmod = true,
 550    .lower_hadd = true,
 551    .lower_add_sat = true,
 552    .lower_pack_snorm_2x16 = true,
 553    .lower_pack_snorm_4x8 = true,
 554    .lower_pack_unorm_2x16 = true,
 555    .lower_pack_unorm_4x8 = true,
 556    .lower_unpack_snorm_2x16 = true,
 557    .lower_unpack_snorm_4x8 = true,
 558    .lower_unpack_unorm_2x16 = true,
 559    .lower_unpack_unorm_4x8 = true,
 560    .lower_extract_byte = true,
 561    .lower_extract_word = true,
 562    .lower_rotate = true,
 563    .lower_ifind_msb = true,
 564    .max_unroll_iterations = 32,
 565    .use_interpolated_input_intrinsics = true,
 566    .lower_to_scalar = true,
 567 };
 568
 569 static void
 570 llvmpipe_finalize_nir(struct pipe_screen *screen,
 571                       void *nirptr,
 572                       bool optimize)
 573 {
 574    struct nir_shader *nir = (struct nir_shader *)nirptr;
 575    lp_build_opt_nir(nir);
 576 }
 577
 578 static inline const void *
 579 llvmpipe_get_compiler_options(struct pipe_screen *screen,
 580                               enum pipe_shader_ir ir,
 581                               enum pipe_shader_type shader)
 582 {
 583    assert(ir == PIPE_SHADER_IR_NIR);
 584    return &gallivm_nir_options;
 585 }
 586
 587 /**
 588  * Query format support for creating a texture, drawing surface, etc.
 589  * \param format  the format to test
 590  * \param type  one of PIPE_TEXTURE, PIPE_SURFACE
 591  */
 592 static bool
 593 llvmpipe_is_format_supported( struct pipe_screen *_screen,
 594                               enum pipe_format format,
 595                               enum pipe_texture_target target,
 596                               unsigned sample_count,
 597                               unsigned storage_sample_count,
 598                               unsigned bind)
 599 {
 600    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
 601    struct sw_winsys *winsys = screen->winsys;
 602    const struct util_format_description *format_desc;
 603
 604    format_desc = util_format_description(format);
 605    if (!format_desc)
 606       return false;
 607
 608    assert(target == PIPE_BUFFER ||
 609           target == PIPE_TEXTURE_1D ||
 610           target == PIPE_TEXTURE_1D_ARRAY ||
 611           target == PIPE_TEXTURE_2D ||
 612           target == PIPE_TEXTURE_2D_ARRAY ||
 613           target == PIPE_TEXTURE_RECT ||
 614           target == PIPE_TEXTURE_3D ||
 615           target == PIPE_TEXTURE_CUBE ||
 616           target == PIPE_TEXTURE_CUBE_ARRAY);
 617
 618    if (sample_count != 0 && sample_count != 1 && sample_count != 4)
 619       return false;
 620
 621    if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
 622       return false;
 623
 624    if (bind & PIPE_BIND_RENDER_TARGET) {
 625       if (format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) {
 626          /* this is a lie actually other formats COULD exist where we would fail */
 627          if (format_desc->nr_channels < 3)
 628             return false;
 629       }
 630       else if (format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB)
 631          return false;
 632
 633       if (format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN &&
 634           format != PIPE_FORMAT_R11G11B10_FLOAT)
 635          return false;
 636
 637       assert(format_desc->block.width == 1);
 638       assert(format_desc->block.height == 1);
 639
 640       if (format_desc->is_mixed)
 641          return false;
 642
 643       if (!format_desc->is_array && !format_desc->is_bitmask &&
 644           format != PIPE_FORMAT_R11G11B10_FLOAT)
 645          return false;
 646    }
 647
 648    if ((bind & (PIPE_BIND_RENDER_TARGET | PIPE_BIND_SAMPLER_VIEW)) &&
 649        ((bind & PIPE_BIND_DISPLAY_TARGET) == 0)) {
 650       /* Disable all 3-channel formats, where channel size != 32 bits.
 651        * In some cases we run into crashes (in generate_unswizzled_blend()),
 652        * for 3-channel RGB16 variants, there was an apparent LLVM bug.
 653        * In any case, disabling the shallower 3-channel formats avoids a
 654        * number of issues with GL_ARB_copy_image support.
 655        */
 656       if (format_desc->is_array &&
 657           format_desc->nr_channels == 3 &&
 658           format_desc->block.bits != 96) {
 659          return false;
 660       }
 661    }
 662
 663    if (bind & PIPE_BIND_DISPLAY_TARGET) {
 664       if(!winsys->is_displaytarget_format_supported(winsys, bind, format))
 665          return false;
 666    }
 667
 668    if (bind & PIPE_BIND_DEPTH_STENCIL) {
 669       if (format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
 670          return false;
 671
 672       if (format_desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
 673          return false;
 674    }
 675
 676    if (format_desc->layout == UTIL_FORMAT_LAYOUT_ASTC ||
 677        format_desc->layout == UTIL_FORMAT_LAYOUT_ATC ||
 678        format_desc->layout == UTIL_FORMAT_LAYOUT_FXT1) {
 679       /* Software decoding is not hooked up. */
 680       return false;
 681    }
 682
 683    if (format_desc->layout == UTIL_FORMAT_LAYOUT_ETC &&
 684        format != PIPE_FORMAT_ETC1_RGB8)
 685       return false;
 686
 687    /*
 688     * Everything can be supported by u_format
 689     * (those without fetch_rgba_float might be not but shouldn't hit that)
 690     */
 691
 692    return true;
 693 }
 694
 695
 696
 697
 698 static void
 699 llvmpipe_flush_frontbuffer(struct pipe_screen *_screen,
 700                            struct pipe_resource *resource,
 701                            unsigned level, unsigned layer,
 702                            void *context_private,
 703                            struct pipe_box *sub_box)
 704 {
 705    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
 706    struct sw_winsys *winsys = screen->winsys;
 707    struct llvmpipe_resource *texture = llvmpipe_resource(resource);
 708
 709    assert(texture->dt);
 710    if (texture->dt)
 711       winsys->displaytarget_display(winsys, texture->dt, context_private, sub_box);
 712 }
 713
 714 static void
 715 llvmpipe_destroy_screen( struct pipe_screen *_screen )
 716 {
 717    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
 718    struct sw_winsys *winsys = screen->winsys;
 719
 720    if (screen->cs_tpool)
 721       lp_cs_tpool_destroy(screen->cs_tpool);
 722
 723    if (screen->rast)
 724       lp_rast_destroy(screen->rast);
 725
 726    lp_jit_screen_cleanup(screen);
 727
 728    if (LP_DEBUG & DEBUG_CACHE_STATS)
 729       printf("disk shader cache:   hits = %u, misses = %u\n", screen->num_disk_shader_cache_hits,
 730              screen->num_disk_shader_cache_misses);
 731    disk_cache_destroy(screen->disk_shader_cache);
 732    if(winsys->destroy)
 733       winsys->destroy(winsys);
 734
 735    glsl_type_singleton_decref();
 736
 737    mtx_destroy(&screen->rast_mutex);
 738    mtx_destroy(&screen->cs_mutex);
 739    FREE(screen);
 740 }
 741
 742
 743
 744
 745 /**
 746  * Fence reference counting.
 747  */
 748 static void
 749 llvmpipe_fence_reference(struct pipe_screen *screen,
 750                          struct pipe_fence_handle **ptr,
 751                          struct pipe_fence_handle *fence)
 752 {
 753    struct lp_fence **old = (struct lp_fence **) ptr;
 754    struct lp_fence *f = (struct lp_fence *) fence;
 755
 756    lp_fence_reference(old, f);
 757 }
 758
 759
 760 /**
 761  * Wait for the fence to finish.
 762  */
 763 static bool
 764 llvmpipe_fence_finish(struct pipe_screen *screen,
 765                       struct pipe_context *ctx,
 766                       struct pipe_fence_handle *fence_handle,
 767                       uint64_t timeout)
 768 {
 769    struct lp_fence *f = (struct lp_fence *) fence_handle;
 770
 771    if (!timeout)
 772       return lp_fence_signalled(f);
 773
 774    if (!lp_fence_signalled(f)) {
 775       if (timeout != PIPE_TIMEOUT_INFINITE)
 776          return lp_fence_timedwait(f, timeout);
 777
 778       lp_fence_wait(f);
 779    }
 780    return true;
 781 }
 782
 783 static uint64_t
 784 llvmpipe_get_timestamp(struct pipe_screen *_screen)
 785 {
 786    return os_time_get_nano();
 787 }
 788
 789 static void lp_disk_cache_create(struct llvmpipe_screen *screen)
 790 {
 791    struct mesa_sha1 ctx;
 792    unsigned char sha1[20];
 793    char cache_id[20 * 2 + 1];
 794    _mesa_sha1_init(&ctx);
 795
 796    if (!disk_cache_get_function_identifier(lp_disk_cache_create, &ctx) ||
 797        !disk_cache_get_function_identifier(LLVMLinkInMCJIT, &ctx))
 798       return;
 799
 800    _mesa_sha1_final(&ctx, sha1);
 801    disk_cache_format_hex_id(cache_id, sha1, 20 * 2);
 802
 803    screen->disk_shader_cache = disk_cache_create("llvmpipe", cache_id, 0);
 804 }
 805
 806 static struct disk_cache *lp_get_disk_shader_cache(struct pipe_screen *_screen)
 807 {
 808    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
 809
 810    return screen->disk_shader_cache;
 811 }
 812
 813 void lp_disk_cache_find_shader(struct llvmpipe_screen *screen,
 814                                struct lp_cached_code *cache,
 815                                unsigned char ir_sha1_cache_key[20])
 816 {
 817    unsigned char sha1[CACHE_KEY_SIZE];
 818
 819    if (!screen->disk_shader_cache)
 820       return;
 821    disk_cache_compute_key(screen->disk_shader_cache, ir_sha1_cache_key, 20, sha1);
 822
 823    size_t binary_size;
 824    uint8_t *buffer = disk_cache_get(screen->disk_shader_cache, sha1, &binary_size);
 825    if (!buffer) {
 826       cache->data_size = 0;
 827       p_atomic_inc(&screen->num_disk_shader_cache_misses);
 828       return;
 829    }
 830    cache->data_size = binary_size;
 831    cache->data = buffer;
 832    p_atomic_inc(&screen->num_disk_shader_cache_hits);
 833 }
 834
 835 void lp_disk_cache_insert_shader(struct llvmpipe_screen *screen,
 836                                  struct lp_cached_code *cache,
 837                                  unsigned char ir_sha1_cache_key[20])
 838 {
 839    unsigned char sha1[CACHE_KEY_SIZE];
 840
 841    if (!screen->disk_shader_cache || !cache->data_size || cache->dont_cache)
 842       return;
 843    disk_cache_compute_key(screen->disk_shader_cache, ir_sha1_cache_key, 20, sha1);
 844    disk_cache_put(screen->disk_shader_cache, sha1, cache->data, cache->data_size, NULL);
 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
 897    screen->base.get_disk_shader_cache = lp_get_disk_shader_cache;
 898    llvmpipe_init_screen_resource_funcs(&screen->base);
 899
 900    screen->use_tgsi = (LP_DEBUG & DEBUG_TGSI_IR);
 901    screen->num_threads = util_cpu_caps.nr_cpus > 1 ? util_cpu_caps.nr_cpus : 0;
 902 #ifdef EMBEDDED_DEVICE
 903    screen->num_threads = 0;
 904 #endif
 905    screen->num_threads = debug_get_num_option("LP_NUM_THREADS", screen->num_threads);
 906    screen->num_threads = MIN2(screen->num_threads, LP_MAX_THREADS);
 907
 908    screen->rast = lp_rast_create(screen->num_threads);
 909    if (!screen->rast) {
 910       lp_jit_screen_cleanup(screen);
 911       FREE(screen);
 912       return NULL;
 913    }
 914    (void) mtx_init(&screen->rast_mutex, mtx_plain);
 915
 916    screen->cs_tpool = lp_cs_tpool_create(screen->num_threads);
 917    if (!screen->cs_tpool) {
 918       lp_rast_destroy(screen->rast);
 919       lp_jit_screen_cleanup(screen);
 920       FREE(screen);
 921       return NULL;
 922    }
 923    (void) mtx_init(&screen->cs_mutex, mtx_plain);
 924
 925    lp_disk_cache_create(screen);
 926    return &screen->base;
 927 }