src/gallium/drivers/freedreno/freedreno_screen.c

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29
  30 #include "pipe/p_defines.h"
  31 #include "pipe/p_screen.h"
  32 #include "pipe/p_state.h"
  33
  34 #include "util/u_memory.h"
  35 #include "util/u_inlines.h"
  36 #include "util/u_format.h"
  37 #include "util/u_format_s3tc.h"
  38 #include "util/u_string.h"
  39 #include "util/u_debug.h"
  40
  41 #include "os/os_time.h"
  42
  43 #include <stdio.h>
  44 #include <errno.h>
  45 #include <stdlib.h>
  46
  47 #include "freedreno_screen.h"
  48 #include "freedreno_resource.h"
  49 #include "freedreno_fence.h"
  50 #include "freedreno_query.h"
  51 #include "freedreno_util.h"
  52
  53 #include "fd2_screen.h"
  54 #include "fd3_screen.h"
  55
  56 /* XXX this should go away */
  57 #include "state_tracker/drm_driver.h"
  58
  59 static const struct debug_named_value debug_options[] = {
  60                 {"msgs",      FD_DBG_MSGS,   "Print debug messages"},
  61                 {"disasm",    FD_DBG_DISASM, "Dump TGSI and adreno shader disassembly"},
  62                 {"dclear",    FD_DBG_DCLEAR, "Mark all state dirty after clear"},
  63                 {"dgmem",     FD_DBG_DGMEM,  "Mark all state dirty after GMEM tile pass"},
  64                 {"dscis",     FD_DBG_DSCIS,  "Disable scissor optimization"},
  65                 {"direct",    FD_DBG_DIRECT, "Force inline (SS_DIRECT) state loads"},
  66                 {"dbypass",   FD_DBG_DBYPASS,"Disable GMEM bypass"},
  67                 {"fraghalf",  FD_DBG_FRAGHALF, "Use half-precision in fragment shader"},
  68                 {"nobin",     FD_DBG_NOBIN,  "Disable hw binning"},
  69                 {"noopt",     FD_DBG_NOOPT , "Disable optimization passes in compiler"},
  70                 {"optmsgs",   FD_DBG_OPTMSGS,"Enable optimizater debug messages"},
  71                 {"optdump",   FD_DBG_OPTDUMP,"Dump shader DAG to .dot files"},
  72                 DEBUG_NAMED_VALUE_END
  73 };
  74
  75 DEBUG_GET_ONCE_FLAGS_OPTION(fd_mesa_debug, "FD_MESA_DEBUG", debug_options, 0)
  76
  77 int fd_mesa_debug = 0;
  78 bool fd_binning_enabled = true;
  79
  80 static const char *
  81 fd_screen_get_name(struct pipe_screen *pscreen)
  82 {
  83         static char buffer[128];
  84         util_snprintf(buffer, sizeof(buffer), "FD%03d",
  85                         fd_screen(pscreen)->device_id);
  86         return buffer;
  87 }
  88
  89 static const char *
  90 fd_screen_get_vendor(struct pipe_screen *pscreen)
  91 {
  92         return "freedreno";
  93 }
  94
  95 static uint64_t
  96 fd_screen_get_timestamp(struct pipe_screen *pscreen)
  97 {
  98         int64_t cpu_time = os_time_get() * 1000;
  99         return cpu_time + fd_screen(pscreen)->cpu_gpu_time_delta;
 100 }
 101
 102 static void
 103 fd_screen_fence_ref(struct pipe_screen *pscreen,
 104                 struct pipe_fence_handle **ptr,
 105                 struct pipe_fence_handle *pfence)
 106 {
 107         fd_fence_ref(fd_fence(pfence), (struct fd_fence **)ptr);
 108 }
 109
 110 static boolean
 111 fd_screen_fence_signalled(struct pipe_screen *screen,
 112                 struct pipe_fence_handle *pfence)
 113 {
 114         return fd_fence_signalled(fd_fence(pfence));
 115 }
 116
 117 static boolean
 118 fd_screen_fence_finish(struct pipe_screen *screen,
 119                 struct pipe_fence_handle *pfence,
 120                 uint64_t timeout)
 121 {
 122         return fd_fence_wait(fd_fence(pfence));
 123 }
 124
 125 static void
 126 fd_screen_destroy(struct pipe_screen *pscreen)
 127 {
 128         struct fd_screen *screen = fd_screen(pscreen);
 129
 130         if (screen->pipe)
 131                 fd_pipe_del(screen->pipe);
 132
 133         if (screen->dev)
 134                 fd_device_del(screen->dev);
 135
 136         free(screen);
 137 }
 138
 139 /*
 140 TODO either move caps to a2xx/a3xx specific code, or maybe have some
 141 tables for things that differ if the delta is not too much..
 142  */
 143 static int
 144 fd_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
 145 {
 146         /* this is probably not totally correct.. but it's a start: */
 147         switch (param) {
 148         /* Supported features (boolean caps). */
 149         case PIPE_CAP_NPOT_TEXTURES:
 150         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 151         case PIPE_CAP_TWO_SIDED_STENCIL:
 152         case PIPE_CAP_ANISOTROPIC_FILTER:
 153         case PIPE_CAP_POINT_SPRITE:
 154         case PIPE_CAP_TEXTURE_SHADOW_MAP:
 155         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 156         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 157         case PIPE_CAP_TEXTURE_SWIZZLE:
 158         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 159         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 160         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 161         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 162         case PIPE_CAP_SM3:
 163         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 164         case PIPE_CAP_PRIMITIVE_RESTART:
 165         case PIPE_CAP_CONDITIONAL_RENDER:
 166         case PIPE_CAP_TEXTURE_BARRIER:
 167         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 168         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 169         case PIPE_CAP_TGSI_INSTANCEID:
 170         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 171         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 172         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 173         case PIPE_CAP_COMPUTE:
 174         case PIPE_CAP_START_INSTANCE:
 175         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 176         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 177         case PIPE_CAP_USER_CONSTANT_BUFFERS:
 178                 return 1;
 179
 180         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 181         case PIPE_CAP_TGSI_TEXCOORD:
 182         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 183                 return 0;
 184
 185         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 186                 return 256;
 187
 188         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 189                 return 120;
 190
 191         /* Unsupported features. */
 192         case PIPE_CAP_INDEP_BLEND_ENABLE:
 193         case PIPE_CAP_INDEP_BLEND_FUNC:
 194         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 195         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 196         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 197         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 198         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 199         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 200         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 201         case PIPE_CAP_USER_VERTEX_BUFFERS:
 202         case PIPE_CAP_USER_INDEX_BUFFERS:
 203         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 204         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 205         case PIPE_CAP_TGSI_VS_LAYER:
 206         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 207         case PIPE_CAP_TEXTURE_GATHER_SM5:
 208         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 209         case PIPE_CAP_FAKE_SW_MSAA:
 210         case PIPE_CAP_TEXTURE_QUERY_LOD:
 211                 return 0;
 212
 213         /* Stream output. */
 214         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 215         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 216         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 217         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 218                 return 0;
 219
 220         /* Geometry shader output, unsupported. */
 221         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 222         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 223                 return 0;
 224
 225         /* Texturing. */
 226         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 227         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 228         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 229                 return MAX_MIP_LEVELS;
 230         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 231                 return 9192;
 232
 233         /* Render targets. */
 234         case PIPE_CAP_MAX_RENDER_TARGETS:
 235                 return 1;
 236
 237         /* Timer queries. */
 238         case PIPE_CAP_QUERY_TIME_ELAPSED:
 239         case PIPE_CAP_OCCLUSION_QUERY:
 240         case PIPE_CAP_QUERY_TIMESTAMP:
 241                 return 0;
 242
 243         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 244         case PIPE_CAP_MIN_TEXEL_OFFSET:
 245                 return -8;
 246
 247         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 248         case PIPE_CAP_MAX_TEXEL_OFFSET:
 249                 return 7;
 250
 251         case PIPE_CAP_ENDIANNESS:
 252                 return PIPE_ENDIAN_LITTLE;
 253
 254         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 255                 return 64;
 256
 257         default:
 258                 DBG("unknown param %d", param);
 259                 return 0;
 260         }
 261 }
 262
 263 static float
 264 fd_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
 265 {
 266         switch (param) {
 267         case PIPE_CAPF_MAX_LINE_WIDTH:
 268         case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 269         case PIPE_CAPF_MAX_POINT_WIDTH:
 270         case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 271                 return 8192.0f;
 272         case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 273                 return 16.0f;
 274         case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 275                 return 16.0f;
 276         case PIPE_CAPF_GUARD_BAND_LEFT:
 277         case PIPE_CAPF_GUARD_BAND_TOP:
 278         case PIPE_CAPF_GUARD_BAND_RIGHT:
 279         case PIPE_CAPF_GUARD_BAND_BOTTOM:
 280                 return 0.0f;
 281         default:
 282                 DBG("unknown paramf %d", param);
 283                 return 0;
 284         }
 285 }
 286
 287 static int
 288 fd_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
 289                 enum pipe_shader_cap param)
 290 {
 291         struct fd_screen *screen = fd_screen(pscreen);
 292
 293         switch(shader)
 294         {
 295         case PIPE_SHADER_FRAGMENT:
 296         case PIPE_SHADER_VERTEX:
 297                 break;
 298         case PIPE_SHADER_COMPUTE:
 299         case PIPE_SHADER_GEOMETRY:
 300                 /* maye we could emulate.. */
 301                 return 0;
 302         default:
 303                 DBG("unknown shader type %d", shader);
 304                 return 0;
 305         }
 306
 307         /* this is probably not totally correct.. but it's a start: */
 308         switch (param) {
 309         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 310         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 311         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 312         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 313                 return 16384;
 314         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 315                 return 8; /* XXX */
 316         case PIPE_SHADER_CAP_MAX_INPUTS:
 317                 return 32;
 318         case PIPE_SHADER_CAP_MAX_TEMPS:
 319                 return 64; /* Max native temporaries. */
 320         case PIPE_SHADER_CAP_MAX_ADDRS:
 321                 return 1; /* Max native address registers */
 322         case PIPE_SHADER_CAP_MAX_CONSTS:
 323                 return (screen->gpu_id >= 300) ? 1024 : 64;
 324         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 325                 return 1;
 326         case PIPE_SHADER_CAP_MAX_PREDS:
 327                 return 0; /* nothing uses this */
 328         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 329                 return 1;
 330         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 331         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 332         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 333         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 334                 return 1;
 335         case PIPE_SHADER_CAP_SUBROUTINES:
 336                 return 0;
 337         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 338                 return 1;
 339         case PIPE_SHADER_CAP_INTEGERS:
 340                 /* we should be able to support this on a3xx, but not
 341                  * implemented yet:
 342                  */
 343                 return 0;
 344         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 345         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 346                 return 16;
 347         case PIPE_SHADER_CAP_PREFERRED_IR:
 348                 return PIPE_SHADER_IR_TGSI;
 349         default:
 350                 DBG("unknown shader param %d", param);
 351                 return 0;
 352         }
 353         return 0;
 354 }
 355
 356 boolean
 357 fd_screen_bo_get_handle(struct pipe_screen *pscreen,
 358                 struct fd_bo *bo,
 359                 unsigned stride,
 360                 struct winsys_handle *whandle)
 361 {
 362         whandle->stride = stride;
 363
 364         if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
 365                 return fd_bo_get_name(bo, &whandle->handle) == 0;
 366         } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
 367                 whandle->handle = fd_bo_handle(bo);
 368                 return TRUE;
 369         } else {
 370                 return FALSE;
 371         }
 372 }
 373
 374 struct fd_bo *
 375 fd_screen_bo_from_handle(struct pipe_screen *pscreen,
 376                 struct winsys_handle *whandle,
 377                 unsigned *out_stride)
 378 {
 379         struct fd_screen *screen = fd_screen(pscreen);
 380         struct fd_bo *bo;
 381
 382         if (whandle->type != DRM_API_HANDLE_TYPE_SHARED) {
 383                 DBG("Attempt to import unsupported handle type %d", whandle->type);
 384                 return NULL;
 385         }
 386
 387         bo = fd_bo_from_name(screen->dev, whandle->handle);
 388         if (!bo) {
 389                 DBG("ref name 0x%08x failed", whandle->handle);
 390                 return NULL;
 391         }
 392
 393         *out_stride = whandle->stride;
 394
 395         return bo;
 396 }
 397
 398 struct pipe_screen *
 399 fd_screen_create(struct fd_device *dev)
 400 {
 401         struct fd_screen *screen = CALLOC_STRUCT(fd_screen);
 402         struct pipe_screen *pscreen;
 403         uint64_t val;
 404
 405         fd_mesa_debug = debug_get_option_fd_mesa_debug();
 406
 407         if (fd_mesa_debug & FD_DBG_NOBIN)
 408                 fd_binning_enabled = false;
 409
 410         if (!screen)
 411                 return NULL;
 412
 413         pscreen = &screen->base;
 414
 415         screen->dev = dev;
 416
 417         // maybe this should be in context?
 418         screen->pipe = fd_pipe_new(screen->dev, FD_PIPE_3D);
 419         if (!screen->pipe) {
 420                 DBG("could not create 3d pipe");
 421                 goto fail;
 422         }
 423
 424         if (fd_pipe_get_param(screen->pipe, FD_GMEM_SIZE, &val)) {
 425                 DBG("could not get GMEM size");
 426                 goto fail;
 427         }
 428         screen->gmemsize_bytes = val;
 429
 430         if (fd_pipe_get_param(screen->pipe, FD_DEVICE_ID, &val)) {
 431                 DBG("could not get device-id");
 432                 goto fail;
 433         }
 434         screen->device_id = val;
 435
 436         if (fd_pipe_get_param(screen->pipe, FD_GPU_ID, &val)) {
 437                 DBG("could not get gpu-id");
 438                 goto fail;
 439         }
 440         screen->gpu_id = val;
 441
 442         /* explicitly checking for GPU revisions that are known to work.  This
 443          * may be overly conservative for a3xx, where spoofing the gpu_id with
 444          * the blob driver seems to generate identical cmdstream dumps.  But
 445          * on a2xx, there seem to be small differences between the GPU revs
 446          * so it is probably better to actually test first on real hardware
 447          * before enabling:
 448          *
 449          * If you have a different adreno version, feel free to add it to one
 450          * of the two cases below and see what happens.  And if it works, please
 451          * send a patch ;-)
 452          */
 453         switch (screen->gpu_id) {
 454         case 220:
 455                 fd2_screen_init(pscreen);
 456                 break;
 457         case 320:
 458         case 330:
 459                 fd3_screen_init(pscreen);
 460                 break;
 461         default:
 462                 debug_printf("unsupported GPU: a%03d\n", screen->gpu_id);
 463                 goto fail;
 464         }
 465
 466         pscreen->destroy = fd_screen_destroy;
 467         pscreen->get_param = fd_screen_get_param;
 468         pscreen->get_paramf = fd_screen_get_paramf;
 469         pscreen->get_shader_param = fd_screen_get_shader_param;
 470
 471         fd_resource_screen_init(pscreen);
 472         fd_query_screen_init(pscreen);
 473
 474         pscreen->get_name = fd_screen_get_name;
 475         pscreen->get_vendor = fd_screen_get_vendor;
 476
 477         pscreen->get_timestamp = fd_screen_get_timestamp;
 478
 479         pscreen->fence_reference = fd_screen_fence_ref;
 480         pscreen->fence_signalled = fd_screen_fence_signalled;
 481         pscreen->fence_finish = fd_screen_fence_finish;
 482
 483         util_format_s3tc_init();
 484
 485         return pscreen;
 486
 487 fail:
 488         fd_screen_destroy(pscreen);
 489         return NULL;
 490 }