src/gallium/drivers/radeonsi/si_pipe.c

   1 /*
   2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
   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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  */
  23 #include <stdio.h>
  24 #include <errno.h>
  25 #include "pipe/p_defines.h"
  26 #include "pipe/p_state.h"
  27 #include "pipe/p_context.h"
  28 #include "tgsi/tgsi_scan.h"
  29 #include "tgsi/tgsi_parse.h"
  30 #include "tgsi/tgsi_util.h"
  31 #include "util/u_blitter.h"
  32 #include "util/u_double_list.h"
  33 #include "util/u_format.h"
  34 #include "util/u_transfer.h"
  35 #include "util/u_surface.h"
  36 #include "util/u_pack_color.h"
  37 #include "util/u_memory.h"
  38 #include "util/u_inlines.h"
  39 #include "util/u_simple_shaders.h"
  40 #include "util/u_upload_mgr.h"
  41 #include "vl/vl_decoder.h"
  42 #include "vl/vl_video_buffer.h"
  43 #include "os/os_time.h"
  44 #include "pipebuffer/pb_buffer.h"
  45 #include "si_pipe.h"
  46 #include "radeon/radeon_uvd.h"
  47 #include "si.h"
  48 #include "sid.h"
  49 #include "si_resource.h"
  50 #include "si_pipe.h"
  51 #include "si_state.h"
  52 #include "../radeon/r600_cs.h"
  53
  54 /*
  55  * pipe_context
  56  */
  57 void si_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
  58               unsigned flags)
  59 {
  60         struct si_context *sctx = (struct si_context *)ctx;
  61         struct pipe_query *render_cond = NULL;
  62         boolean render_cond_cond = FALSE;
  63         unsigned render_cond_mode = 0;
  64
  65         if (fence) {
  66                 *fence = sctx->b.ws->cs_create_fence(sctx->b.rings.gfx.cs);
  67         }
  68
  69         /* Disable render condition. */
  70         if (sctx->b.current_render_cond) {
  71                 render_cond = sctx->b.current_render_cond;
  72                 render_cond_cond = sctx->b.current_render_cond_cond;
  73                 render_cond_mode = sctx->b.current_render_cond_mode;
  74                 ctx->render_condition(ctx, NULL, FALSE, 0);
  75         }
  76
  77         si_context_flush(sctx, flags);
  78
  79         /* Re-enable render condition. */
  80         if (render_cond) {
  81                 ctx->render_condition(ctx, render_cond, render_cond_cond, render_cond_mode);
  82         }
  83 }
  84
  85 static void si_flush_from_st(struct pipe_context *ctx,
  86                              struct pipe_fence_handle **fence,
  87                              unsigned flags)
  88 {
  89         si_flush(ctx, fence,
  90                  flags & PIPE_FLUSH_END_OF_FRAME ? RADEON_FLUSH_END_OF_FRAME : 0);
  91 }
  92
  93 static void si_flush_from_winsys(void *ctx, unsigned flags)
  94 {
  95         si_flush((struct pipe_context*)ctx, NULL, flags);
  96 }
  97
  98 static void si_destroy_context(struct pipe_context *context)
  99 {
 100         struct si_context *sctx = (struct si_context *)context;
 101
 102         si_release_all_descriptors(sctx);
 103
 104         pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
 105         r600_resource_reference(&sctx->border_color_table, NULL);
 106
 107         if (sctx->dummy_pixel_shader) {
 108                 sctx->b.b.delete_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
 109         }
 110         for (int i = 0; i < 8; i++) {
 111                 sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth_stencil[i]);
 112                 sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth[i]);
 113                 sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_stencil[i]);
 114         }
 115         sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_inplace);
 116         sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
 117         sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
 118         util_unreference_framebuffer_state(&sctx->framebuffer);
 119
 120         util_blitter_destroy(sctx->blitter);
 121
 122         r600_common_context_cleanup(&sctx->b);
 123         FREE(sctx);
 124 }
 125
 126 static struct pipe_context *si_create_context(struct pipe_screen *screen, void *priv)
 127 {
 128         struct si_context *sctx = CALLOC_STRUCT(si_context);
 129         struct si_screen* sscreen = (struct si_screen *)screen;
 130         int shader, i;
 131
 132         if (sctx == NULL)
 133                 return NULL;
 134
 135         sctx->b.b.screen = screen; /* this must be set first */
 136         sctx->b.b.priv = priv;
 137         sctx->b.b.destroy = si_destroy_context;
 138         sctx->b.b.flush = si_flush_from_st;
 139         sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
 140
 141         if (!r600_common_context_init(&sctx->b, &sscreen->b))
 142                 goto fail;
 143
 144         si_init_blit_functions(sctx);
 145         si_init_compute_functions(sctx);
 146
 147         if (sscreen->b.info.has_uvd) {
 148                 sctx->b.b.create_video_codec = si_uvd_create_decoder;
 149                 sctx->b.b.create_video_buffer = si_video_buffer_create;
 150         } else {
 151                 sctx->b.b.create_video_codec = vl_create_decoder;
 152                 sctx->b.b.create_video_buffer = vl_video_buffer_create;
 153         }
 154
 155         sctx->b.rings.gfx.cs = sctx->b.ws->cs_create(sctx->b.ws, RING_GFX, NULL);
 156         sctx->b.rings.gfx.flush = si_flush_from_winsys;
 157
 158         si_init_all_descriptors(sctx);
 159
 160         /* Initialize cache_flush. */
 161         sctx->cache_flush = si_atom_cache_flush;
 162         sctx->atoms.cache_flush = &sctx->cache_flush;
 163
 164         sctx->atoms.streamout_begin = &sctx->b.streamout.begin_atom;
 165
 166         switch (sctx->b.chip_class) {
 167         case SI:
 168         case CIK:
 169                 si_init_state_functions(sctx);
 170                 si_init_config(sctx);
 171                 break;
 172         default:
 173                 R600_ERR("Unsupported chip class %d.\n", sctx->b.chip_class);
 174                 goto fail;
 175         }
 176
 177         sctx->b.ws->cs_set_flush_callback(sctx->b.rings.gfx.cs, si_flush_from_winsys, sctx);
 178
 179         sctx->blitter = util_blitter_create(&sctx->b.b);
 180         if (sctx->blitter == NULL)
 181                 goto fail;
 182
 183         sctx->dummy_pixel_shader =
 184                 util_make_fragment_cloneinput_shader(&sctx->b.b, 0,
 185                                                      TGSI_SEMANTIC_GENERIC,
 186                                                      TGSI_INTERPOLATE_CONSTANT);
 187         sctx->b.b.bind_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
 188
 189         /* these must be last */
 190         si_begin_new_cs(sctx);
 191         r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
 192
 193         /* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
 194          * with a NULL buffer). We need to use a dummy buffer instead. */
 195         if (sctx->b.chip_class == CIK) {
 196                 sctx->null_const_buf.buffer = pipe_buffer_create(screen, PIPE_BIND_CONSTANT_BUFFER,
 197                                                                  PIPE_USAGE_STATIC, 16);
 198                 sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
 199
 200                 for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
 201                         for (i = 0; i < NUM_CONST_BUFFERS; i++) {
 202                                 sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
 203                                                               &sctx->null_const_buf);
 204                         }
 205                 }
 206
 207                 /* Clear the NULL constant buffer, because loads should return zeros. */
 208                 sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
 209                                      sctx->null_const_buf.buffer->width0, 0);
 210         }
 211
 212         return &sctx->b.b;
 213 fail:
 214         si_destroy_context(&sctx->b.b);
 215         return NULL;
 216 }
 217
 218 /*
 219  * pipe_screen
 220  */
 221
 222 const char *si_get_llvm_processor_name(enum radeon_family family)
 223 {
 224         switch (family) {
 225                 case CHIP_TAHITI: return "tahiti";
 226                 case CHIP_PITCAIRN: return "pitcairn";
 227                 case CHIP_VERDE: return "verde";
 228                 case CHIP_OLAND: return "oland";
 229 #if HAVE_LLVM <= 0x0303
 230                 default: return "SI";
 231 #else
 232                 case CHIP_HAINAN: return "hainan";
 233                 case CHIP_BONAIRE: return "bonaire";
 234                 case CHIP_KABINI: return "kabini";
 235                 case CHIP_KAVERI: return "kaveri";
 236                 case CHIP_HAWAII: return "hawaii";
 237                 default: return "";
 238 #endif
 239         }
 240 }
 241
 242 static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 243 {
 244         struct si_screen *sscreen = (struct si_screen *)pscreen;
 245
 246         switch (param) {
 247         /* Supported features (boolean caps). */
 248         case PIPE_CAP_TWO_SIDED_STENCIL:
 249         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 250         case PIPE_CAP_ANISOTROPIC_FILTER:
 251         case PIPE_CAP_POINT_SPRITE:
 252         case PIPE_CAP_OCCLUSION_QUERY:
 253         case PIPE_CAP_TEXTURE_SHADOW_MAP:
 254         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 255         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 256         case PIPE_CAP_TEXTURE_SWIZZLE:
 257         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 258         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 259         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 260         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 261         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 262         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 263         case PIPE_CAP_SM3:
 264         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 265         case PIPE_CAP_PRIMITIVE_RESTART:
 266         case PIPE_CAP_CONDITIONAL_RENDER:
 267         case PIPE_CAP_TEXTURE_BARRIER:
 268         case PIPE_CAP_INDEP_BLEND_ENABLE:
 269         case PIPE_CAP_INDEP_BLEND_FUNC:
 270         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 271         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 272         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 273         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 274         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 275         case PIPE_CAP_USER_INDEX_BUFFERS:
 276         case PIPE_CAP_USER_CONSTANT_BUFFERS:
 277         case PIPE_CAP_START_INSTANCE:
 278         case PIPE_CAP_NPOT_TEXTURES:
 279         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 280         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 281         case PIPE_CAP_TGSI_INSTANCEID:
 282         case PIPE_CAP_COMPUTE:
 283         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 284         case PIPE_CAP_TGSI_VS_LAYER:
 285         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 286                 return 1;
 287
 288         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 289                 /* 2D tiling on CIK is supported since DRM 2.35.0 */
 290                 return HAVE_LLVM >= 0x0304 && (sscreen->b.chip_class < CIK ||
 291                                                sscreen->b.info.drm_minor >= 35);
 292
 293         case PIPE_CAP_TGSI_TEXCOORD:
 294                 return 0;
 295
 296         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 297                 return 64;
 298
 299         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 300                 return 256;
 301
 302         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 303                 return 140;
 304
 305         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 306                 return 1;
 307         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 308                 return MIN2(sscreen->b.info.vram_size, 0xFFFFFFFF);
 309
 310         /* Unsupported features. */
 311         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 312         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 313         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 314         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 315         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 316         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 317         case PIPE_CAP_USER_VERTEX_BUFFERS:
 318         case PIPE_CAP_CUBE_MAP_ARRAY:
 319                 return 0;
 320
 321         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 322                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 323
 324         /* Stream output. */
 325         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 326                 return sscreen->b.has_streamout ? 4 : 0;
 327         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 328                 return sscreen->b.has_streamout ? 1 : 0;
 329         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 330         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 331                 return sscreen->b.has_streamout ? 32*4 : 0;
 332
 333         /* Texturing. */
 334         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 335         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 336         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 337                         return 15;
 338         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 339                 return 16384;
 340         case PIPE_CAP_MAX_COMBINED_SAMPLERS:
 341                 return 32;
 342
 343         /* Render targets. */
 344         case PIPE_CAP_MAX_RENDER_TARGETS:
 345                 return 8;
 346
 347         case PIPE_CAP_MAX_VIEWPORTS:
 348                 return 1;
 349
 350         /* Timer queries, present when the clock frequency is non zero. */
 351         case PIPE_CAP_QUERY_TIMESTAMP:
 352         case PIPE_CAP_QUERY_TIME_ELAPSED:
 353                 return sscreen->b.info.r600_clock_crystal_freq != 0;
 354
 355         case PIPE_CAP_MIN_TEXEL_OFFSET:
 356                 return -8;
 357
 358         case PIPE_CAP_MAX_TEXEL_OFFSET:
 359                 return 7;
 360         case PIPE_CAP_ENDIANNESS:
 361                 return PIPE_ENDIAN_LITTLE;
 362         }
 363         return 0;
 364 }
 365
 366 static float si_get_paramf(struct pipe_screen* pscreen,
 367                              enum pipe_capf param)
 368 {
 369         switch (param) {
 370         case PIPE_CAPF_MAX_LINE_WIDTH:
 371         case PIPE_CAPF_MAX_LINE_WIDTH_AA:
 372         case PIPE_CAPF_MAX_POINT_WIDTH:
 373         case PIPE_CAPF_MAX_POINT_WIDTH_AA:
 374                 return 16384.0f;
 375         case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
 376                 return 16.0f;
 377         case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
 378                 return 16.0f;
 379         case PIPE_CAPF_GUARD_BAND_LEFT:
 380         case PIPE_CAPF_GUARD_BAND_TOP:
 381         case PIPE_CAPF_GUARD_BAND_RIGHT:
 382         case PIPE_CAPF_GUARD_BAND_BOTTOM:
 383                 return 0.0f;
 384         }
 385         return 0.0f;
 386 }
 387
 388 static int si_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
 389 {
 390         switch(shader)
 391         {
 392         case PIPE_SHADER_FRAGMENT:
 393         case PIPE_SHADER_VERTEX:
 394                 break;
 395         case PIPE_SHADER_GEOMETRY:
 396                 /* TODO: support and enable geometry programs */
 397                 return 0;
 398         case PIPE_SHADER_COMPUTE:
 399                 switch (param) {
 400                 case PIPE_SHADER_CAP_PREFERRED_IR:
 401                         return PIPE_SHADER_IR_LLVM;
 402                 default:
 403                         return 0;
 404                 }
 405         default:
 406                 /* TODO: support tessellation */
 407                 return 0;
 408         }
 409
 410         switch (param) {
 411         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 412         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 413         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 414         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 415                 return 16384;
 416         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 417                 return 32;
 418         case PIPE_SHADER_CAP_MAX_INPUTS:
 419                 return 32;
 420         case PIPE_SHADER_CAP_MAX_TEMPS:
 421                 return 256; /* Max native temporaries. */
 422         case PIPE_SHADER_CAP_MAX_ADDRS:
 423                 /* FIXME Isn't this equal to TEMPS? */
 424                 return 1; /* Max native address registers */
 425         case PIPE_SHADER_CAP_MAX_CONSTS:
 426                 return 4096; /* actually only memory limits this */
 427         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 428                 return NUM_PIPE_CONST_BUFFERS;
 429         case PIPE_SHADER_CAP_MAX_PREDS:
 430                 return 0; /* FIXME */
 431         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 432                 return 1;
 433         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 434                 return 0;
 435         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 436         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 437         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 438         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 439                 return 1;
 440         case PIPE_SHADER_CAP_INTEGERS:
 441                 return 1;
 442         case PIPE_SHADER_CAP_SUBROUTINES:
 443                 return 0;
 444         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 445         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 446                 return 16;
 447         case PIPE_SHADER_CAP_PREFERRED_IR:
 448                 return PIPE_SHADER_IR_TGSI;
 449         }
 450         return 0;
 451 }
 452
 453 static int si_get_video_param(struct pipe_screen *screen,
 454                               enum pipe_video_profile profile,
 455                               enum pipe_video_entrypoint entrypoint,
 456                               enum pipe_video_cap param)
 457 {
 458         switch (param) {
 459         case PIPE_VIDEO_CAP_SUPPORTED:
 460                 return vl_profile_supported(screen, profile, entrypoint);
 461         case PIPE_VIDEO_CAP_NPOT_TEXTURES:
 462                 return 1;
 463         case PIPE_VIDEO_CAP_MAX_WIDTH:
 464         case PIPE_VIDEO_CAP_MAX_HEIGHT:
 465                 return vl_video_buffer_max_size(screen);
 466         case PIPE_VIDEO_CAP_PREFERED_FORMAT:
 467                 return PIPE_FORMAT_NV12;
 468         case PIPE_VIDEO_CAP_MAX_LEVEL:
 469                 return vl_level_supported(screen, profile);
 470         default:
 471                 return 0;
 472         }
 473 }
 474
 475 static int si_get_compute_param(struct pipe_screen *screen,
 476         enum pipe_compute_cap param,
 477         void *ret)
 478 {
 479         struct si_screen *sscreen = (struct si_screen *)screen;
 480         //TODO: select these params by asic
 481         switch (param) {
 482         case PIPE_COMPUTE_CAP_IR_TARGET: {
 483                 const char *gpu = si_get_llvm_processor_name(sscreen->b.family);
 484                 if (ret) {
 485                         sprintf(ret, "%s-r600--", gpu);
 486                 }
 487                 return (8 + strlen(gpu)) * sizeof(char);
 488         }
 489         case PIPE_COMPUTE_CAP_GRID_DIMENSION:
 490                 if (ret) {
 491                         uint64_t * grid_dimension = ret;
 492                         grid_dimension[0] = 3;
 493                 }
 494                 return 1 * sizeof(uint64_t);
 495         case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
 496                 if (ret) {
 497                         uint64_t * grid_size = ret;
 498                         grid_size[0] = 65535;
 499                         grid_size[1] = 65535;
 500                         grid_size[2] = 1;
 501                 }
 502                 return 3 * sizeof(uint64_t) ;
 503
 504         case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
 505                 if (ret) {
 506                         uint64_t * block_size = ret;
 507                         block_size[0] = 256;
 508                         block_size[1] = 256;
 509                         block_size[2] = 256;
 510                 }
 511                 return 3 * sizeof(uint64_t);
 512         case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
 513                 if (ret) {
 514                         uint64_t * max_threads_per_block = ret;
 515                         *max_threads_per_block = 256;
 516                 }
 517                 return sizeof(uint64_t);
 518
 519         case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
 520                 if (ret) {
 521                         uint64_t *max_global_size = ret;
 522                         /* XXX: Not sure what to put here. */
 523                         *max_global_size = 2000000000;
 524                 }
 525                 return sizeof(uint64_t);
 526         case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
 527                 if (ret) {
 528                         uint64_t *max_local_size = ret;
 529                         /* Value reported by the closed source driver. */
 530                         *max_local_size = 32768;
 531                 }
 532                 return sizeof(uint64_t);
 533         case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
 534                 if (ret) {
 535                         uint64_t *max_input_size = ret;
 536                         /* Value reported by the closed source driver. */
 537                         *max_input_size = 1024;
 538                 }
 539                 return sizeof(uint64_t);
 540         case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
 541                 if (ret) {
 542                         uint64_t max_global_size;
 543                         uint64_t *max_mem_alloc_size = ret;
 544                         si_get_compute_param(screen, PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE, &max_global_size);
 545                         *max_mem_alloc_size = max_global_size / 4;
 546                 }
 547                 return sizeof(uint64_t);
 548         default:
 549                 fprintf(stderr, "unknown PIPE_COMPUTE_CAP %d\n", param);
 550                 return 0;
 551         }
 552 }
 553
 554 static void si_destroy_screen(struct pipe_screen* pscreen)
 555 {
 556         struct si_screen *sscreen = (struct si_screen *)pscreen;
 557
 558         if (sscreen == NULL)
 559                 return;
 560
 561         if (!radeon_winsys_unref(sscreen->b.ws))
 562                 return;
 563
 564         r600_common_screen_cleanup(&sscreen->b);
 565
 566 #if SI_TRACE_CS
 567         if (sscreen->trace_bo) {
 568                 sscreen->ws->buffer_unmap(sscreen->trace_bo->cs_buf);
 569                 pipe_resource_reference((struct pipe_resource**)&sscreen->trace_bo, NULL);
 570         }
 571 #endif
 572
 573         sscreen->b.ws->destroy(sscreen->b.ws);
 574         FREE(sscreen);
 575 }
 576
 577 struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
 578 {
 579         struct si_screen *sscreen = CALLOC_STRUCT(si_screen);
 580         if (sscreen == NULL) {
 581                 return NULL;
 582         }
 583
 584         ws->query_info(ws, &sscreen->b.info);
 585
 586         /* Set functions first. */
 587         sscreen->b.b.context_create = si_create_context;
 588         sscreen->b.b.destroy = si_destroy_screen;
 589         sscreen->b.b.get_param = si_get_param;
 590         sscreen->b.b.get_shader_param = si_get_shader_param;
 591         sscreen->b.b.get_paramf = si_get_paramf;
 592         sscreen->b.b.get_compute_param = si_get_compute_param;
 593         sscreen->b.b.is_format_supported = si_is_format_supported;
 594         if (sscreen->b.info.has_uvd) {
 595                 sscreen->b.b.get_video_param = ruvd_get_video_param;
 596                 sscreen->b.b.is_video_format_supported = ruvd_is_format_supported;
 597         } else {
 598                 sscreen->b.b.get_video_param = si_get_video_param;
 599                 sscreen->b.b.is_video_format_supported = vl_video_buffer_is_format_supported;
 600         }
 601
 602         if (!r600_common_screen_init(&sscreen->b, ws)) {
 603                 FREE(sscreen);
 604                 return NULL;
 605         }
 606
 607         sscreen->b.has_cp_dma = true;
 608         sscreen->b.has_streamout = HAVE_LLVM >= 0x0304;
 609
 610         if (debug_get_bool_option("RADEON_DUMP_SHADERS", FALSE))
 611                 sscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
 612
 613 #if SI_TRACE_CS
 614         sscreen->cs_count = 0;
 615         if (sscreen->info.drm_minor >= 28) {
 616                 sscreen->trace_bo = (struct r600_resource*)pipe_buffer_create(&sscreen->screen,
 617                                                                                 PIPE_BIND_CUSTOM,
 618                                                                                 PIPE_USAGE_STAGING,
 619                                                                                 4096);
 620                 if (sscreen->trace_bo) {
 621                         sscreen->trace_ptr = sscreen->ws->buffer_map(sscreen->trace_bo->cs_buf, NULL,
 622                                                                         PIPE_TRANSFER_UNSYNCHRONIZED);
 623                 }
 624         }
 625 #endif
 626
 627         /* Create the auxiliary context. This must be done last. */
 628         sscreen->b.aux_context = sscreen->b.b.context_create(&sscreen->b.b, NULL);
 629
 630         return &sscreen->b.b;
 631 }