src/gallium/drivers/r600/r600_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 "r600_pipe.h"
  24 #include "r600_public.h"
  25 #include "r600_isa.h"
  26 #include "evergreen_compute.h"
  27 #include "r600d.h"
  28
  29 #include "sb/sb_public.h"
  30
  31 #include <errno.h>
  32 #include "pipe/p_shader_tokens.h"
  33 #include "util/u_debug.h"
  34 #include "util/u_memory.h"
  35 #include "util/u_simple_shaders.h"
  36 #include "util/u_upload_mgr.h"
  37 #include "util/u_math.h"
  38 #include "vl/vl_decoder.h"
  39 #include "vl/vl_video_buffer.h"
  40 #include "radeon_video.h"
  41 #include "radeon_uvd.h"
  42 #include "util/os_time.h"
  43
  44 static const struct debug_named_value r600_debug_options[] = {
  45         /* features */
  46         { "nocpdma", DBG_NO_CP_DMA, "Disable CP DMA" },
  47
  48         /* shader backend */
  49         { "nosb", DBG_NO_SB, "Disable sb backend for graphics shaders" },
  50         { "sbcl", DBG_SB_CS, "Enable sb backend for compute shaders" },
  51         { "sbdry", DBG_SB_DRY_RUN, "Don't use optimized bytecode (just print the dumps)" },
  52         { "sbstat", DBG_SB_STAT, "Print optimization statistics for shaders" },
  53         { "sbdump", DBG_SB_DUMP, "Print IR dumps after some optimization passes" },
  54         { "sbnofallback", DBG_SB_NO_FALLBACK, "Abort on errors instead of fallback" },
  55         { "sbdisasm", DBG_SB_DISASM, "Use sb disassembler for shader dumps" },
  56         { "sbsafemath", DBG_SB_SAFEMATH, "Disable unsafe math optimizations" },
  57
  58         DEBUG_NAMED_VALUE_END /* must be last */
  59 };
  60
  61 /*
  62  * pipe_context
  63  */
  64
  65 static void r600_destroy_context(struct pipe_context *context)
  66 {
  67         struct r600_context *rctx = (struct r600_context *)context;
  68         unsigned sh;
  69
  70         r600_isa_destroy(rctx->isa);
  71
  72         r600_sb_context_destroy(rctx->sb_context);
  73
  74         for (sh = 0; sh < (rctx->b.chip_class < EVERGREEN ? R600_NUM_HW_STAGES : EG_NUM_HW_STAGES); sh++) {
  75                 r600_resource_reference(&rctx->scratch_buffers[sh].buffer, NULL);
  76         }
  77         r600_resource_reference(&rctx->dummy_cmask, NULL);
  78         r600_resource_reference(&rctx->dummy_fmask, NULL);
  79
  80         if (rctx->append_fence)
  81                 pipe_resource_reference((struct pipe_resource**)&rctx->append_fence, NULL);
  82         for (sh = 0; sh < PIPE_SHADER_TYPES; sh++) {
  83                 rctx->b.b.set_constant_buffer(&rctx->b.b, sh, R600_BUFFER_INFO_CONST_BUFFER, NULL);
  84                 free(rctx->driver_consts[sh].constants);
  85         }
  86
  87         if (rctx->fixed_func_tcs_shader)
  88                 rctx->b.b.delete_tcs_state(&rctx->b.b, rctx->fixed_func_tcs_shader);
  89
  90         if (rctx->dummy_pixel_shader) {
  91                 rctx->b.b.delete_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
  92         }
  93         if (rctx->custom_dsa_flush) {
  94                 rctx->b.b.delete_depth_stencil_alpha_state(&rctx->b.b, rctx->custom_dsa_flush);
  95         }
  96         if (rctx->custom_blend_resolve) {
  97                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_resolve);
  98         }
  99         if (rctx->custom_blend_decompress) {
 100                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_decompress);
 101         }
 102         if (rctx->custom_blend_fastclear) {
 103                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_fastclear);
 104         }
 105         util_unreference_framebuffer_state(&rctx->framebuffer.state);
 106
 107         if (rctx->blitter) {
 108                 util_blitter_destroy(rctx->blitter);
 109         }
 110         if (rctx->allocator_fetch_shader) {
 111                 u_suballocator_destroy(rctx->allocator_fetch_shader);
 112         }
 113
 114         r600_release_command_buffer(&rctx->start_cs_cmd);
 115
 116         FREE(rctx->start_compute_cs_cmd.buf);
 117
 118         r600_common_context_cleanup(&rctx->b);
 119
 120         r600_resource_reference(&rctx->trace_buf, NULL);
 121         r600_resource_reference(&rctx->last_trace_buf, NULL);
 122         radeon_clear_saved_cs(&rctx->last_gfx);
 123
 124         FREE(rctx);
 125 }
 126
 127 static struct pipe_context *r600_create_context(struct pipe_screen *screen,
 128                                                 void *priv, unsigned flags)
 129 {
 130         struct r600_context *rctx = CALLOC_STRUCT(r600_context);
 131         struct r600_screen* rscreen = (struct r600_screen *)screen;
 132         struct radeon_winsys *ws = rscreen->b.ws;
 133
 134         if (!rctx)
 135                 return NULL;
 136
 137         rctx->b.b.screen = screen;
 138         assert(!priv);
 139         rctx->b.b.priv = NULL; /* for threaded_context_unwrap_sync */
 140         rctx->b.b.destroy = r600_destroy_context;
 141         rctx->b.set_atom_dirty = (void *)r600_set_atom_dirty;
 142
 143         if (!r600_common_context_init(&rctx->b, &rscreen->b, flags))
 144                 goto fail;
 145
 146         rctx->screen = rscreen;
 147         LIST_INITHEAD(&rctx->texture_buffers);
 148
 149         r600_init_blit_functions(rctx);
 150
 151         if (rscreen->b.info.has_hw_decode) {
 152                 rctx->b.b.create_video_codec = r600_uvd_create_decoder;
 153                 rctx->b.b.create_video_buffer = r600_video_buffer_create;
 154         } else {
 155                 rctx->b.b.create_video_codec = vl_create_decoder;
 156                 rctx->b.b.create_video_buffer = vl_video_buffer_create;
 157         }
 158
 159         if (getenv("R600_TRACE"))
 160                 rctx->is_debug = true;
 161         r600_init_common_state_functions(rctx);
 162
 163         switch (rctx->b.chip_class) {
 164         case R600:
 165         case R700:
 166                 r600_init_state_functions(rctx);
 167                 r600_init_atom_start_cs(rctx);
 168                 rctx->custom_dsa_flush = r600_create_db_flush_dsa(rctx);
 169                 rctx->custom_blend_resolve = rctx->b.chip_class == R700 ? r700_create_resolve_blend(rctx)
 170                                                                       : r600_create_resolve_blend(rctx);
 171                 rctx->custom_blend_decompress = r600_create_decompress_blend(rctx);
 172                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_RV610 ||
 173                                            rctx->b.family == CHIP_RV620 ||
 174                                            rctx->b.family == CHIP_RS780 ||
 175                                            rctx->b.family == CHIP_RS880 ||
 176                                            rctx->b.family == CHIP_RV710);
 177                 break;
 178         case EVERGREEN:
 179         case CAYMAN:
 180                 evergreen_init_state_functions(rctx);
 181                 evergreen_init_atom_start_cs(rctx);
 182                 evergreen_init_atom_start_compute_cs(rctx);
 183                 rctx->custom_dsa_flush = evergreen_create_db_flush_dsa(rctx);
 184                 rctx->custom_blend_resolve = evergreen_create_resolve_blend(rctx);
 185                 rctx->custom_blend_decompress = evergreen_create_decompress_blend(rctx);
 186                 rctx->custom_blend_fastclear = evergreen_create_fastclear_blend(rctx);
 187                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_CEDAR ||
 188                                            rctx->b.family == CHIP_PALM ||
 189                                            rctx->b.family == CHIP_SUMO ||
 190                                            rctx->b.family == CHIP_SUMO2 ||
 191                                            rctx->b.family == CHIP_CAICOS ||
 192                                            rctx->b.family == CHIP_CAYMAN ||
 193                                            rctx->b.family == CHIP_ARUBA);
 194
 195                 rctx->append_fence = pipe_buffer_create(rctx->b.b.screen, PIPE_BIND_CUSTOM,
 196                                                          PIPE_USAGE_DEFAULT, 32);
 197                 break;
 198         default:
 199                 R600_ERR("Unsupported chip class %d.\n", rctx->b.chip_class);
 200                 goto fail;
 201         }
 202
 203         rctx->b.gfx.cs = ws->cs_create(rctx->b.ctx, RING_GFX,
 204                                        r600_context_gfx_flush, rctx);
 205         rctx->b.gfx.flush = r600_context_gfx_flush;
 206
 207         rctx->allocator_fetch_shader =
 208                 u_suballocator_create(&rctx->b.b, 64 * 1024,
 209                                       0, PIPE_USAGE_DEFAULT, 0, FALSE);
 210         if (!rctx->allocator_fetch_shader)
 211                 goto fail;
 212
 213         rctx->isa = calloc(1, sizeof(struct r600_isa));
 214         if (!rctx->isa || r600_isa_init(rctx, rctx->isa))
 215                 goto fail;
 216
 217         if (rscreen->b.debug_flags & DBG_FORCE_DMA)
 218                 rctx->b.b.resource_copy_region = rctx->b.dma_copy;
 219
 220         rctx->blitter = util_blitter_create(&rctx->b.b);
 221         if (rctx->blitter == NULL)
 222                 goto fail;
 223         util_blitter_set_texture_multisample(rctx->blitter, rscreen->has_msaa);
 224         rctx->blitter->draw_rectangle = r600_draw_rectangle;
 225
 226         r600_begin_new_cs(rctx);
 227
 228         rctx->dummy_pixel_shader =
 229                 util_make_fragment_cloneinput_shader(&rctx->b.b, 0,
 230                                                      TGSI_SEMANTIC_GENERIC,
 231                                                      TGSI_INTERPOLATE_CONSTANT);
 232         rctx->b.b.bind_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
 233
 234         return &rctx->b.b;
 235
 236 fail:
 237         r600_destroy_context(&rctx->b.b);
 238         return NULL;
 239 }
 240
 241 /*
 242  * pipe_screen
 243  */
 244
 245 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 246 {
 247         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 248         enum radeon_family family = rscreen->b.family;
 249
 250         switch (param) {
 251         /* Supported features (boolean caps). */
 252         case PIPE_CAP_NPOT_TEXTURES:
 253         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 254         case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
 255         case PIPE_CAP_ANISOTROPIC_FILTER:
 256         case PIPE_CAP_POINT_SPRITE:
 257         case PIPE_CAP_OCCLUSION_QUERY:
 258         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 259         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 260         case PIPE_CAP_TEXTURE_SWIZZLE:
 261         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 262         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 263         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 264         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 265         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 266         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 267         case PIPE_CAP_SM3:
 268         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 269         case PIPE_CAP_PRIMITIVE_RESTART:
 270         case PIPE_CAP_CONDITIONAL_RENDER:
 271         case PIPE_CAP_TEXTURE_BARRIER:
 272         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 273         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 274         case PIPE_CAP_TGSI_INSTANCEID:
 275         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 276         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 277         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 278         case PIPE_CAP_START_INSTANCE:
 279         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 280         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 281         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 282         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 283         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 284         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 285         case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 286         case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 287         case PIPE_CAP_SAMPLE_SHADING:
 288         case PIPE_CAP_CLIP_HALFZ:
 289         case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 290         case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 291         case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 292         case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 293         case PIPE_CAP_TGSI_TXQS:
 294         case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 295         case PIPE_CAP_INVALIDATE_BUFFER:
 296         case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 297         case PIPE_CAP_QUERY_MEMORY_INFO:
 298         case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 299         case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 300         case PIPE_CAP_CLEAR_TEXTURE:
 301         case PIPE_CAP_TGSI_MUL_ZERO_WINS:
 302         case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
 303         case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 304         case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 305                 return 1;
 306
 307         case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 308                 return rscreen->b.info.drm_major == 2 && rscreen->b.info.drm_minor >= 43;
 309
 310         case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 311                 return !R600_BIG_ENDIAN && rscreen->b.info.has_userptr;
 312
 313         case PIPE_CAP_COMPUTE:
 314                 return rscreen->b.chip_class > R700;
 315
 316         case PIPE_CAP_TGSI_TEXCOORD:
 317                 return 0;
 318
 319         case PIPE_CAP_FAKE_SW_MSAA:
 320                 return 0;
 321
 322         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 323                 return MIN2(rscreen->b.info.max_alloc_size, INT_MAX);
 324
 325         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 326                 return R600_MAP_BUFFER_ALIGNMENT;
 327
 328         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 329                 return 256;
 330
 331         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 332                 return 1;
 333
 334         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 335                 if (family >= CHIP_CEDAR)
 336                    return 430;
 337                 /* pre-evergreen geom shaders need newer kernel */
 338                 if (rscreen->b.info.drm_minor >= 37)
 339                    return 330;
 340                 return 140;
 341
 342         /* Supported except the original R600. */
 343         case PIPE_CAP_INDEP_BLEND_ENABLE:
 344         case PIPE_CAP_INDEP_BLEND_FUNC:
 345                 /* R600 doesn't support per-MRT blends */
 346                 return family == CHIP_R600 ? 0 : 1;
 347
 348         /* Supported on Evergreen. */
 349         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 350         case PIPE_CAP_CUBE_MAP_ARRAY:
 351         case PIPE_CAP_TEXTURE_GATHER_SM5:
 352         case PIPE_CAP_TEXTURE_QUERY_LOD:
 353         case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 354         case PIPE_CAP_SAMPLER_VIEW_TARGET:
 355         case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 356         case PIPE_CAP_TGSI_CLOCK:
 357         case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 358         case PIPE_CAP_QUERY_BUFFER_OBJECT:
 359                 return family >= CHIP_CEDAR ? 1 : 0;
 360         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 361                 return family >= CHIP_CEDAR ? 4 : 0;
 362         case PIPE_CAP_DRAW_INDIRECT:
 363                 /* kernel command checker support is also required */
 364                 return family >= CHIP_CEDAR && rscreen->b.info.drm_minor >= 41;
 365
 366         case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 367                 return family >= CHIP_CEDAR ? 0 : 1;
 368
 369         case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
 370                 return 8;
 371
 372         /* Unsupported features. */
 373         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 374         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 375         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 376         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 377         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 378         case PIPE_CAP_USER_VERTEX_BUFFERS:
 379         case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 380         case PIPE_CAP_VERTEXID_NOBASE:
 381         case PIPE_CAP_DEPTH_BOUNDS_TEST:
 382         case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 383         case PIPE_CAP_SHAREABLE_SHADERS:
 384         case PIPE_CAP_DRAW_PARAMETERS:
 385         case PIPE_CAP_MULTI_DRAW_INDIRECT:
 386         case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 387         case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 388         case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 389         case PIPE_CAP_GENERATE_MIPMAP:
 390         case PIPE_CAP_STRING_MARKER:
 391         case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 392         case PIPE_CAP_TGSI_VOTE:
 393         case PIPE_CAP_MAX_WINDOW_RECTANGLES:
 394         case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
 395         case PIPE_CAP_NATIVE_FENCE_FD:
 396         case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 397         case PIPE_CAP_TGSI_FS_FBFETCH:
 398         case PIPE_CAP_INT64:
 399         case PIPE_CAP_INT64_DIVMOD:
 400         case PIPE_CAP_TGSI_TEX_TXF_LZ:
 401         case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
 402         case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
 403         case PIPE_CAP_TGSI_BALLOT:
 404         case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
 405         case PIPE_CAP_POST_DEPTH_COVERAGE:
 406         case PIPE_CAP_BINDLESS_TEXTURE:
 407         case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
 408         case PIPE_CAP_QUERY_SO_OVERFLOW:
 409         case PIPE_CAP_MEMOBJ:
 410         case PIPE_CAP_LOAD_CONSTBUF:
 411         case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
 412         case PIPE_CAP_TILE_RASTER_ORDER:
 413         case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
 414         case PIPE_CAP_CONTEXT_PRIORITY_MASK:
 415         case PIPE_CAP_FENCE_SIGNAL:
 416                 return 0;
 417
 418         case PIPE_CAP_DOUBLES:
 419                 if (rscreen->b.family == CHIP_ARUBA ||
 420                     rscreen->b.family == CHIP_CAYMAN ||
 421                     rscreen->b.family == CHIP_CYPRESS ||
 422                     rscreen->b.family == CHIP_HEMLOCK)
 423                         return 1;
 424                 return 0;
 425         case PIPE_CAP_CULL_DISTANCE:
 426                 return 1;
 427
 428         case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 429                 if (family >= CHIP_CEDAR)
 430                         return 256;
 431                 return 0;
 432
 433         case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 434                 if (family >= CHIP_CEDAR)
 435                         return 30;
 436                 else
 437                         return 0;
 438         /* Stream output. */
 439         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 440                 return rscreen->b.has_streamout ? 4 : 0;
 441         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 442         case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 443                 return rscreen->b.has_streamout ? 1 : 0;
 444         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 445         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 446                 return 32*4;
 447
 448         /* Geometry shader output. */
 449         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 450                 return 1024;
 451         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 452                 return 16384;
 453         case PIPE_CAP_MAX_VERTEX_STREAMS:
 454                 return family >= CHIP_CEDAR ? 4 : 1;
 455
 456         case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 457                 return 2047;
 458
 459         /* Texturing. */
 460         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 461         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 462                 if (family >= CHIP_CEDAR)
 463                         return 15;
 464                 else
 465                         return 14;
 466         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 467                 /* textures support 8192, but layered rendering supports 2048 */
 468                 return 12;
 469         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 470                 /* textures support 8192, but layered rendering supports 2048 */
 471                 return 2048;
 472
 473         /* Render targets. */
 474         case PIPE_CAP_MAX_RENDER_TARGETS:
 475                 /* XXX some r6xx are buggy and can only do 4 */
 476                 return 8;
 477
 478         case PIPE_CAP_MAX_VIEWPORTS:
 479                 return R600_MAX_VIEWPORTS;
 480         case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 481                 return 8;
 482
 483         /* Timer queries, present when the clock frequency is non zero. */
 484         case PIPE_CAP_QUERY_TIME_ELAPSED:
 485                 return rscreen->b.info.clock_crystal_freq != 0;
 486         case PIPE_CAP_QUERY_TIMESTAMP:
 487                 return rscreen->b.info.drm_minor >= 20 &&
 488                        rscreen->b.info.clock_crystal_freq != 0;
 489
 490         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 491         case PIPE_CAP_MIN_TEXEL_OFFSET:
 492                 return -8;
 493
 494         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 495         case PIPE_CAP_MAX_TEXEL_OFFSET:
 496                 return 7;
 497
 498         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 499                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 500         case PIPE_CAP_ENDIANNESS:
 501                 return PIPE_ENDIAN_LITTLE;
 502
 503         case PIPE_CAP_VENDOR_ID:
 504                 return ATI_VENDOR_ID;
 505         case PIPE_CAP_DEVICE_ID:
 506                 return rscreen->b.info.pci_id;
 507         case PIPE_CAP_ACCELERATED:
 508                 return 1;
 509         case PIPE_CAP_VIDEO_MEMORY:
 510                 return rscreen->b.info.vram_size >> 20;
 511         case PIPE_CAP_UMA:
 512                 return 0;
 513         case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 514                 return rscreen->b.chip_class >= R700;
 515         case PIPE_CAP_PCI_GROUP:
 516                 return rscreen->b.info.pci_domain;
 517         case PIPE_CAP_PCI_BUS:
 518                 return rscreen->b.info.pci_bus;
 519         case PIPE_CAP_PCI_DEVICE:
 520                 return rscreen->b.info.pci_dev;
 521         case PIPE_CAP_PCI_FUNCTION:
 522                 return rscreen->b.info.pci_func;
 523         }
 524         return 0;
 525 }
 526
 527 static int r600_get_shader_param(struct pipe_screen* pscreen,
 528                                  enum pipe_shader_type shader,
 529                                  enum pipe_shader_cap param)
 530 {
 531         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 532
 533         switch(shader)
 534         {
 535         case PIPE_SHADER_FRAGMENT:
 536         case PIPE_SHADER_VERTEX:
 537         case PIPE_SHADER_COMPUTE:
 538                 break;
 539         case PIPE_SHADER_GEOMETRY:
 540                 if (rscreen->b.family >= CHIP_CEDAR)
 541                         break;
 542                 /* pre-evergreen geom shaders need newer kernel */
 543                 if (rscreen->b.info.drm_minor >= 37)
 544                         break;
 545                 return 0;
 546         case PIPE_SHADER_TESS_CTRL:
 547         case PIPE_SHADER_TESS_EVAL:
 548                 if (rscreen->b.family >= CHIP_CEDAR)
 549                         break;
 550         default:
 551                 return 0;
 552         }
 553
 554         switch (param) {
 555         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 556         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 557         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 558         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 559                 return 16384;
 560         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 561                 return 32;
 562         case PIPE_SHADER_CAP_MAX_INPUTS:
 563                 return shader == PIPE_SHADER_VERTEX ? 16 : 32;
 564         case PIPE_SHADER_CAP_MAX_OUTPUTS:
 565                 return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
 566         case PIPE_SHADER_CAP_MAX_TEMPS:
 567                 return 256; /* Max native temporaries. */
 568         case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 569                 if (shader == PIPE_SHADER_COMPUTE) {
 570                         uint64_t max_const_buffer_size;
 571                         pscreen->get_compute_param(pscreen, PIPE_SHADER_IR_TGSI,
 572                                 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
 573                                 &max_const_buffer_size);
 574                         return MIN2(max_const_buffer_size, INT_MAX);
 575
 576                 } else {
 577                         return R600_MAX_CONST_BUFFER_SIZE;
 578                 }
 579         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 580                 return R600_MAX_USER_CONST_BUFFERS;
 581         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 582                 return 1;
 583         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 584                 return 1;
 585         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 586         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 587         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 588         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 589                 return 1;
 590         case PIPE_SHADER_CAP_SUBROUTINES:
 591         case PIPE_SHADER_CAP_INT64_ATOMICS:
 592         case PIPE_SHADER_CAP_FP16:
 593                 return 0;
 594         case PIPE_SHADER_CAP_INTEGERS:
 595         case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 596                 return 1;
 597         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 598         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 599                 return 16;
 600         case PIPE_SHADER_CAP_PREFERRED_IR:
 601                 if (shader == PIPE_SHADER_COMPUTE) {
 602                         return PIPE_SHADER_IR_NATIVE;
 603                 } else {
 604                         return PIPE_SHADER_IR_TGSI;
 605                 }
 606         case PIPE_SHADER_CAP_SUPPORTED_IRS:
 607                 if (rscreen->b.family >= CHIP_CEDAR)
 608                         return (1 << PIPE_SHADER_IR_TGSI);
 609                 return 0;
 610         case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 611                 if (rscreen->b.family == CHIP_ARUBA ||
 612                     rscreen->b.family == CHIP_CAYMAN ||
 613                     rscreen->b.family == CHIP_CYPRESS ||
 614                     rscreen->b.family == CHIP_HEMLOCK)
 615                         return 1;
 616                 return 0;
 617         case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 618         case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 619         case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
 620         case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
 621         case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
 622                 return 0;
 623         case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 624         case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 625                 if (rscreen->b.family >= CHIP_CEDAR &&
 626                     (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE))
 627                     return 8;
 628                 return 0;
 629         case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
 630                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics)
 631                         return 8;
 632                 return 0;
 633         case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
 634                 /* having to allocate the atomics out amongst shaders stages is messy,
 635                    so give compute 8 buffers and all the others one */
 636                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics) {
 637                         return EG_MAX_ATOMIC_BUFFERS;
 638                 }
 639                 return 0;
 640         case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 641                 /* due to a bug in the shader compiler, some loops hang
 642                  * if they are not unrolled, see:
 643                  *    https://bugs.freedesktop.org/show_bug.cgi?id=86720
 644                  */
 645                 return 255;
 646         }
 647         return 0;
 648 }
 649
 650 static void r600_destroy_screen(struct pipe_screen* pscreen)
 651 {
 652         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 653
 654         if (!rscreen)
 655                 return;
 656
 657         if (!rscreen->b.ws->unref(rscreen->b.ws))
 658                 return;
 659
 660         if (rscreen->global_pool) {
 661                 compute_memory_pool_delete(rscreen->global_pool);
 662         }
 663
 664         r600_destroy_common_screen(&rscreen->b);
 665 }
 666
 667 static struct pipe_resource *r600_resource_create(struct pipe_screen *screen,
 668                                                   const struct pipe_resource *templ)
 669 {
 670         if (templ->target == PIPE_BUFFER &&
 671             (templ->bind & PIPE_BIND_GLOBAL))
 672                 return r600_compute_global_buffer_create(screen, templ);
 673
 674         return r600_resource_create_common(screen, templ);
 675 }
 676
 677 struct pipe_screen *r600_screen_create(struct radeon_winsys *ws,
 678                                        const struct pipe_screen_config *config)
 679 {
 680         struct r600_screen *rscreen = CALLOC_STRUCT(r600_screen);
 681
 682         if (!rscreen) {
 683                 return NULL;
 684         }
 685
 686         /* Set functions first. */
 687         rscreen->b.b.context_create = r600_create_context;
 688         rscreen->b.b.destroy = r600_destroy_screen;
 689         rscreen->b.b.get_param = r600_get_param;
 690         rscreen->b.b.get_shader_param = r600_get_shader_param;
 691         rscreen->b.b.resource_create = r600_resource_create;
 692
 693         if (!r600_common_screen_init(&rscreen->b, ws)) {
 694                 FREE(rscreen);
 695                 return NULL;
 696         }
 697
 698         if (rscreen->b.info.chip_class >= EVERGREEN) {
 699                 rscreen->b.b.is_format_supported = evergreen_is_format_supported;
 700         } else {
 701                 rscreen->b.b.is_format_supported = r600_is_format_supported;
 702         }
 703
 704         rscreen->b.debug_flags |= debug_get_flags_option("R600_DEBUG", r600_debug_options, 0);
 705         if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE))
 706                 rscreen->b.debug_flags |= DBG_COMPUTE;
 707         if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE))
 708                 rscreen->b.debug_flags |= DBG_ALL_SHADERS | DBG_FS;
 709         if (!debug_get_bool_option("R600_HYPERZ", TRUE))
 710                 rscreen->b.debug_flags |= DBG_NO_HYPERZ;
 711
 712         if (rscreen->b.family == CHIP_UNKNOWN) {
 713                 fprintf(stderr, "r600: Unknown chipset 0x%04X\n", rscreen->b.info.pci_id);
 714                 FREE(rscreen);
 715                 return NULL;
 716         }
 717
 718         /* Figure out streamout kernel support. */
 719         switch (rscreen->b.chip_class) {
 720         case R600:
 721                 if (rscreen->b.family < CHIP_RS780) {
 722                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 723                 } else {
 724                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 23;
 725                 }
 726                 break;
 727         case R700:
 728                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 17;
 729                 break;
 730         case EVERGREEN:
 731         case CAYMAN:
 732                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 733                 break;
 734         default:
 735                 rscreen->b.has_streamout = FALSE;
 736                 break;
 737         }
 738
 739         /* MSAA support. */
 740         switch (rscreen->b.chip_class) {
 741         case R600:
 742         case R700:
 743                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 22;
 744                 rscreen->has_compressed_msaa_texturing = false;
 745                 break;
 746         case EVERGREEN:
 747                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 748                 rscreen->has_compressed_msaa_texturing = rscreen->b.info.drm_minor >= 24;
 749                 break;
 750         case CAYMAN:
 751                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 752                 rscreen->has_compressed_msaa_texturing = true;
 753                 break;
 754         default:
 755                 rscreen->has_msaa = FALSE;
 756                 rscreen->has_compressed_msaa_texturing = false;
 757         }
 758
 759         rscreen->b.has_cp_dma = rscreen->b.info.drm_minor >= 27 &&
 760                               !(rscreen->b.debug_flags & DBG_NO_CP_DMA);
 761
 762         rscreen->b.barrier_flags.cp_to_L2 =
 763                 R600_CONTEXT_INV_VERTEX_CACHE |
 764                 R600_CONTEXT_INV_TEX_CACHE |
 765                 R600_CONTEXT_INV_CONST_CACHE;
 766         rscreen->b.barrier_flags.compute_to_L2 = R600_CONTEXT_CS_PARTIAL_FLUSH | R600_CONTEXT_FLUSH_AND_INV;
 767
 768         rscreen->global_pool = compute_memory_pool_new(rscreen);
 769
 770         /* Create the auxiliary context. This must be done last. */
 771         rscreen->b.aux_context = rscreen->b.b.context_create(&rscreen->b.b, NULL, 0);
 772
 773         rscreen->has_atomics = rscreen->b.info.drm_minor >= 44;
 774 #if 0 /* This is for testing whether aux_context and buffer clearing work correctly. */
 775         struct pipe_resource templ = {};
 776
 777         templ.width0 = 4;
 778         templ.height0 = 2048;
 779         templ.depth0 = 1;
 780         templ.array_size = 1;
 781         templ.target = PIPE_TEXTURE_2D;
 782         templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 783         templ.usage = PIPE_USAGE_DEFAULT;
 784
 785         struct r600_resource *res = r600_resource(rscreen->screen.resource_create(&rscreen->screen, &templ));
 786         unsigned char *map = ws->buffer_map(res->buf, NULL, PIPE_TRANSFER_WRITE);
 787
 788         memset(map, 0, 256);
 789
 790         r600_screen_clear_buffer(rscreen, &res->b.b, 4, 4, 0xCC);
 791         r600_screen_clear_buffer(rscreen, &res->b.b, 8, 4, 0xDD);
 792         r600_screen_clear_buffer(rscreen, &res->b.b, 12, 4, 0xEE);
 793         r600_screen_clear_buffer(rscreen, &res->b.b, 20, 4, 0xFF);
 794         r600_screen_clear_buffer(rscreen, &res->b.b, 32, 20, 0x87);
 795
 796         ws->buffer_wait(res->buf, RADEON_USAGE_WRITE);
 797
 798         int i;
 799         for (i = 0; i < 256; i++) {
 800                 printf("%02X", map[i]);
 801                 if (i % 16 == 15)
 802                         printf("\n");
 803         }
 804 #endif
 805
 806         if (rscreen->b.debug_flags & DBG_TEST_DMA)
 807                 r600_test_dma(&rscreen->b);
 808
 809         r600_query_fix_enabled_rb_mask(&rscreen->b);
 810         return &rscreen->b.b;
 811 }