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/radeon_video.h"
  41 #include "radeon/radeon_uvd.h"
  42 #include "os/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         r600_resource_reference(&rctx->dummy_cmask, NULL);
  75         r600_resource_reference(&rctx->dummy_fmask, NULL);
  76
  77         for (sh = 0; sh < PIPE_SHADER_TYPES; sh++) {
  78                 rctx->b.b.set_constant_buffer(&rctx->b.b, sh, R600_BUFFER_INFO_CONST_BUFFER, NULL);
  79                 free(rctx->driver_consts[sh].constants);
  80         }
  81
  82         if (rctx->fixed_func_tcs_shader)
  83                 rctx->b.b.delete_tcs_state(&rctx->b.b, rctx->fixed_func_tcs_shader);
  84
  85         if (rctx->dummy_pixel_shader) {
  86                 rctx->b.b.delete_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
  87         }
  88         if (rctx->custom_dsa_flush) {
  89                 rctx->b.b.delete_depth_stencil_alpha_state(&rctx->b.b, rctx->custom_dsa_flush);
  90         }
  91         if (rctx->custom_blend_resolve) {
  92                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_resolve);
  93         }
  94         if (rctx->custom_blend_decompress) {
  95                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_decompress);
  96         }
  97         if (rctx->custom_blend_fastclear) {
  98                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_fastclear);
  99         }
 100         util_unreference_framebuffer_state(&rctx->framebuffer.state);
 101
 102         if (rctx->blitter) {
 103                 util_blitter_destroy(rctx->blitter);
 104         }
 105         if (rctx->allocator_fetch_shader) {
 106                 u_suballocator_destroy(rctx->allocator_fetch_shader);
 107         }
 108
 109         r600_release_command_buffer(&rctx->start_cs_cmd);
 110
 111         FREE(rctx->start_compute_cs_cmd.buf);
 112
 113         r600_common_context_cleanup(&rctx->b);
 114
 115         r600_resource_reference(&rctx->trace_buf, NULL);
 116         r600_resource_reference(&rctx->last_trace_buf, NULL);
 117         radeon_clear_saved_cs(&rctx->last_gfx);
 118
 119         FREE(rctx);
 120 }
 121
 122 static struct pipe_context *r600_create_context(struct pipe_screen *screen,
 123                                                 void *priv, unsigned flags)
 124 {
 125         struct r600_context *rctx = CALLOC_STRUCT(r600_context);
 126         struct r600_screen* rscreen = (struct r600_screen *)screen;
 127         struct radeon_winsys *ws = rscreen->b.ws;
 128
 129         if (!rctx)
 130                 return NULL;
 131
 132         rctx->b.b.screen = screen;
 133         rctx->b.b.priv = priv;
 134         rctx->b.b.destroy = r600_destroy_context;
 135         rctx->b.set_atom_dirty = (void *)r600_set_atom_dirty;
 136
 137         if (!r600_common_context_init(&rctx->b, &rscreen->b, flags))
 138                 goto fail;
 139
 140         rctx->screen = rscreen;
 141         LIST_INITHEAD(&rctx->texture_buffers);
 142
 143         r600_init_blit_functions(rctx);
 144
 145         if (rscreen->b.info.has_hw_decode) {
 146                 rctx->b.b.create_video_codec = r600_uvd_create_decoder;
 147                 rctx->b.b.create_video_buffer = r600_video_buffer_create;
 148         } else {
 149                 rctx->b.b.create_video_codec = vl_create_decoder;
 150                 rctx->b.b.create_video_buffer = vl_video_buffer_create;
 151         }
 152
 153         if (getenv("R600_TRACE"))
 154                 rctx->is_debug = true;
 155         r600_init_common_state_functions(rctx);
 156
 157         switch (rctx->b.chip_class) {
 158         case R600:
 159         case R700:
 160                 r600_init_state_functions(rctx);
 161                 r600_init_atom_start_cs(rctx);
 162                 rctx->custom_dsa_flush = r600_create_db_flush_dsa(rctx);
 163                 rctx->custom_blend_resolve = rctx->b.chip_class == R700 ? r700_create_resolve_blend(rctx)
 164                                                                       : r600_create_resolve_blend(rctx);
 165                 rctx->custom_blend_decompress = r600_create_decompress_blend(rctx);
 166                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_RV610 ||
 167                                            rctx->b.family == CHIP_RV620 ||
 168                                            rctx->b.family == CHIP_RS780 ||
 169                                            rctx->b.family == CHIP_RS880 ||
 170                                            rctx->b.family == CHIP_RV710);
 171                 break;
 172         case EVERGREEN:
 173         case CAYMAN:
 174                 evergreen_init_state_functions(rctx);
 175                 evergreen_init_atom_start_cs(rctx);
 176                 evergreen_init_atom_start_compute_cs(rctx);
 177                 rctx->custom_dsa_flush = evergreen_create_db_flush_dsa(rctx);
 178                 rctx->custom_blend_resolve = evergreen_create_resolve_blend(rctx);
 179                 rctx->custom_blend_decompress = evergreen_create_decompress_blend(rctx);
 180                 rctx->custom_blend_fastclear = evergreen_create_fastclear_blend(rctx);
 181                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_CEDAR ||
 182                                            rctx->b.family == CHIP_PALM ||
 183                                            rctx->b.family == CHIP_SUMO ||
 184                                            rctx->b.family == CHIP_SUMO2 ||
 185                                            rctx->b.family == CHIP_CAICOS ||
 186                                            rctx->b.family == CHIP_CAYMAN ||
 187                                            rctx->b.family == CHIP_ARUBA);
 188                 break;
 189         default:
 190                 R600_ERR("Unsupported chip class %d.\n", rctx->b.chip_class);
 191                 goto fail;
 192         }
 193
 194         rctx->b.gfx.cs = ws->cs_create(rctx->b.ctx, RING_GFX,
 195                                        r600_context_gfx_flush, rctx);
 196         rctx->b.gfx.flush = r600_context_gfx_flush;
 197
 198         rctx->allocator_fetch_shader =
 199                 u_suballocator_create(&rctx->b.b, 64 * 1024,
 200                                       0, PIPE_USAGE_DEFAULT, 0, FALSE);
 201         if (!rctx->allocator_fetch_shader)
 202                 goto fail;
 203
 204         rctx->isa = calloc(1, sizeof(struct r600_isa));
 205         if (!rctx->isa || r600_isa_init(rctx, rctx->isa))
 206                 goto fail;
 207
 208         if (rscreen->b.debug_flags & DBG_FORCE_DMA)
 209                 rctx->b.b.resource_copy_region = rctx->b.dma_copy;
 210
 211         rctx->blitter = util_blitter_create(&rctx->b.b);
 212         if (rctx->blitter == NULL)
 213                 goto fail;
 214         util_blitter_set_texture_multisample(rctx->blitter, rscreen->has_msaa);
 215         rctx->blitter->draw_rectangle = r600_draw_rectangle;
 216
 217         r600_begin_new_cs(rctx);
 218
 219         rctx->dummy_pixel_shader =
 220                 util_make_fragment_cloneinput_shader(&rctx->b.b, 0,
 221                                                      TGSI_SEMANTIC_GENERIC,
 222                                                      TGSI_INTERPOLATE_CONSTANT);
 223         rctx->b.b.bind_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
 224
 225         return &rctx->b.b;
 226
 227 fail:
 228         r600_destroy_context(&rctx->b.b);
 229         return NULL;
 230 }
 231
 232 /*
 233  * pipe_screen
 234  */
 235
 236 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 237 {
 238         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 239         enum radeon_family family = rscreen->b.family;
 240
 241         switch (param) {
 242         /* Supported features (boolean caps). */
 243         case PIPE_CAP_NPOT_TEXTURES:
 244         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 245         case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
 246         case PIPE_CAP_TWO_SIDED_STENCIL:
 247         case PIPE_CAP_ANISOTROPIC_FILTER:
 248         case PIPE_CAP_POINT_SPRITE:
 249         case PIPE_CAP_OCCLUSION_QUERY:
 250         case PIPE_CAP_TEXTURE_SHADOW_MAP:
 251         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 252         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 253         case PIPE_CAP_TEXTURE_SWIZZLE:
 254         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 255         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 256         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 257         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 258         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 259         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 260         case PIPE_CAP_SM3:
 261         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 262         case PIPE_CAP_PRIMITIVE_RESTART:
 263         case PIPE_CAP_CONDITIONAL_RENDER:
 264         case PIPE_CAP_TEXTURE_BARRIER:
 265         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 266         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 267         case PIPE_CAP_TGSI_INSTANCEID:
 268         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 269         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 270         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 271         case PIPE_CAP_USER_CONSTANT_BUFFERS:
 272         case PIPE_CAP_START_INSTANCE:
 273         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 274         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 275         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 276         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 277         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 278         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 279         case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 280         case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 281         case PIPE_CAP_SAMPLE_SHADING:
 282         case PIPE_CAP_CLIP_HALFZ:
 283         case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 284         case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 285         case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 286         case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 287         case PIPE_CAP_TGSI_TXQS:
 288         case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 289         case PIPE_CAP_INVALIDATE_BUFFER:
 290         case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 291         case PIPE_CAP_QUERY_MEMORY_INFO:
 292         case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 293         case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 294         case PIPE_CAP_CLEAR_TEXTURE:
 295         case PIPE_CAP_TGSI_MUL_ZERO_WINS:
 296         case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
 297         case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 298                 return 1;
 299
 300         case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 301                 return rscreen->b.info.drm_major == 2 && rscreen->b.info.drm_minor >= 43;
 302
 303         case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 304                 return !R600_BIG_ENDIAN && rscreen->b.info.has_userptr;
 305
 306         case PIPE_CAP_COMPUTE:
 307                 return rscreen->b.chip_class > R700;
 308
 309         case PIPE_CAP_TGSI_TEXCOORD:
 310                 return 0;
 311
 312         case PIPE_CAP_FAKE_SW_MSAA:
 313                 return 0;
 314
 315         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 316                 return MIN2(rscreen->b.info.max_alloc_size, INT_MAX);
 317
 318         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 319                 return R600_MAP_BUFFER_ALIGNMENT;
 320
 321         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 322                 return 256;
 323
 324         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 325                 return 1;
 326
 327         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 328                 if (family >= CHIP_CEDAR)
 329                    return 410;
 330                 /* pre-evergreen geom shaders need newer kernel */
 331                 if (rscreen->b.info.drm_minor >= 37)
 332                    return 330;
 333                 return 140;
 334
 335         /* Supported except the original R600. */
 336         case PIPE_CAP_INDEP_BLEND_ENABLE:
 337         case PIPE_CAP_INDEP_BLEND_FUNC:
 338                 /* R600 doesn't support per-MRT blends */
 339                 return family == CHIP_R600 ? 0 : 1;
 340
 341         /* Supported on Evergreen. */
 342         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 343         case PIPE_CAP_CUBE_MAP_ARRAY:
 344         case PIPE_CAP_TEXTURE_GATHER_SM5:
 345         case PIPE_CAP_TEXTURE_QUERY_LOD:
 346         case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 347         case PIPE_CAP_SAMPLER_VIEW_TARGET:
 348         case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 349                 return family >= CHIP_CEDAR ? 1 : 0;
 350         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 351                 return family >= CHIP_CEDAR ? 4 : 0;
 352         case PIPE_CAP_DRAW_INDIRECT:
 353                 /* kernel command checker support is also required */
 354                 return family >= CHIP_CEDAR && rscreen->b.info.drm_minor >= 41;
 355
 356         case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 357                 return family >= CHIP_CEDAR ? 0 : 1;
 358
 359         /* Unsupported features. */
 360         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 361         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 362         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 363         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 364         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 365         case PIPE_CAP_USER_VERTEX_BUFFERS:
 366         case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 367         case PIPE_CAP_VERTEXID_NOBASE:
 368         case PIPE_CAP_DEPTH_BOUNDS_TEST:
 369         case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 370         case PIPE_CAP_SHAREABLE_SHADERS:
 371         case PIPE_CAP_DRAW_PARAMETERS:
 372         case PIPE_CAP_MULTI_DRAW_INDIRECT:
 373         case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 374         case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 375         case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 376         case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 377         case PIPE_CAP_GENERATE_MIPMAP:
 378         case PIPE_CAP_STRING_MARKER:
 379         case PIPE_CAP_QUERY_BUFFER_OBJECT:
 380         case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 381         case PIPE_CAP_CULL_DISTANCE:
 382         case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 383         case PIPE_CAP_TGSI_VOTE:
 384         case PIPE_CAP_MAX_WINDOW_RECTANGLES:
 385         case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 386         case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
 387         case PIPE_CAP_NATIVE_FENCE_FD:
 388         case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 389         case PIPE_CAP_TGSI_FS_FBFETCH:
 390         case PIPE_CAP_INT64:
 391         case PIPE_CAP_INT64_DIVMOD:
 392         case PIPE_CAP_TGSI_TEX_TXF_LZ:
 393         case PIPE_CAP_TGSI_CLOCK:
 394         case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
 395         case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
 396         case PIPE_CAP_TGSI_BALLOT:
 397         case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
 398                 return 0;
 399
 400         case PIPE_CAP_DOUBLES:
 401                 if (rscreen->b.family == CHIP_ARUBA ||
 402                     rscreen->b.family == CHIP_CAYMAN ||
 403                     rscreen->b.family == CHIP_CYPRESS ||
 404                     rscreen->b.family == CHIP_HEMLOCK)
 405                         return 1;
 406                 return 0;
 407
 408         case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 409                 if (family >= CHIP_CEDAR)
 410                         return 30;
 411                 else
 412                         return 0;
 413         /* Stream output. */
 414         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 415                 return rscreen->b.has_streamout ? 4 : 0;
 416         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 417         case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 418                 return rscreen->b.has_streamout ? 1 : 0;
 419         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 420         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 421                 return 32*4;
 422
 423         /* Geometry shader output. */
 424         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 425                 return 1024;
 426         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 427                 return 16384;
 428         case PIPE_CAP_MAX_VERTEX_STREAMS:
 429                 return family >= CHIP_CEDAR ? 4 : 1;
 430
 431         case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 432                 return 2047;
 433
 434         /* Texturing. */
 435         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 436         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 437                 if (family >= CHIP_CEDAR)
 438                         return 15;
 439                 else
 440                         return 14;
 441         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 442                 /* textures support 8192, but layered rendering supports 2048 */
 443                 return 12;
 444         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 445                 /* textures support 8192, but layered rendering supports 2048 */
 446                 return 2048;
 447
 448         /* Render targets. */
 449         case PIPE_CAP_MAX_RENDER_TARGETS:
 450                 /* XXX some r6xx are buggy and can only do 4 */
 451                 return 8;
 452
 453         case PIPE_CAP_MAX_VIEWPORTS:
 454                 return R600_MAX_VIEWPORTS;
 455         case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 456                 return 8;
 457
 458         /* Timer queries, present when the clock frequency is non zero. */
 459         case PIPE_CAP_QUERY_TIME_ELAPSED:
 460                 return rscreen->b.info.clock_crystal_freq != 0;
 461         case PIPE_CAP_QUERY_TIMESTAMP:
 462                 return rscreen->b.info.drm_minor >= 20 &&
 463                        rscreen->b.info.clock_crystal_freq != 0;
 464
 465         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 466         case PIPE_CAP_MIN_TEXEL_OFFSET:
 467                 return -8;
 468
 469         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 470         case PIPE_CAP_MAX_TEXEL_OFFSET:
 471                 return 7;
 472
 473         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 474                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 475         case PIPE_CAP_ENDIANNESS:
 476                 return PIPE_ENDIAN_LITTLE;
 477
 478         case PIPE_CAP_VENDOR_ID:
 479                 return ATI_VENDOR_ID;
 480         case PIPE_CAP_DEVICE_ID:
 481                 return rscreen->b.info.pci_id;
 482         case PIPE_CAP_ACCELERATED:
 483                 return 1;
 484         case PIPE_CAP_VIDEO_MEMORY:
 485                 return rscreen->b.info.vram_size >> 20;
 486         case PIPE_CAP_UMA:
 487                 return 0;
 488         case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 489                 return rscreen->b.chip_class >= R700;
 490         case PIPE_CAP_PCI_GROUP:
 491                 return rscreen->b.info.pci_domain;
 492         case PIPE_CAP_PCI_BUS:
 493                 return rscreen->b.info.pci_bus;
 494         case PIPE_CAP_PCI_DEVICE:
 495                 return rscreen->b.info.pci_dev;
 496         case PIPE_CAP_PCI_FUNCTION:
 497                 return rscreen->b.info.pci_func;
 498         }
 499         return 0;
 500 }
 501
 502 static int r600_get_shader_param(struct pipe_screen* pscreen,
 503                                  enum pipe_shader_type shader,
 504                                  enum pipe_shader_cap param)
 505 {
 506         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 507
 508         switch(shader)
 509         {
 510         case PIPE_SHADER_FRAGMENT:
 511         case PIPE_SHADER_VERTEX:
 512         case PIPE_SHADER_COMPUTE:
 513                 break;
 514         case PIPE_SHADER_GEOMETRY:
 515                 if (rscreen->b.family >= CHIP_CEDAR)
 516                         break;
 517                 /* pre-evergreen geom shaders need newer kernel */
 518                 if (rscreen->b.info.drm_minor >= 37)
 519                         break;
 520                 return 0;
 521         case PIPE_SHADER_TESS_CTRL:
 522         case PIPE_SHADER_TESS_EVAL:
 523                 if (rscreen->b.family >= CHIP_CEDAR)
 524                         break;
 525         default:
 526                 return 0;
 527         }
 528
 529         switch (param) {
 530         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 531         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 532         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 533         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 534                 return 16384;
 535         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 536                 return 32;
 537         case PIPE_SHADER_CAP_MAX_INPUTS:
 538                 return shader == PIPE_SHADER_VERTEX ? 16 : 32;
 539         case PIPE_SHADER_CAP_MAX_OUTPUTS:
 540                 return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
 541         case PIPE_SHADER_CAP_MAX_TEMPS:
 542                 return 256; /* Max native temporaries. */
 543         case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 544                 if (shader == PIPE_SHADER_COMPUTE) {
 545                         uint64_t max_const_buffer_size;
 546                         pscreen->get_compute_param(pscreen, PIPE_SHADER_IR_TGSI,
 547                                 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
 548                                 &max_const_buffer_size);
 549                         return MIN2(max_const_buffer_size, INT_MAX);
 550
 551                 } else {
 552                         return R600_MAX_CONST_BUFFER_SIZE;
 553                 }
 554         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 555                 return R600_MAX_USER_CONST_BUFFERS;
 556         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 557                 return 1;
 558         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 559                 return 1;
 560         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 561         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 562         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 563         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 564                 return 1;
 565         case PIPE_SHADER_CAP_SUBROUTINES:
 566                 return 0;
 567         case PIPE_SHADER_CAP_INTEGERS:
 568         case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 569                 return 1;
 570         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 571         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 572                 return 16;
 573         case PIPE_SHADER_CAP_PREFERRED_IR:
 574                 if (shader == PIPE_SHADER_COMPUTE) {
 575                         return PIPE_SHADER_IR_NATIVE;
 576                 } else {
 577                         return PIPE_SHADER_IR_TGSI;
 578                 }
 579         case PIPE_SHADER_CAP_SUPPORTED_IRS:
 580                 return 0;
 581         case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 582                 if (rscreen->b.family == CHIP_ARUBA ||
 583                     rscreen->b.family == CHIP_CAYMAN ||
 584                     rscreen->b.family == CHIP_CYPRESS ||
 585                     rscreen->b.family == CHIP_HEMLOCK)
 586                         return 1;
 587                 return 0;
 588         case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 589         case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 590         case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 591         case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 592         case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
 593         case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
 594                 return 0;
 595         case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 596                 /* due to a bug in the shader compiler, some loops hang
 597                  * if they are not unrolled, see:
 598                  *    https://bugs.freedesktop.org/show_bug.cgi?id=86720
 599                  */
 600                 return 255;
 601         }
 602         return 0;
 603 }
 604
 605 static void r600_destroy_screen(struct pipe_screen* pscreen)
 606 {
 607         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 608
 609         if (!rscreen)
 610                 return;
 611
 612         if (!rscreen->b.ws->unref(rscreen->b.ws))
 613                 return;
 614
 615         if (rscreen->global_pool) {
 616                 compute_memory_pool_delete(rscreen->global_pool);
 617         }
 618
 619         r600_destroy_common_screen(&rscreen->b);
 620 }
 621
 622 static struct pipe_resource *r600_resource_create(struct pipe_screen *screen,
 623                                                   const struct pipe_resource *templ)
 624 {
 625         if (templ->target == PIPE_BUFFER &&
 626             (templ->bind & PIPE_BIND_GLOBAL))
 627                 return r600_compute_global_buffer_create(screen, templ);
 628
 629         return r600_resource_create_common(screen, templ);
 630 }
 631
 632 struct pipe_screen *r600_screen_create(struct radeon_winsys *ws)
 633 {
 634         struct r600_screen *rscreen = CALLOC_STRUCT(r600_screen);
 635
 636         if (!rscreen) {
 637                 return NULL;
 638         }
 639
 640         /* Set functions first. */
 641         rscreen->b.b.context_create = r600_create_context;
 642         rscreen->b.b.destroy = r600_destroy_screen;
 643         rscreen->b.b.get_param = r600_get_param;
 644         rscreen->b.b.get_shader_param = r600_get_shader_param;
 645         rscreen->b.b.resource_create = r600_resource_create;
 646
 647         if (!r600_common_screen_init(&rscreen->b, ws)) {
 648                 FREE(rscreen);
 649                 return NULL;
 650         }
 651
 652         if (rscreen->b.info.chip_class >= EVERGREEN) {
 653                 rscreen->b.b.is_format_supported = evergreen_is_format_supported;
 654         } else {
 655                 rscreen->b.b.is_format_supported = r600_is_format_supported;
 656         }
 657
 658         rscreen->b.debug_flags |= debug_get_flags_option("R600_DEBUG", r600_debug_options, 0);
 659         if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE))
 660                 rscreen->b.debug_flags |= DBG_COMPUTE;
 661         if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE))
 662                 rscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS | DBG_TCS | DBG_TES;
 663         if (!debug_get_bool_option("R600_HYPERZ", TRUE))
 664                 rscreen->b.debug_flags |= DBG_NO_HYPERZ;
 665
 666         if (rscreen->b.family == CHIP_UNKNOWN) {
 667                 fprintf(stderr, "r600: Unknown chipset 0x%04X\n", rscreen->b.info.pci_id);
 668                 FREE(rscreen);
 669                 return NULL;
 670         }
 671
 672         /* Figure out streamout kernel support. */
 673         switch (rscreen->b.chip_class) {
 674         case R600:
 675                 if (rscreen->b.family < CHIP_RS780) {
 676                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 677                 } else {
 678                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 23;
 679                 }
 680                 break;
 681         case R700:
 682                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 17;
 683                 break;
 684         case EVERGREEN:
 685         case CAYMAN:
 686                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 687                 break;
 688         default:
 689                 rscreen->b.has_streamout = FALSE;
 690                 break;
 691         }
 692
 693         /* MSAA support. */
 694         switch (rscreen->b.chip_class) {
 695         case R600:
 696         case R700:
 697                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 22;
 698                 rscreen->has_compressed_msaa_texturing = false;
 699                 break;
 700         case EVERGREEN:
 701                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 702                 rscreen->has_compressed_msaa_texturing = rscreen->b.info.drm_minor >= 24;
 703                 break;
 704         case CAYMAN:
 705                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 706                 rscreen->has_compressed_msaa_texturing = true;
 707                 break;
 708         default:
 709                 rscreen->has_msaa = FALSE;
 710                 rscreen->has_compressed_msaa_texturing = false;
 711         }
 712
 713         rscreen->b.has_cp_dma = rscreen->b.info.drm_minor >= 27 &&
 714                               !(rscreen->b.debug_flags & DBG_NO_CP_DMA);
 715
 716         rscreen->b.barrier_flags.cp_to_L2 =
 717                 R600_CONTEXT_INV_VERTEX_CACHE |
 718                 R600_CONTEXT_INV_TEX_CACHE |
 719                 R600_CONTEXT_INV_CONST_CACHE;
 720         rscreen->b.barrier_flags.compute_to_L2 = R600_CONTEXT_PS_PARTIAL_FLUSH;
 721
 722         rscreen->global_pool = compute_memory_pool_new(rscreen);
 723
 724         /* Create the auxiliary context. This must be done last. */
 725         rscreen->b.aux_context = rscreen->b.b.context_create(&rscreen->b.b, NULL, 0);
 726
 727 #if 0 /* This is for testing whether aux_context and buffer clearing work correctly. */
 728         struct pipe_resource templ = {};
 729
 730         templ.width0 = 4;
 731         templ.height0 = 2048;
 732         templ.depth0 = 1;
 733         templ.array_size = 1;
 734         templ.target = PIPE_TEXTURE_2D;
 735         templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 736         templ.usage = PIPE_USAGE_DEFAULT;
 737
 738         struct r600_resource *res = r600_resource(rscreen->screen.resource_create(&rscreen->screen, &templ));
 739         unsigned char *map = ws->buffer_map(res->buf, NULL, PIPE_TRANSFER_WRITE);
 740
 741         memset(map, 0, 256);
 742
 743         r600_screen_clear_buffer(rscreen, &res->b.b, 4, 4, 0xCC);
 744         r600_screen_clear_buffer(rscreen, &res->b.b, 8, 4, 0xDD);
 745         r600_screen_clear_buffer(rscreen, &res->b.b, 12, 4, 0xEE);
 746         r600_screen_clear_buffer(rscreen, &res->b.b, 20, 4, 0xFF);
 747         r600_screen_clear_buffer(rscreen, &res->b.b, 32, 20, 0x87);
 748
 749         ws->buffer_wait(res->buf, RADEON_USAGE_WRITE);
 750
 751         int i;
 752         for (i = 0; i < 256; i++) {
 753                 printf("%02X", map[i]);
 754                 if (i % 16 == 15)
 755                         printf("\n");
 756         }
 757 #endif
 758
 759         if (rscreen->b.debug_flags & DBG_TEST_DMA)
 760                 r600_test_dma(&rscreen->b);
 761
 762         r600_query_fix_enabled_rb_mask(&rscreen->b);
 763         return &rscreen->b.b;
 764 }