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, i;
  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         for (sh = 0; sh < PIPE_SHADER_TYPES; ++sh)
 108                 for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; ++i)
 109                         rctx->b.b.set_constant_buffer(context, sh, i, NULL);
 110
 111         if (rctx->blitter) {
 112                 util_blitter_destroy(rctx->blitter);
 113         }
 114         if (rctx->allocator_fetch_shader) {
 115                 u_suballocator_destroy(rctx->allocator_fetch_shader);
 116         }
 117
 118         r600_release_command_buffer(&rctx->start_cs_cmd);
 119
 120         FREE(rctx->start_compute_cs_cmd.buf);
 121
 122         r600_common_context_cleanup(&rctx->b);
 123
 124         r600_resource_reference(&rctx->trace_buf, NULL);
 125         r600_resource_reference(&rctx->last_trace_buf, NULL);
 126         radeon_clear_saved_cs(&rctx->last_gfx);
 127
 128         FREE(rctx);
 129 }
 130
 131 static struct pipe_context *r600_create_context(struct pipe_screen *screen,
 132                                                 void *priv, unsigned flags)
 133 {
 134         struct r600_context *rctx = CALLOC_STRUCT(r600_context);
 135         struct r600_screen* rscreen = (struct r600_screen *)screen;
 136         struct radeon_winsys *ws = rscreen->b.ws;
 137
 138         if (!rctx)
 139                 return NULL;
 140
 141         rctx->b.b.screen = screen;
 142         assert(!priv);
 143         rctx->b.b.priv = NULL; /* for threaded_context_unwrap_sync */
 144         rctx->b.b.destroy = r600_destroy_context;
 145         rctx->b.set_atom_dirty = (void *)r600_set_atom_dirty;
 146
 147         if (!r600_common_context_init(&rctx->b, &rscreen->b, flags))
 148                 goto fail;
 149
 150         rctx->screen = rscreen;
 151         LIST_INITHEAD(&rctx->texture_buffers);
 152
 153         r600_init_blit_functions(rctx);
 154
 155         if (rscreen->b.info.has_hw_decode) {
 156                 rctx->b.b.create_video_codec = r600_uvd_create_decoder;
 157                 rctx->b.b.create_video_buffer = r600_video_buffer_create;
 158         } else {
 159                 rctx->b.b.create_video_codec = vl_create_decoder;
 160                 rctx->b.b.create_video_buffer = vl_video_buffer_create;
 161         }
 162
 163         if (getenv("R600_TRACE"))
 164                 rctx->is_debug = true;
 165         r600_init_common_state_functions(rctx);
 166
 167         switch (rctx->b.chip_class) {
 168         case R600:
 169         case R700:
 170                 r600_init_state_functions(rctx);
 171                 r600_init_atom_start_cs(rctx);
 172                 rctx->custom_dsa_flush = r600_create_db_flush_dsa(rctx);
 173                 rctx->custom_blend_resolve = rctx->b.chip_class == R700 ? r700_create_resolve_blend(rctx)
 174                                                                       : r600_create_resolve_blend(rctx);
 175                 rctx->custom_blend_decompress = r600_create_decompress_blend(rctx);
 176                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_RV610 ||
 177                                            rctx->b.family == CHIP_RV620 ||
 178                                            rctx->b.family == CHIP_RS780 ||
 179                                            rctx->b.family == CHIP_RS880 ||
 180                                            rctx->b.family == CHIP_RV710);
 181                 break;
 182         case EVERGREEN:
 183         case CAYMAN:
 184                 evergreen_init_state_functions(rctx);
 185                 evergreen_init_atom_start_cs(rctx);
 186                 evergreen_init_atom_start_compute_cs(rctx);
 187                 rctx->custom_dsa_flush = evergreen_create_db_flush_dsa(rctx);
 188                 rctx->custom_blend_resolve = evergreen_create_resolve_blend(rctx);
 189                 rctx->custom_blend_decompress = evergreen_create_decompress_blend(rctx);
 190                 rctx->custom_blend_fastclear = evergreen_create_fastclear_blend(rctx);
 191                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_CEDAR ||
 192                                            rctx->b.family == CHIP_PALM ||
 193                                            rctx->b.family == CHIP_SUMO ||
 194                                            rctx->b.family == CHIP_SUMO2 ||
 195                                            rctx->b.family == CHIP_CAICOS ||
 196                                            rctx->b.family == CHIP_CAYMAN ||
 197                                            rctx->b.family == CHIP_ARUBA);
 198
 199                 rctx->append_fence = pipe_buffer_create(rctx->b.b.screen, PIPE_BIND_CUSTOM,
 200                                                          PIPE_USAGE_DEFAULT, 32);
 201                 break;
 202         default:
 203                 R600_ERR("Unsupported chip class %d.\n", rctx->b.chip_class);
 204                 goto fail;
 205         }
 206
 207         rctx->b.gfx.cs = ws->cs_create(rctx->b.ctx, RING_GFX,
 208                                        r600_context_gfx_flush, rctx);
 209         rctx->b.gfx.flush = r600_context_gfx_flush;
 210
 211         rctx->allocator_fetch_shader =
 212                 u_suballocator_create(&rctx->b.b, 64 * 1024,
 213                                       0, PIPE_USAGE_DEFAULT, 0, FALSE);
 214         if (!rctx->allocator_fetch_shader)
 215                 goto fail;
 216
 217         rctx->isa = calloc(1, sizeof(struct r600_isa));
 218         if (!rctx->isa || r600_isa_init(rctx, rctx->isa))
 219                 goto fail;
 220
 221         if (rscreen->b.debug_flags & DBG_FORCE_DMA)
 222                 rctx->b.b.resource_copy_region = rctx->b.dma_copy;
 223
 224         rctx->blitter = util_blitter_create(&rctx->b.b);
 225         if (rctx->blitter == NULL)
 226                 goto fail;
 227         util_blitter_set_texture_multisample(rctx->blitter, rscreen->has_msaa);
 228         rctx->blitter->draw_rectangle = r600_draw_rectangle;
 229
 230         r600_begin_new_cs(rctx);
 231
 232         rctx->dummy_pixel_shader =
 233                 util_make_fragment_cloneinput_shader(&rctx->b.b, 0,
 234                                                      TGSI_SEMANTIC_GENERIC,
 235                                                      TGSI_INTERPOLATE_CONSTANT);
 236         rctx->b.b.bind_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
 237
 238         return &rctx->b.b;
 239
 240 fail:
 241         r600_destroy_context(&rctx->b.b);
 242         return NULL;
 243 }
 244
 245 /*
 246  * pipe_screen
 247  */
 248
 249 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 250 {
 251         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 252         enum radeon_family family = rscreen->b.family;
 253
 254         switch (param) {
 255         /* Supported features (boolean caps). */
 256         case PIPE_CAP_NPOT_TEXTURES:
 257         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 258         case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
 259         case PIPE_CAP_ANISOTROPIC_FILTER:
 260         case PIPE_CAP_POINT_SPRITE:
 261         case PIPE_CAP_OCCLUSION_QUERY:
 262         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 263         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 264         case PIPE_CAP_TEXTURE_SWIZZLE:
 265         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 266         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 267         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 268         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 269         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 270         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 271         case PIPE_CAP_SM3:
 272         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 273         case PIPE_CAP_PRIMITIVE_RESTART:
 274         case PIPE_CAP_CONDITIONAL_RENDER:
 275         case PIPE_CAP_TEXTURE_BARRIER:
 276         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 277         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 278         case PIPE_CAP_TGSI_INSTANCEID:
 279         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 280         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 281         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 282         case PIPE_CAP_START_INSTANCE:
 283         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 284         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 285         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 286         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 287         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 288         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 289         case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 290         case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 291         case PIPE_CAP_SAMPLE_SHADING:
 292         case PIPE_CAP_CLIP_HALFZ:
 293         case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 294         case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 295         case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 296         case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 297         case PIPE_CAP_TGSI_TXQS:
 298         case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 299         case PIPE_CAP_INVALIDATE_BUFFER:
 300         case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 301         case PIPE_CAP_QUERY_MEMORY_INFO:
 302         case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 303         case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 304         case PIPE_CAP_CLEAR_TEXTURE:
 305         case PIPE_CAP_TGSI_MUL_ZERO_WINS:
 306         case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
 307         case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 308         case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 309                 return 1;
 310
 311         case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 312                 return rscreen->b.info.drm_major == 2 && rscreen->b.info.drm_minor >= 43;
 313
 314         case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 315                 return !R600_BIG_ENDIAN && rscreen->b.info.has_userptr;
 316
 317         case PIPE_CAP_COMPUTE:
 318                 return rscreen->b.chip_class > R700;
 319
 320         case PIPE_CAP_TGSI_TEXCOORD:
 321                 return 0;
 322
 323         case PIPE_CAP_FAKE_SW_MSAA:
 324                 return 0;
 325
 326         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 327                 return MIN2(rscreen->b.info.max_alloc_size, INT_MAX);
 328
 329         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 330                 return R600_MAP_BUFFER_ALIGNMENT;
 331
 332         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 333                 return 256;
 334
 335         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 336                 return 1;
 337
 338         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 339                 if (family >= CHIP_CEDAR)
 340                    return 430;
 341                 /* pre-evergreen geom shaders need newer kernel */
 342                 if (rscreen->b.info.drm_minor >= 37)
 343                    return 330;
 344                 return 140;
 345
 346         case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
 347                 return 140;
 348
 349         /* Supported except the original R600. */
 350         case PIPE_CAP_INDEP_BLEND_ENABLE:
 351         case PIPE_CAP_INDEP_BLEND_FUNC:
 352                 /* R600 doesn't support per-MRT blends */
 353                 return family == CHIP_R600 ? 0 : 1;
 354
 355         /* Supported on Evergreen. */
 356         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 357         case PIPE_CAP_CUBE_MAP_ARRAY:
 358         case PIPE_CAP_TEXTURE_GATHER_SM5:
 359         case PIPE_CAP_TEXTURE_QUERY_LOD:
 360         case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 361         case PIPE_CAP_SAMPLER_VIEW_TARGET:
 362         case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 363         case PIPE_CAP_TGSI_CLOCK:
 364         case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 365         case PIPE_CAP_QUERY_BUFFER_OBJECT:
 366                 return family >= CHIP_CEDAR ? 1 : 0;
 367         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 368                 return family >= CHIP_CEDAR ? 4 : 0;
 369         case PIPE_CAP_DRAW_INDIRECT:
 370                 /* kernel command checker support is also required */
 371                 return family >= CHIP_CEDAR && rscreen->b.info.drm_minor >= 41;
 372
 373         case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 374                 return family >= CHIP_CEDAR ? 0 : 1;
 375
 376         case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
 377                 return 8;
 378
 379         case PIPE_CAP_MAX_GS_INVOCATIONS:
 380                 return 32;
 381
 382         /* Unsupported features. */
 383         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 384         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 385         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 386         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 387         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 388         case PIPE_CAP_USER_VERTEX_BUFFERS:
 389         case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 390         case PIPE_CAP_VERTEXID_NOBASE:
 391         case PIPE_CAP_DEPTH_BOUNDS_TEST:
 392         case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 393         case PIPE_CAP_SHAREABLE_SHADERS:
 394         case PIPE_CAP_DRAW_PARAMETERS:
 395         case PIPE_CAP_MULTI_DRAW_INDIRECT:
 396         case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 397         case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 398         case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 399         case PIPE_CAP_GENERATE_MIPMAP:
 400         case PIPE_CAP_STRING_MARKER:
 401         case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
 402         case PIPE_CAP_TGSI_VOTE:
 403         case PIPE_CAP_MAX_WINDOW_RECTANGLES:
 404         case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
 405         case PIPE_CAP_NATIVE_FENCE_FD:
 406         case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 407         case PIPE_CAP_TGSI_FS_FBFETCH:
 408         case PIPE_CAP_INT64:
 409         case PIPE_CAP_INT64_DIVMOD:
 410         case PIPE_CAP_TGSI_TEX_TXF_LZ:
 411         case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
 412         case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
 413         case PIPE_CAP_TGSI_BALLOT:
 414         case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
 415         case PIPE_CAP_POST_DEPTH_COVERAGE:
 416         case PIPE_CAP_BINDLESS_TEXTURE:
 417         case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
 418         case PIPE_CAP_QUERY_SO_OVERFLOW:
 419         case PIPE_CAP_MEMOBJ:
 420         case PIPE_CAP_LOAD_CONSTBUF:
 421         case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
 422         case PIPE_CAP_TILE_RASTER_ORDER:
 423         case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
 424         case PIPE_CAP_CONTEXT_PRIORITY_MASK:
 425         case PIPE_CAP_FENCE_SIGNAL:
 426         case PIPE_CAP_CONSTBUF0_FLAGS:
 427         case PIPE_CAP_PACKED_UNIFORMS:
 428         case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS:
 429         case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
 430         case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
 431         case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
 432         case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
 433         case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
 434         case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
 435         case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
 436                 return 0;
 437
 438         case PIPE_CAP_DOUBLES:
 439                 if (rscreen->b.family == CHIP_ARUBA ||
 440                     rscreen->b.family == CHIP_CAYMAN ||
 441                     rscreen->b.family == CHIP_CYPRESS ||
 442                     rscreen->b.family == CHIP_HEMLOCK)
 443                         return 1;
 444                 return 0;
 445         case PIPE_CAP_CULL_DISTANCE:
 446                 return 1;
 447
 448         case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 449                 if (family >= CHIP_CEDAR)
 450                         return 256;
 451                 return 0;
 452
 453         case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 454                 if (family >= CHIP_CEDAR)
 455                         return 30;
 456                 else
 457                         return 0;
 458         /* Stream output. */
 459         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 460                 return rscreen->b.has_streamout ? 4 : 0;
 461         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 462         case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 463                 return rscreen->b.has_streamout ? 1 : 0;
 464         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 465         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 466                 return 32*4;
 467
 468         /* Geometry shader output. */
 469         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 470                 return 1024;
 471         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 472                 return 16384;
 473         case PIPE_CAP_MAX_VERTEX_STREAMS:
 474                 return family >= CHIP_CEDAR ? 4 : 1;
 475
 476         case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 477                 /* Should be 2047, but 2048 is a requirement for GL 4.4 */
 478                 return 2048;
 479
 480         /* Texturing. */
 481         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 482         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 483                 if (family >= CHIP_CEDAR)
 484                         return 15;
 485                 else
 486                         return 14;
 487         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 488                 /* textures support 8192, but layered rendering supports 2048 */
 489                 return 12;
 490         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 491                 /* textures support 8192, but layered rendering supports 2048 */
 492                 return 2048;
 493
 494         /* Render targets. */
 495         case PIPE_CAP_MAX_RENDER_TARGETS:
 496                 /* XXX some r6xx are buggy and can only do 4 */
 497                 return 8;
 498
 499         case PIPE_CAP_MAX_VIEWPORTS:
 500                 return R600_MAX_VIEWPORTS;
 501         case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 502                 return 8;
 503
 504         /* Timer queries, present when the clock frequency is non zero. */
 505         case PIPE_CAP_QUERY_TIME_ELAPSED:
 506                 return rscreen->b.info.clock_crystal_freq != 0;
 507         case PIPE_CAP_QUERY_TIMESTAMP:
 508                 return rscreen->b.info.drm_minor >= 20 &&
 509                        rscreen->b.info.clock_crystal_freq != 0;
 510
 511         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 512         case PIPE_CAP_MIN_TEXEL_OFFSET:
 513                 return -8;
 514
 515         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 516         case PIPE_CAP_MAX_TEXEL_OFFSET:
 517                 return 7;
 518
 519         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 520                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 521         case PIPE_CAP_ENDIANNESS:
 522                 return PIPE_ENDIAN_LITTLE;
 523
 524         case PIPE_CAP_VENDOR_ID:
 525                 return ATI_VENDOR_ID;
 526         case PIPE_CAP_DEVICE_ID:
 527                 return rscreen->b.info.pci_id;
 528         case PIPE_CAP_ACCELERATED:
 529                 return 1;
 530         case PIPE_CAP_VIDEO_MEMORY:
 531                 return rscreen->b.info.vram_size >> 20;
 532         case PIPE_CAP_UMA:
 533                 return 0;
 534         case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 535                 return rscreen->b.chip_class >= R700;
 536         case PIPE_CAP_PCI_GROUP:
 537                 return rscreen->b.info.pci_domain;
 538         case PIPE_CAP_PCI_BUS:
 539                 return rscreen->b.info.pci_bus;
 540         case PIPE_CAP_PCI_DEVICE:
 541                 return rscreen->b.info.pci_dev;
 542         case PIPE_CAP_PCI_FUNCTION:
 543                 return rscreen->b.info.pci_func;
 544         }
 545         return 0;
 546 }
 547
 548 static int r600_get_shader_param(struct pipe_screen* pscreen,
 549                                  enum pipe_shader_type shader,
 550                                  enum pipe_shader_cap param)
 551 {
 552         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 553
 554         switch(shader)
 555         {
 556         case PIPE_SHADER_FRAGMENT:
 557         case PIPE_SHADER_VERTEX:
 558         case PIPE_SHADER_COMPUTE:
 559                 break;
 560         case PIPE_SHADER_GEOMETRY:
 561                 if (rscreen->b.family >= CHIP_CEDAR)
 562                         break;
 563                 /* pre-evergreen geom shaders need newer kernel */
 564                 if (rscreen->b.info.drm_minor >= 37)
 565                         break;
 566                 return 0;
 567         case PIPE_SHADER_TESS_CTRL:
 568         case PIPE_SHADER_TESS_EVAL:
 569                 if (rscreen->b.family >= CHIP_CEDAR)
 570                         break;
 571         default:
 572                 return 0;
 573         }
 574
 575         switch (param) {
 576         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 577         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 578         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 579         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 580                 return 16384;
 581         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 582                 return 32;
 583         case PIPE_SHADER_CAP_MAX_INPUTS:
 584                 return shader == PIPE_SHADER_VERTEX ? 16 : 32;
 585         case PIPE_SHADER_CAP_MAX_OUTPUTS:
 586                 return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
 587         case PIPE_SHADER_CAP_MAX_TEMPS:
 588                 return 256; /* Max native temporaries. */
 589         case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 590                 if (shader == PIPE_SHADER_COMPUTE) {
 591                         uint64_t max_const_buffer_size;
 592                         pscreen->get_compute_param(pscreen, PIPE_SHADER_IR_TGSI,
 593                                 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
 594                                 &max_const_buffer_size);
 595                         return MIN2(max_const_buffer_size, INT_MAX);
 596
 597                 } else {
 598                         return R600_MAX_CONST_BUFFER_SIZE;
 599                 }
 600         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 601                 return R600_MAX_USER_CONST_BUFFERS;
 602         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 603                 return 1;
 604         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 605                 return 1;
 606         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 607         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 608         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 609         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 610                 return 1;
 611         case PIPE_SHADER_CAP_SUBROUTINES:
 612         case PIPE_SHADER_CAP_INT64_ATOMICS:
 613         case PIPE_SHADER_CAP_FP16:
 614                 return 0;
 615         case PIPE_SHADER_CAP_INTEGERS:
 616         case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 617                 return 1;
 618         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 619         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 620                 return 16;
 621         case PIPE_SHADER_CAP_PREFERRED_IR:
 622                 return PIPE_SHADER_IR_TGSI;
 623         case PIPE_SHADER_CAP_SUPPORTED_IRS: {
 624                 int ir = 0;
 625                 if (shader == PIPE_SHADER_COMPUTE)
 626                         ir = 1 << PIPE_SHADER_IR_NATIVE;
 627                 if (rscreen->b.family >= CHIP_CEDAR)
 628                         ir |= 1 << PIPE_SHADER_IR_TGSI;
 629                 return ir;
 630         }
 631         case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 632                 if (rscreen->b.family == CHIP_ARUBA ||
 633                     rscreen->b.family == CHIP_CAYMAN ||
 634                     rscreen->b.family == CHIP_CYPRESS ||
 635                     rscreen->b.family == CHIP_HEMLOCK)
 636                         return 1;
 637                 return 0;
 638         case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 639         case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 640         case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
 641         case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
 642         case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
 643                 return 0;
 644         case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 645         case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 646                 if (rscreen->b.family >= CHIP_CEDAR &&
 647                     (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE))
 648                     return 8;
 649                 return 0;
 650         case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
 651                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics)
 652                         return 8;
 653                 return 0;
 654         case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
 655                 /* having to allocate the atomics out amongst shaders stages is messy,
 656                    so give compute 8 buffers and all the others one */
 657                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics) {
 658                         return EG_MAX_ATOMIC_BUFFERS;
 659                 }
 660                 return 0;
 661         case PIPE_SHADER_CAP_SCALAR_ISA:
 662                 return 0;
 663         case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 664                 /* due to a bug in the shader compiler, some loops hang
 665                  * if they are not unrolled, see:
 666                  *    https://bugs.freedesktop.org/show_bug.cgi?id=86720
 667                  */
 668                 return 255;
 669         }
 670         return 0;
 671 }
 672
 673 static void r600_destroy_screen(struct pipe_screen* pscreen)
 674 {
 675         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 676
 677         if (!rscreen)
 678                 return;
 679
 680         if (!rscreen->b.ws->unref(rscreen->b.ws))
 681                 return;
 682
 683         if (rscreen->global_pool) {
 684                 compute_memory_pool_delete(rscreen->global_pool);
 685         }
 686
 687         r600_destroy_common_screen(&rscreen->b);
 688 }
 689
 690 static struct pipe_resource *r600_resource_create(struct pipe_screen *screen,
 691                                                   const struct pipe_resource *templ)
 692 {
 693         if (templ->target == PIPE_BUFFER &&
 694             (templ->bind & PIPE_BIND_GLOBAL))
 695                 return r600_compute_global_buffer_create(screen, templ);
 696
 697         return r600_resource_create_common(screen, templ);
 698 }
 699
 700 struct pipe_screen *r600_screen_create(struct radeon_winsys *ws,
 701                                        const struct pipe_screen_config *config)
 702 {
 703         struct r600_screen *rscreen = CALLOC_STRUCT(r600_screen);
 704
 705         if (!rscreen) {
 706                 return NULL;
 707         }
 708
 709         /* Set functions first. */
 710         rscreen->b.b.context_create = r600_create_context;
 711         rscreen->b.b.destroy = r600_destroy_screen;
 712         rscreen->b.b.get_param = r600_get_param;
 713         rscreen->b.b.get_shader_param = r600_get_shader_param;
 714         rscreen->b.b.resource_create = r600_resource_create;
 715
 716         if (!r600_common_screen_init(&rscreen->b, ws)) {
 717                 FREE(rscreen);
 718                 return NULL;
 719         }
 720
 721         if (rscreen->b.info.chip_class >= EVERGREEN) {
 722                 rscreen->b.b.is_format_supported = evergreen_is_format_supported;
 723         } else {
 724                 rscreen->b.b.is_format_supported = r600_is_format_supported;
 725         }
 726
 727         rscreen->b.debug_flags |= debug_get_flags_option("R600_DEBUG", r600_debug_options, 0);
 728         if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE))
 729                 rscreen->b.debug_flags |= DBG_COMPUTE;
 730         if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE))
 731                 rscreen->b.debug_flags |= DBG_ALL_SHADERS | DBG_FS;
 732         if (!debug_get_bool_option("R600_HYPERZ", TRUE))
 733                 rscreen->b.debug_flags |= DBG_NO_HYPERZ;
 734
 735         if (rscreen->b.family == CHIP_UNKNOWN) {
 736                 fprintf(stderr, "r600: Unknown chipset 0x%04X\n", rscreen->b.info.pci_id);
 737                 FREE(rscreen);
 738                 return NULL;
 739         }
 740
 741         /* Figure out streamout kernel support. */
 742         switch (rscreen->b.chip_class) {
 743         case R600:
 744                 if (rscreen->b.family < CHIP_RS780) {
 745                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 746                 } else {
 747                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 23;
 748                 }
 749                 break;
 750         case R700:
 751                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 17;
 752                 break;
 753         case EVERGREEN:
 754         case CAYMAN:
 755                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 756                 break;
 757         default:
 758                 rscreen->b.has_streamout = FALSE;
 759                 break;
 760         }
 761
 762         /* MSAA support. */
 763         switch (rscreen->b.chip_class) {
 764         case R600:
 765         case R700:
 766                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 22;
 767                 rscreen->has_compressed_msaa_texturing = false;
 768                 break;
 769         case EVERGREEN:
 770                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 771                 rscreen->has_compressed_msaa_texturing = rscreen->b.info.drm_minor >= 24;
 772                 break;
 773         case CAYMAN:
 774                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 775                 rscreen->has_compressed_msaa_texturing = true;
 776                 break;
 777         default:
 778                 rscreen->has_msaa = FALSE;
 779                 rscreen->has_compressed_msaa_texturing = false;
 780         }
 781
 782         rscreen->b.has_cp_dma = rscreen->b.info.drm_minor >= 27 &&
 783                               !(rscreen->b.debug_flags & DBG_NO_CP_DMA);
 784
 785         rscreen->b.barrier_flags.cp_to_L2 =
 786                 R600_CONTEXT_INV_VERTEX_CACHE |
 787                 R600_CONTEXT_INV_TEX_CACHE |
 788                 R600_CONTEXT_INV_CONST_CACHE;
 789         rscreen->b.barrier_flags.compute_to_L2 = R600_CONTEXT_CS_PARTIAL_FLUSH | R600_CONTEXT_FLUSH_AND_INV;
 790
 791         rscreen->global_pool = compute_memory_pool_new(rscreen);
 792
 793         /* Create the auxiliary context. This must be done last. */
 794         rscreen->b.aux_context = rscreen->b.b.context_create(&rscreen->b.b, NULL, 0);
 795
 796         rscreen->has_atomics = rscreen->b.info.drm_minor >= 44;
 797 #if 0 /* This is for testing whether aux_context and buffer clearing work correctly. */
 798         struct pipe_resource templ = {};
 799
 800         templ.width0 = 4;
 801         templ.height0 = 2048;
 802         templ.depth0 = 1;
 803         templ.array_size = 1;
 804         templ.target = PIPE_TEXTURE_2D;
 805         templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 806         templ.usage = PIPE_USAGE_DEFAULT;
 807
 808         struct r600_resource *res = r600_resource(rscreen->screen.resource_create(&rscreen->screen, &templ));
 809         unsigned char *map = ws->buffer_map(res->buf, NULL, PIPE_TRANSFER_WRITE);
 810
 811         memset(map, 0, 256);
 812
 813         r600_screen_clear_buffer(rscreen, &res->b.b, 4, 4, 0xCC);
 814         r600_screen_clear_buffer(rscreen, &res->b.b, 8, 4, 0xDD);
 815         r600_screen_clear_buffer(rscreen, &res->b.b, 12, 4, 0xEE);
 816         r600_screen_clear_buffer(rscreen, &res->b.b, 20, 4, 0xFF);
 817         r600_screen_clear_buffer(rscreen, &res->b.b, 32, 20, 0x87);
 818
 819         ws->buffer_wait(res->buf, RADEON_USAGE_WRITE);
 820
 821         int i;
 822         for (i = 0; i < 256; i++) {
 823                 printf("%02X", map[i]);
 824                 if (i % 16 == 15)
 825                         printf("\n");
 826         }
 827 #endif
 828
 829         if (rscreen->b.debug_flags & DBG_TEST_DMA)
 830                 r600_test_dma(&rscreen->b);
 831
 832         r600_query_fix_enabled_rb_mask(&rscreen->b);
 833         return &rscreen->b.b;
 834 }