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_screen.h"
  36 #include "util/u_simple_shaders.h"
  37 #include "util/u_upload_mgr.h"
  38 #include "util/u_math.h"
  39 #include "vl/vl_decoder.h"
  40 #include "vl/vl_video_buffer.h"
  41 #include "radeon_video.h"
  42 #include "radeon_uvd.h"
  43 #include "util/os_time.h"
  44
  45 static const struct debug_named_value r600_debug_options[] = {
  46         /* features */
  47         { "nocpdma", DBG_NO_CP_DMA, "Disable CP DMA" },
  48
  49         /* shader backend */
  50         { "nosb", DBG_NO_SB, "Disable sb backend for graphics shaders" },
  51         { "sbcl", DBG_SB_CS, "Enable sb backend for compute shaders" },
  52         { "sbdry", DBG_SB_DRY_RUN, "Don't use optimized bytecode (just print the dumps)" },
  53         { "sbstat", DBG_SB_STAT, "Print optimization statistics for shaders" },
  54         { "sbdump", DBG_SB_DUMP, "Print IR dumps after some optimization passes" },
  55         { "sbnofallback", DBG_SB_NO_FALLBACK, "Abort on errors instead of fallback" },
  56         { "sbdisasm", DBG_SB_DISASM, "Use sb disassembler for shader dumps" },
  57         { "sbsafemath", DBG_SB_SAFEMATH, "Disable unsafe math optimizations" },
  58
  59         DEBUG_NAMED_VALUE_END /* must be last */
  60 };
  61
  62 /*
  63  * pipe_context
  64  */
  65
  66 static void r600_destroy_context(struct pipe_context *context)
  67 {
  68         struct r600_context *rctx = (struct r600_context *)context;
  69         unsigned sh, i;
  70
  71         r600_isa_destroy(rctx->isa);
  72
  73         r600_sb_context_destroy(rctx->sb_context);
  74
  75         for (sh = 0; sh < (rctx->b.chip_class < EVERGREEN ? R600_NUM_HW_STAGES : EG_NUM_HW_STAGES); sh++) {
  76                 r600_resource_reference(&rctx->scratch_buffers[sh].buffer, NULL);
  77         }
  78         r600_resource_reference(&rctx->dummy_cmask, NULL);
  79         r600_resource_reference(&rctx->dummy_fmask, NULL);
  80
  81         if (rctx->append_fence)
  82                 pipe_resource_reference((struct pipe_resource**)&rctx->append_fence, NULL);
  83         for (sh = 0; sh < PIPE_SHADER_TYPES; sh++) {
  84                 rctx->b.b.set_constant_buffer(&rctx->b.b, sh, R600_BUFFER_INFO_CONST_BUFFER, NULL);
  85                 free(rctx->driver_consts[sh].constants);
  86         }
  87
  88         if (rctx->fixed_func_tcs_shader)
  89                 rctx->b.b.delete_tcs_state(&rctx->b.b, rctx->fixed_func_tcs_shader);
  90
  91         if (rctx->dummy_pixel_shader) {
  92                 rctx->b.b.delete_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
  93         }
  94         if (rctx->custom_dsa_flush) {
  95                 rctx->b.b.delete_depth_stencil_alpha_state(&rctx->b.b, rctx->custom_dsa_flush);
  96         }
  97         if (rctx->custom_blend_resolve) {
  98                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_resolve);
  99         }
 100         if (rctx->custom_blend_decompress) {
 101                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_decompress);
 102         }
 103         if (rctx->custom_blend_fastclear) {
 104                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_fastclear);
 105         }
 106         util_unreference_framebuffer_state(&rctx->framebuffer.state);
 107
 108         if (rctx->gs_rings.gsvs_ring.buffer)
 109                 pipe_resource_reference(&rctx->gs_rings.gsvs_ring.buffer, NULL);
 110
 111         if (rctx->gs_rings.esgs_ring.buffer)
 112                 pipe_resource_reference(&rctx->gs_rings.esgs_ring.buffer, NULL);
 113
 114         for (sh = 0; sh < PIPE_SHADER_TYPES; ++sh)
 115                 for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; ++i)
 116                         rctx->b.b.set_constant_buffer(context, sh, i, NULL);
 117
 118         if (rctx->blitter) {
 119                 util_blitter_destroy(rctx->blitter);
 120         }
 121         if (rctx->allocator_fetch_shader) {
 122                 u_suballocator_destroy(rctx->allocator_fetch_shader);
 123         }
 124
 125         r600_release_command_buffer(&rctx->start_cs_cmd);
 126
 127         FREE(rctx->start_compute_cs_cmd.buf);
 128
 129         r600_common_context_cleanup(&rctx->b);
 130
 131         r600_resource_reference(&rctx->trace_buf, NULL);
 132         r600_resource_reference(&rctx->last_trace_buf, NULL);
 133         radeon_clear_saved_cs(&rctx->last_gfx);
 134
 135         FREE(rctx);
 136 }
 137
 138 static struct pipe_context *r600_create_context(struct pipe_screen *screen,
 139                                                 void *priv, unsigned flags)
 140 {
 141         struct r600_context *rctx = CALLOC_STRUCT(r600_context);
 142         struct r600_screen* rscreen = (struct r600_screen *)screen;
 143         struct radeon_winsys *ws = rscreen->b.ws;
 144
 145         if (!rctx)
 146                 return NULL;
 147
 148         rctx->b.b.screen = screen;
 149         assert(!priv);
 150         rctx->b.b.priv = NULL; /* for threaded_context_unwrap_sync */
 151         rctx->b.b.destroy = r600_destroy_context;
 152         rctx->b.set_atom_dirty = (void *)r600_set_atom_dirty;
 153
 154         if (!r600_common_context_init(&rctx->b, &rscreen->b, flags))
 155                 goto fail;
 156
 157         rctx->screen = rscreen;
 158         list_inithead(&rctx->texture_buffers);
 159
 160         r600_init_blit_functions(rctx);
 161
 162         if (rscreen->b.info.has_hw_decode) {
 163                 rctx->b.b.create_video_codec = r600_uvd_create_decoder;
 164                 rctx->b.b.create_video_buffer = r600_video_buffer_create;
 165         } else {
 166                 rctx->b.b.create_video_codec = vl_create_decoder;
 167                 rctx->b.b.create_video_buffer = vl_video_buffer_create;
 168         }
 169
 170         if (getenv("R600_TRACE"))
 171                 rctx->is_debug = true;
 172         r600_init_common_state_functions(rctx);
 173
 174         switch (rctx->b.chip_class) {
 175         case R600:
 176         case R700:
 177                 r600_init_state_functions(rctx);
 178                 r600_init_atom_start_cs(rctx);
 179                 rctx->custom_dsa_flush = r600_create_db_flush_dsa(rctx);
 180                 rctx->custom_blend_resolve = rctx->b.chip_class == R700 ? r700_create_resolve_blend(rctx)
 181                                                                       : r600_create_resolve_blend(rctx);
 182                 rctx->custom_blend_decompress = r600_create_decompress_blend(rctx);
 183                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_RV610 ||
 184                                            rctx->b.family == CHIP_RV620 ||
 185                                            rctx->b.family == CHIP_RS780 ||
 186                                            rctx->b.family == CHIP_RS880 ||
 187                                            rctx->b.family == CHIP_RV710);
 188                 break;
 189         case EVERGREEN:
 190         case CAYMAN:
 191                 evergreen_init_state_functions(rctx);
 192                 evergreen_init_atom_start_cs(rctx);
 193                 evergreen_init_atom_start_compute_cs(rctx);
 194                 rctx->custom_dsa_flush = evergreen_create_db_flush_dsa(rctx);
 195                 rctx->custom_blend_resolve = evergreen_create_resolve_blend(rctx);
 196                 rctx->custom_blend_decompress = evergreen_create_decompress_blend(rctx);
 197                 rctx->custom_blend_fastclear = evergreen_create_fastclear_blend(rctx);
 198                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_CEDAR ||
 199                                            rctx->b.family == CHIP_PALM ||
 200                                            rctx->b.family == CHIP_SUMO ||
 201                                            rctx->b.family == CHIP_SUMO2 ||
 202                                            rctx->b.family == CHIP_CAICOS ||
 203                                            rctx->b.family == CHIP_CAYMAN ||
 204                                            rctx->b.family == CHIP_ARUBA);
 205
 206                 rctx->append_fence = pipe_buffer_create(rctx->b.b.screen, PIPE_BIND_CUSTOM,
 207                                                          PIPE_USAGE_DEFAULT, 32);
 208                 break;
 209         default:
 210                 R600_ERR("Unsupported chip class %d.\n", rctx->b.chip_class);
 211                 goto fail;
 212         }
 213
 214         rctx->b.gfx.cs = ws->cs_create(rctx->b.ctx, RING_GFX,
 215                                        r600_context_gfx_flush, rctx, false);
 216         rctx->b.gfx.flush = r600_context_gfx_flush;
 217
 218         rctx->allocator_fetch_shader =
 219                 u_suballocator_create(&rctx->b.b, 64 * 1024,
 220                                       0, PIPE_USAGE_DEFAULT, 0, FALSE);
 221         if (!rctx->allocator_fetch_shader)
 222                 goto fail;
 223
 224         rctx->isa = calloc(1, sizeof(struct r600_isa));
 225         if (!rctx->isa || r600_isa_init(rctx, rctx->isa))
 226                 goto fail;
 227
 228         if (rscreen->b.debug_flags & DBG_FORCE_DMA)
 229                 rctx->b.b.resource_copy_region = rctx->b.dma_copy;
 230
 231         rctx->blitter = util_blitter_create(&rctx->b.b);
 232         if (rctx->blitter == NULL)
 233                 goto fail;
 234         util_blitter_set_texture_multisample(rctx->blitter, rscreen->has_msaa);
 235         rctx->blitter->draw_rectangle = r600_draw_rectangle;
 236
 237         r600_begin_new_cs(rctx);
 238
 239         rctx->dummy_pixel_shader =
 240                 util_make_fragment_cloneinput_shader(&rctx->b.b, 0,
 241                                                      TGSI_SEMANTIC_GENERIC,
 242                                                      TGSI_INTERPOLATE_CONSTANT);
 243         rctx->b.b.bind_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
 244
 245         return &rctx->b.b;
 246
 247 fail:
 248         r600_destroy_context(&rctx->b.b);
 249         return NULL;
 250 }
 251
 252 static bool is_nir_enabled(struct r600_common_screen *screen) {
 253    return (screen->debug_flags & DBG_NIR &&
 254        screen->family >= CHIP_CEDAR &&
 255        screen->family < CHIP_CAYMAN);
 256 }
 257
 258 /*
 259  * pipe_screen
 260  */
 261
 262 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 263 {
 264         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 265         enum radeon_family family = rscreen->b.family;
 266
 267         switch (param) {
 268         /* Supported features (boolean caps). */
 269         case PIPE_CAP_NPOT_TEXTURES:
 270         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 271         case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
 272         case PIPE_CAP_ANISOTROPIC_FILTER:
 273         case PIPE_CAP_POINT_SPRITE:
 274         case PIPE_CAP_OCCLUSION_QUERY:
 275         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 276         case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
 277         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 278         case PIPE_CAP_TEXTURE_SWIZZLE:
 279         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 280         case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
 281         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 282         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 283         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 284         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 285         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 286         case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD:
 287         case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES:
 288         case PIPE_CAP_VERTEX_SHADER_SATURATE:
 289         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 290         case PIPE_CAP_PRIMITIVE_RESTART:
 291         case PIPE_CAP_PRIMITIVE_RESTART_FIXED_INDEX:
 292         case PIPE_CAP_CONDITIONAL_RENDER:
 293         case PIPE_CAP_TEXTURE_BARRIER:
 294         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 295         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 296         case PIPE_CAP_TGSI_INSTANCEID:
 297         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 298         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 299         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 300         case PIPE_CAP_START_INSTANCE:
 301         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 302         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 303         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 304         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 305         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 306         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 307         case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 308         case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 309         case PIPE_CAP_SAMPLE_SHADING:
 310         case PIPE_CAP_CLIP_HALFZ:
 311         case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 312         case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 313         case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 314         case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 315         case PIPE_CAP_TGSI_TXQS:
 316         case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 317         case PIPE_CAP_INVALIDATE_BUFFER:
 318         case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 319         case PIPE_CAP_QUERY_MEMORY_INFO:
 320         case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
 321         case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
 322         case PIPE_CAP_CLEAR_TEXTURE:
 323         case PIPE_CAP_TGSI_MUL_ZERO_WINS:
 324         case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
 325         case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
 326         case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
 327         case PIPE_CAP_NIR_ATOMICS_AS_DEREF:
 328                 return 1;
 329
 330         case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
 331                 /* Optimal number for good TexSubImage performance on Polaris10. */
 332                 return 64 * 1024 * 1024;
 333
 334         case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 335                 return rscreen->b.info.drm_minor >= 43;
 336
 337         case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 338                 return !R600_BIG_ENDIAN && rscreen->b.info.has_userptr;
 339
 340         case PIPE_CAP_COMPUTE:
 341                 return rscreen->b.chip_class > R700;
 342
 343         case PIPE_CAP_TGSI_TEXCOORD:
 344                 return 1;
 345
 346         case PIPE_CAP_NIR_IMAGES_AS_DEREF:
 347         case PIPE_CAP_FAKE_SW_MSAA:
 348                 return 0;
 349
 350         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 351                 return MIN2(rscreen->b.info.max_alloc_size, INT_MAX);
 352
 353         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 354                 return R600_MAP_BUFFER_ALIGNMENT;
 355
 356         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 357                 return 256;
 358
 359         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 360                 return 1;
 361
 362         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 363                 if (family >= CHIP_CEDAR)
 364                    return 430;
 365                 /* pre-evergreen geom shaders need newer kernel */
 366                 if (rscreen->b.info.drm_minor >= 37)
 367                    return 330;
 368                 return 140;
 369
 370         case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
 371                 return 140;
 372
 373         /* Supported except the original R600. */
 374         case PIPE_CAP_INDEP_BLEND_ENABLE:
 375         case PIPE_CAP_INDEP_BLEND_FUNC:
 376                 /* R600 doesn't support per-MRT blends */
 377                 return family == CHIP_R600 ? 0 : 1;
 378
 379         /* Supported on Evergreen. */
 380         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 381         case PIPE_CAP_CUBE_MAP_ARRAY:
 382         case PIPE_CAP_TEXTURE_GATHER_SM5:
 383         case PIPE_CAP_TEXTURE_QUERY_LOD:
 384         case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 385         case PIPE_CAP_SAMPLER_VIEW_TARGET:
 386         case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 387         case PIPE_CAP_TGSI_CLOCK:
 388         case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
 389         case PIPE_CAP_QUERY_BUFFER_OBJECT:
 390                 return family >= CHIP_CEDAR ? 1 : 0;
 391         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 392                 return family >= CHIP_CEDAR ? 4 : 0;
 393         case PIPE_CAP_DRAW_INDIRECT:
 394                 /* kernel command checker support is also required */
 395                 return family >= CHIP_CEDAR && rscreen->b.info.drm_minor >= 41;
 396
 397         case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 398                 return family >= CHIP_CEDAR ? 0 : 1;
 399
 400         case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
 401                 return 8;
 402
 403         case PIPE_CAP_MAX_GS_INVOCATIONS:
 404                 return 32;
 405
 406         /* shader buffer objects */
 407         case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
 408                 return 1 << 27;
 409         case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS:
 410                 return 8;
 411
 412         case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
 413                 return 0;
 414
 415         case PIPE_CAP_DOUBLES:
 416                 if (rscreen->b.family == CHIP_ARUBA ||
 417                     rscreen->b.family == CHIP_CAYMAN ||
 418                     rscreen->b.family == CHIP_CYPRESS ||
 419                     rscreen->b.family == CHIP_HEMLOCK)
 420                         return 1;
 421                 return 0;
 422         case PIPE_CAP_CULL_DISTANCE:
 423                 return 1;
 424
 425         case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 426                 if (family >= CHIP_CEDAR)
 427                         return 256;
 428                 return 0;
 429
 430         case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 431                 if (family >= CHIP_CEDAR)
 432                         return 30;
 433                 else
 434                         return 0;
 435         /* Stream output. */
 436         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 437                 return rscreen->b.has_streamout ? 4 : 0;
 438         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 439         case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
 440                 return rscreen->b.has_streamout ? 1 : 0;
 441         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 442         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 443                 return 32*4;
 444
 445         /* Geometry shader output. */
 446         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 447                 return 1024;
 448         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 449                 return 16384;
 450         case PIPE_CAP_MAX_VERTEX_STREAMS:
 451                 return family >= CHIP_CEDAR ? 4 : 1;
 452
 453         case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 454                 /* Should be 2047, but 2048 is a requirement for GL 4.4 */
 455                 return 2048;
 456
 457         /* Texturing. */
 458         case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
 459                 if (family >= CHIP_CEDAR)
 460                         return 16384;
 461                 else
 462                         return 8192;
 463         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 464                 if (family >= CHIP_CEDAR)
 465                         return 15;
 466                 else
 467                         return 14;
 468         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 469                 /* textures support 8192, but layered rendering supports 2048 */
 470                 return 12;
 471         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 472                 /* textures support 8192, but layered rendering supports 2048 */
 473                 return 2048;
 474
 475         /* Render targets. */
 476         case PIPE_CAP_MAX_RENDER_TARGETS:
 477                 /* XXX some r6xx are buggy and can only do 4 */
 478                 return 8;
 479
 480         case PIPE_CAP_MAX_VIEWPORTS:
 481                 return R600_MAX_VIEWPORTS;
 482         case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
 483         case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
 484                 return 8;
 485
 486         /* Timer queries, present when the clock frequency is non zero. */
 487         case PIPE_CAP_QUERY_TIME_ELAPSED:
 488                 return rscreen->b.info.clock_crystal_freq != 0;
 489         case PIPE_CAP_QUERY_TIMESTAMP:
 490                 return rscreen->b.info.drm_minor >= 20 &&
 491                        rscreen->b.info.clock_crystal_freq != 0;
 492
 493         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 494         case PIPE_CAP_MIN_TEXEL_OFFSET:
 495                 return -8;
 496
 497         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 498         case PIPE_CAP_MAX_TEXEL_OFFSET:
 499                 return 7;
 500
 501         case PIPE_CAP_MAX_VARYINGS:
 502                 return 32;
 503
 504         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 505                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 506         case PIPE_CAP_ENDIANNESS:
 507                 return PIPE_ENDIAN_LITTLE;
 508
 509         case PIPE_CAP_VENDOR_ID:
 510                 return ATI_VENDOR_ID;
 511         case PIPE_CAP_DEVICE_ID:
 512                 return rscreen->b.info.pci_id;
 513         case PIPE_CAP_ACCELERATED:
 514                 return 1;
 515         case PIPE_CAP_VIDEO_MEMORY:
 516                 return rscreen->b.info.vram_size >> 20;
 517         case PIPE_CAP_UMA:
 518                 return 0;
 519         case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 520                 return rscreen->b.chip_class >= R700;
 521         case PIPE_CAP_PCI_GROUP:
 522                 return rscreen->b.info.pci_domain;
 523         case PIPE_CAP_PCI_BUS:
 524                 return rscreen->b.info.pci_bus;
 525         case PIPE_CAP_PCI_DEVICE:
 526                 return rscreen->b.info.pci_dev;
 527         case PIPE_CAP_PCI_FUNCTION:
 528                 return rscreen->b.info.pci_func;
 529
 530         case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS:
 531                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics)
 532                         return 8;
 533                 return 0;
 534         case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS:
 535                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics)
 536                         return EG_MAX_ATOMIC_BUFFERS;
 537                 return 0;
 538
 539         default:
 540                 return u_pipe_screen_get_param_defaults(pscreen, param);
 541         }
 542 }
 543
 544 static int r600_get_shader_param(struct pipe_screen* pscreen,
 545                                  enum pipe_shader_type shader,
 546                                  enum pipe_shader_cap param)
 547 {
 548         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 549
 550         switch(shader)
 551         {
 552         case PIPE_SHADER_FRAGMENT:
 553         case PIPE_SHADER_VERTEX:
 554                 break;
 555         case PIPE_SHADER_GEOMETRY:
 556                 if (rscreen->b.family >= CHIP_CEDAR)
 557                         break;
 558                 /* pre-evergreen geom shaders need newer kernel */
 559                 if (rscreen->b.info.drm_minor >= 37)
 560                         break;
 561                 return 0;
 562         case PIPE_SHADER_TESS_CTRL:
 563         case PIPE_SHADER_TESS_EVAL:
 564         case PIPE_SHADER_COMPUTE:
 565                 if (rscreen->b.family >= CHIP_CEDAR)
 566                         break;
 567                 /* fallthrough */
 568         default:
 569                 return 0;
 570         }
 571
 572         switch (param) {
 573         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 574         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 575         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 576         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 577                 return 16384;
 578         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 579                 return 32;
 580         case PIPE_SHADER_CAP_MAX_INPUTS:
 581                 return shader == PIPE_SHADER_VERTEX ? 16 : 32;
 582         case PIPE_SHADER_CAP_MAX_OUTPUTS:
 583                 return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
 584         case PIPE_SHADER_CAP_MAX_TEMPS:
 585                 return 256; /* Max native temporaries. */
 586         case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 587                 if (shader == PIPE_SHADER_COMPUTE) {
 588                         uint64_t max_const_buffer_size;
 589                         enum pipe_shader_ir ir_type = is_nir_enabled(&rscreen->b) ?
 590                                 PIPE_SHADER_IR_NIR: PIPE_SHADER_IR_TGSI;
 591                         pscreen->get_compute_param(pscreen, ir_type,
 592                                                    PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
 593                                                    &max_const_buffer_size);
 594                         return MIN2(max_const_buffer_size, INT_MAX);
 595
 596                 } else {
 597                         return R600_MAX_CONST_BUFFER_SIZE;
 598                 }
 599         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 600                 return R600_MAX_USER_CONST_BUFFERS;
 601         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 602                 return 1;
 603         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 604                 return 1;
 605         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 606         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 607         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 608         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 609                 return 1;
 610         case PIPE_SHADER_CAP_SUBROUTINES:
 611         case PIPE_SHADER_CAP_INT64_ATOMICS:
 612         case PIPE_SHADER_CAP_FP16:
 613         case PIPE_SHADER_CAP_FP16_DERIVATIVES:
 614         case PIPE_SHADER_CAP_INT16:
 615         case PIPE_SHADER_CAP_GLSL_16BIT_CONSTS:
 616                 return 0;
 617         case PIPE_SHADER_CAP_INTEGERS:
 618         case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 619                 return 1;
 620         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 621         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 622                 return 16;
 623         case PIPE_SHADER_CAP_PREFERRED_IR:
 624                 if (is_nir_enabled(&rscreen->b))
 625                         return PIPE_SHADER_IR_NIR;
 626                 return PIPE_SHADER_IR_TGSI;
 627         case PIPE_SHADER_CAP_SUPPORTED_IRS: {
 628                 int ir = 0;
 629                 if (shader == PIPE_SHADER_COMPUTE)
 630                         ir = 1 << PIPE_SHADER_IR_NATIVE;
 631                 if (rscreen->b.family >= CHIP_CEDAR) {
 632                         ir |= 1 << PIPE_SHADER_IR_TGSI;
 633                         if (is_nir_enabled(&rscreen->b))
 634                                 ir |= 1 << PIPE_SHADER_IR_NIR;
 635                 }
 636                 return ir;
 637         }
 638         case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 639                 if (rscreen->b.family == CHIP_ARUBA ||
 640                     rscreen->b.family == CHIP_CAYMAN ||
 641                     rscreen->b.family == CHIP_CYPRESS ||
 642                     rscreen->b.family == CHIP_HEMLOCK)
 643                         return 1;
 644                 return 0;
 645         case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 646         case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 647         case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
 648         case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
 649         case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
 650                 return 0;
 651         case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 652         case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
 653                 if (rscreen->b.family >= CHIP_CEDAR &&
 654                     (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE))
 655                     return 8;
 656                 return 0;
 657         case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
 658                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics)
 659                         return 8;
 660                 return 0;
 661         case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
 662                 /* having to allocate the atomics out amongst shaders stages is messy,
 663                    so give compute 8 buffers and all the others one */
 664                 if (rscreen->b.family >= CHIP_CEDAR && rscreen->has_atomics) {
 665                         return EG_MAX_ATOMIC_BUFFERS;
 666                 }
 667                 return 0;
 668         case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 669                 /* due to a bug in the shader compiler, some loops hang
 670                  * if they are not unrolled, see:
 671                  *    https://bugs.freedesktop.org/show_bug.cgi?id=86720
 672                  */
 673                 return 255;
 674         }
 675         return 0;
 676 }
 677
 678 static void r600_destroy_screen(struct pipe_screen* pscreen)
 679 {
 680         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 681
 682         if (!rscreen)
 683                 return;
 684
 685         if (!rscreen->b.ws->unref(rscreen->b.ws))
 686                 return;
 687
 688         if (rscreen->global_pool) {
 689                 compute_memory_pool_delete(rscreen->global_pool);
 690         }
 691
 692         r600_destroy_common_screen(&rscreen->b);
 693 }
 694
 695 static struct pipe_resource *r600_resource_create(struct pipe_screen *screen,
 696                                                   const struct pipe_resource *templ)
 697 {
 698         if (templ->target == PIPE_BUFFER &&
 699             (templ->bind & PIPE_BIND_GLOBAL))
 700                 return r600_compute_global_buffer_create(screen, templ);
 701
 702         return r600_resource_create_common(screen, templ);
 703 }
 704
 705 struct pipe_screen *r600_screen_create(struct radeon_winsys *ws,
 706                                        const struct pipe_screen_config *config)
 707 {
 708         struct r600_screen *rscreen = CALLOC_STRUCT(r600_screen);
 709
 710         if (!rscreen) {
 711                 return NULL;
 712         }
 713
 714         /* Set functions first. */
 715         rscreen->b.b.context_create = r600_create_context;
 716         rscreen->b.b.destroy = r600_destroy_screen;
 717         rscreen->b.b.get_param = r600_get_param;
 718         rscreen->b.b.get_shader_param = r600_get_shader_param;
 719         rscreen->b.b.resource_create = r600_resource_create;
 720
 721         if (!r600_common_screen_init(&rscreen->b, ws)) {
 722                 FREE(rscreen);
 723                 return NULL;
 724         }
 725
 726         if (rscreen->b.info.chip_class >= EVERGREEN) {
 727                 rscreen->b.b.is_format_supported = evergreen_is_format_supported;
 728         } else {
 729                 rscreen->b.b.is_format_supported = r600_is_format_supported;
 730         }
 731
 732         rscreen->b.debug_flags |= debug_get_flags_option("R600_DEBUG", r600_debug_options, 0);
 733         if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE))
 734                 rscreen->b.debug_flags |= DBG_COMPUTE;
 735         if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE))
 736                 rscreen->b.debug_flags |= DBG_ALL_SHADERS | DBG_FS;
 737         if (!debug_get_bool_option("R600_HYPERZ", TRUE))
 738                 rscreen->b.debug_flags |= DBG_NO_HYPERZ;
 739
 740         if (rscreen->b.family == CHIP_UNKNOWN) {
 741                 fprintf(stderr, "r600: Unknown chipset 0x%04X\n", rscreen->b.info.pci_id);
 742                 FREE(rscreen);
 743                 return NULL;
 744         }
 745
 746         /* Figure out streamout kernel support. */
 747         switch (rscreen->b.chip_class) {
 748         case R600:
 749                 if (rscreen->b.family < CHIP_RS780) {
 750                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 751                 } else {
 752                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 23;
 753                 }
 754                 break;
 755         case R700:
 756                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 17;
 757                 break;
 758         case EVERGREEN:
 759         case CAYMAN:
 760                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 761                 break;
 762         default:
 763                 rscreen->b.has_streamout = FALSE;
 764                 break;
 765         }
 766
 767         /* MSAA support. */
 768         switch (rscreen->b.chip_class) {
 769         case R600:
 770         case R700:
 771                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 22;
 772                 rscreen->has_compressed_msaa_texturing = false;
 773                 break;
 774         case EVERGREEN:
 775                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 776                 rscreen->has_compressed_msaa_texturing = rscreen->b.info.drm_minor >= 24;
 777                 break;
 778         case CAYMAN:
 779                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 780                 rscreen->has_compressed_msaa_texturing = true;
 781                 break;
 782         default:
 783                 rscreen->has_msaa = FALSE;
 784                 rscreen->has_compressed_msaa_texturing = false;
 785         }
 786
 787         rscreen->b.has_cp_dma = rscreen->b.info.drm_minor >= 27 &&
 788                               !(rscreen->b.debug_flags & DBG_NO_CP_DMA);
 789
 790         rscreen->b.barrier_flags.cp_to_L2 =
 791                 R600_CONTEXT_INV_VERTEX_CACHE |
 792                 R600_CONTEXT_INV_TEX_CACHE |
 793                 R600_CONTEXT_INV_CONST_CACHE;
 794         rscreen->b.barrier_flags.compute_to_L2 = R600_CONTEXT_CS_PARTIAL_FLUSH | R600_CONTEXT_FLUSH_AND_INV;
 795
 796         rscreen->global_pool = compute_memory_pool_new(rscreen);
 797
 798         /* Create the auxiliary context. This must be done last. */
 799         rscreen->b.aux_context = rscreen->b.b.context_create(&rscreen->b.b, NULL, 0);
 800
 801         rscreen->has_atomics = rscreen->b.info.drm_minor >= 44;
 802 #if 0 /* This is for testing whether aux_context and buffer clearing work correctly. */
 803         struct pipe_resource templ = {};
 804
 805         templ.width0 = 4;
 806         templ.height0 = 2048;
 807         templ.depth0 = 1;
 808         templ.array_size = 1;
 809         templ.target = PIPE_TEXTURE_2D;
 810         templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 811         templ.usage = PIPE_USAGE_DEFAULT;
 812
 813         struct r600_resource *res = r600_resource(rscreen->screen.resource_create(&rscreen->screen, &templ));
 814         unsigned char *map = ws->buffer_map(res->buf, NULL, PIPE_TRANSFER_WRITE);
 815
 816         memset(map, 0, 256);
 817
 818         r600_screen_clear_buffer(rscreen, &res->b.b, 4, 4, 0xCC);
 819         r600_screen_clear_buffer(rscreen, &res->b.b, 8, 4, 0xDD);
 820         r600_screen_clear_buffer(rscreen, &res->b.b, 12, 4, 0xEE);
 821         r600_screen_clear_buffer(rscreen, &res->b.b, 20, 4, 0xFF);
 822         r600_screen_clear_buffer(rscreen, &res->b.b, 32, 20, 0x87);
 823
 824         ws->buffer_wait(res->buf, RADEON_USAGE_WRITE);
 825
 826         int i;
 827         for (i = 0; i < 256; i++) {
 828                 printf("%02X", map[i]);
 829                 if (i % 16 == 15)
 830                         printf("\n");
 831         }
 832 #endif
 833
 834         if (rscreen->b.debug_flags & DBG_TEST_DMA)
 835                 r600_test_dma(&rscreen->b);
 836
 837         r600_query_fix_enabled_rb_mask(&rscreen->b);
 838         return &rscreen->b.b;
 839 }