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 #if defined(R600_USE_LLVM)
  47         { "llvm", DBG_LLVM, "Enable the LLVM shader compiler" },
  48 #endif
  49         { "nocpdma", DBG_NO_CP_DMA, "Disable CP DMA" },
  50
  51         /* shader backend */
  52         { "nosb", DBG_NO_SB, "Disable sb backend for graphics shaders" },
  53         { "sbcl", DBG_SB_CS, "Enable sb backend for compute shaders" },
  54         { "sbdry", DBG_SB_DRY_RUN, "Don't use optimized bytecode (just print the dumps)" },
  55         { "sbstat", DBG_SB_STAT, "Print optimization statistics for shaders" },
  56         { "sbdump", DBG_SB_DUMP, "Print IR dumps after some optimization passes" },
  57         { "sbnofallback", DBG_SB_NO_FALLBACK, "Abort on errors instead of fallback" },
  58         { "sbdisasm", DBG_SB_DISASM, "Use sb disassembler for shader dumps" },
  59         { "sbsafemath", DBG_SB_SAFEMATH, "Disable unsafe math optimizations" },
  60
  61         DEBUG_NAMED_VALUE_END /* must be last */
  62 };
  63
  64 /*
  65  * pipe_context
  66  */
  67
  68 static void r600_destroy_context(struct pipe_context *context)
  69 {
  70         struct r600_context *rctx = (struct r600_context *)context;
  71
  72         r600_isa_destroy(rctx->isa);
  73
  74         r600_sb_context_destroy(rctx->sb_context);
  75
  76         pipe_resource_reference((struct pipe_resource**)&rctx->dummy_cmask, NULL);
  77         pipe_resource_reference((struct pipe_resource**)&rctx->dummy_fmask, NULL);
  78
  79         if (rctx->dummy_pixel_shader) {
  80                 rctx->b.b.delete_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
  81         }
  82         if (rctx->custom_dsa_flush) {
  83                 rctx->b.b.delete_depth_stencil_alpha_state(&rctx->b.b, rctx->custom_dsa_flush);
  84         }
  85         if (rctx->custom_blend_resolve) {
  86                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_resolve);
  87         }
  88         if (rctx->custom_blend_decompress) {
  89                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_decompress);
  90         }
  91         if (rctx->custom_blend_fastclear) {
  92                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_fastclear);
  93         }
  94         util_unreference_framebuffer_state(&rctx->framebuffer.state);
  95
  96         if (rctx->blitter) {
  97                 util_blitter_destroy(rctx->blitter);
  98         }
  99         if (rctx->allocator_fetch_shader) {
 100                 u_suballocator_destroy(rctx->allocator_fetch_shader);
 101         }
 102
 103         r600_release_command_buffer(&rctx->start_cs_cmd);
 104
 105         FREE(rctx->start_compute_cs_cmd.buf);
 106
 107         r600_common_context_cleanup(&rctx->b);
 108         FREE(rctx);
 109 }
 110
 111 static struct pipe_context *r600_create_context(struct pipe_screen *screen, void *priv)
 112 {
 113         struct r600_context *rctx = CALLOC_STRUCT(r600_context);
 114         struct r600_screen* rscreen = (struct r600_screen *)screen;
 115         struct radeon_winsys *ws = rscreen->b.ws;
 116
 117         if (rctx == NULL)
 118                 return NULL;
 119
 120         rctx->b.b.screen = screen;
 121         rctx->b.b.priv = priv;
 122         rctx->b.b.destroy = r600_destroy_context;
 123
 124         if (!r600_common_context_init(&rctx->b, &rscreen->b))
 125                 goto fail;
 126
 127         rctx->screen = rscreen;
 128         rctx->keep_tiling_flags = rscreen->b.info.drm_minor >= 12;
 129
 130         r600_init_blit_functions(rctx);
 131
 132         if (rscreen->b.info.has_uvd) {
 133                 rctx->b.b.create_video_codec = r600_uvd_create_decoder;
 134                 rctx->b.b.create_video_buffer = r600_video_buffer_create;
 135         } else {
 136                 rctx->b.b.create_video_codec = vl_create_decoder;
 137                 rctx->b.b.create_video_buffer = vl_video_buffer_create;
 138         }
 139
 140         r600_init_common_state_functions(rctx);
 141
 142         switch (rctx->b.chip_class) {
 143         case R600:
 144         case R700:
 145                 r600_init_state_functions(rctx);
 146                 r600_init_atom_start_cs(rctx);
 147                 rctx->custom_dsa_flush = r600_create_db_flush_dsa(rctx);
 148                 rctx->custom_blend_resolve = rctx->b.chip_class == R700 ? r700_create_resolve_blend(rctx)
 149                                                                       : r600_create_resolve_blend(rctx);
 150                 rctx->custom_blend_decompress = r600_create_decompress_blend(rctx);
 151                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_RV610 ||
 152                                            rctx->b.family == CHIP_RV620 ||
 153                                            rctx->b.family == CHIP_RS780 ||
 154                                            rctx->b.family == CHIP_RS880 ||
 155                                            rctx->b.family == CHIP_RV710);
 156                 break;
 157         case EVERGREEN:
 158         case CAYMAN:
 159                 evergreen_init_state_functions(rctx);
 160                 evergreen_init_atom_start_cs(rctx);
 161                 evergreen_init_atom_start_compute_cs(rctx);
 162                 rctx->custom_dsa_flush = evergreen_create_db_flush_dsa(rctx);
 163                 rctx->custom_blend_resolve = evergreen_create_resolve_blend(rctx);
 164                 rctx->custom_blend_decompress = evergreen_create_decompress_blend(rctx);
 165                 rctx->custom_blend_fastclear = evergreen_create_fastclear_blend(rctx);
 166                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_CEDAR ||
 167                                            rctx->b.family == CHIP_PALM ||
 168                                            rctx->b.family == CHIP_SUMO ||
 169                                            rctx->b.family == CHIP_SUMO2 ||
 170                                            rctx->b.family == CHIP_CAICOS ||
 171                                            rctx->b.family == CHIP_CAYMAN ||
 172                                            rctx->b.family == CHIP_ARUBA);
 173                 break;
 174         default:
 175                 R600_ERR("Unsupported chip class %d.\n", rctx->b.chip_class);
 176                 goto fail;
 177         }
 178
 179         rctx->b.rings.gfx.cs = ws->cs_create(ws, RING_GFX,
 180                                              r600_context_gfx_flush, rctx,
 181                                              rscreen->b.trace_bo ?
 182                                                      rscreen->b.trace_bo->cs_buf : NULL);
 183         rctx->b.rings.gfx.flush = r600_context_gfx_flush;
 184
 185         rctx->allocator_fetch_shader = u_suballocator_create(&rctx->b.b, 64 * 1024, 256,
 186                                                              0, PIPE_USAGE_DEFAULT, FALSE);
 187         if (!rctx->allocator_fetch_shader)
 188                 goto fail;
 189
 190         rctx->isa = calloc(1, sizeof(struct r600_isa));
 191         if (!rctx->isa || r600_isa_init(rctx, rctx->isa))
 192                 goto fail;
 193
 194         if (rscreen->b.debug_flags & DBG_FORCE_DMA)
 195                 rctx->b.b.resource_copy_region = rctx->b.dma_copy;
 196
 197         rctx->blitter = util_blitter_create(&rctx->b.b);
 198         if (rctx->blitter == NULL)
 199                 goto fail;
 200         util_blitter_set_texture_multisample(rctx->blitter, rscreen->has_msaa);
 201         rctx->blitter->draw_rectangle = r600_draw_rectangle;
 202
 203         r600_begin_new_cs(rctx);
 204         r600_query_init_backend_mask(&rctx->b); /* this emits commands and must be last */
 205
 206         rctx->dummy_pixel_shader =
 207                 util_make_fragment_cloneinput_shader(&rctx->b.b, 0,
 208                                                      TGSI_SEMANTIC_GENERIC,
 209                                                      TGSI_INTERPOLATE_CONSTANT);
 210         rctx->b.b.bind_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
 211
 212         return &rctx->b.b;
 213
 214 fail:
 215         r600_destroy_context(&rctx->b.b);
 216         return NULL;
 217 }
 218
 219 /*
 220  * pipe_screen
 221  */
 222
 223 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 224 {
 225         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 226         enum radeon_family family = rscreen->b.family;
 227
 228         switch (param) {
 229         /* Supported features (boolean caps). */
 230         case PIPE_CAP_NPOT_TEXTURES:
 231         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 232         case PIPE_CAP_TWO_SIDED_STENCIL:
 233         case PIPE_CAP_ANISOTROPIC_FILTER:
 234         case PIPE_CAP_POINT_SPRITE:
 235         case PIPE_CAP_OCCLUSION_QUERY:
 236         case PIPE_CAP_TEXTURE_SHADOW_MAP:
 237         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 238         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 239         case PIPE_CAP_TEXTURE_SWIZZLE:
 240         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 241         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 242         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 243         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 244         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 245         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 246         case PIPE_CAP_SM3:
 247         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 248         case PIPE_CAP_PRIMITIVE_RESTART:
 249         case PIPE_CAP_CONDITIONAL_RENDER:
 250         case PIPE_CAP_TEXTURE_BARRIER:
 251         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 252         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 253         case PIPE_CAP_TGSI_INSTANCEID:
 254         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 255         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 256         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 257         case PIPE_CAP_USER_INDEX_BUFFERS:
 258         case PIPE_CAP_USER_CONSTANT_BUFFERS:
 259         case PIPE_CAP_START_INSTANCE:
 260         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 261         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 262         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 263         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 264         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 265         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 266         case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 267         case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 268         case PIPE_CAP_SAMPLE_SHADING:
 269         case PIPE_CAP_CLIP_HALFZ:
 270         case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 271                 return 1;
 272
 273         case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 274                 return !R600_BIG_ENDIAN && rscreen->b.info.has_userptr;
 275
 276         case PIPE_CAP_COMPUTE:
 277                 return rscreen->b.chip_class > R700;
 278
 279         case PIPE_CAP_TGSI_TEXCOORD:
 280                 return 0;
 281
 282         case PIPE_CAP_FAKE_SW_MSAA:
 283                 return 0;
 284
 285         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 286                 return MIN2(rscreen->b.info.vram_size, 0xFFFFFFFF);
 287
 288         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 289                 return R600_MAP_BUFFER_ALIGNMENT;
 290
 291         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 292                 return 256;
 293
 294         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 295                 return 1;
 296
 297         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 298                 if (family >= CHIP_CEDAR)
 299                    return 330;
 300                 /* pre-evergreen geom shaders need newer kernel */
 301                 if (rscreen->b.info.drm_minor >= 37)
 302                    return 330;
 303                 return 140;
 304
 305         /* Supported except the original R600. */
 306         case PIPE_CAP_INDEP_BLEND_ENABLE:
 307         case PIPE_CAP_INDEP_BLEND_FUNC:
 308                 /* R600 doesn't support per-MRT blends */
 309                 return family == CHIP_R600 ? 0 : 1;
 310
 311         /* Supported on Evergreen. */
 312         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 313         case PIPE_CAP_CUBE_MAP_ARRAY:
 314         case PIPE_CAP_TEXTURE_GATHER_SM5:
 315         case PIPE_CAP_TEXTURE_QUERY_LOD:
 316         case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 317                 return family >= CHIP_CEDAR ? 1 : 0;
 318         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 319                 return family >= CHIP_CEDAR ? 4 : 0;
 320         case PIPE_CAP_DRAW_INDIRECT:
 321                 /* kernel command checker support is also required */
 322                 return family >= CHIP_CEDAR && rscreen->b.info.drm_minor >= 41;
 323
 324         /* Unsupported features. */
 325         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 326         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 327         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 328         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 329         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 330         case PIPE_CAP_USER_VERTEX_BUFFERS:
 331         case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 332         case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 333         case PIPE_CAP_SAMPLER_VIEW_TARGET:
 334         case PIPE_CAP_VERTEXID_NOBASE:
 335                 return 0;
 336
 337         /* Stream output. */
 338         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 339                 return rscreen->b.has_streamout ? 4 : 0;
 340         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 341                 return rscreen->b.has_streamout ? 1 : 0;
 342         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 343         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 344                 return 32*4;
 345
 346         /* Geometry shader output. */
 347         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 348                 return 1024;
 349         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 350                 return 16384;
 351         case PIPE_CAP_MAX_VERTEX_STREAMS:
 352                 return 1;
 353
 354         case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 355                 return 2047;
 356
 357         /* Texturing. */
 358         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 359         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 360                 if (family >= CHIP_CEDAR)
 361                         return 15;
 362                 else
 363                         return 14;
 364         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 365                 /* textures support 8192, but layered rendering supports 2048 */
 366                 return 12;
 367         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 368                 /* textures support 8192, but layered rendering supports 2048 */
 369                 return rscreen->b.info.drm_minor >= 9 ? 2048 : 0;
 370
 371         /* Render targets. */
 372         case PIPE_CAP_MAX_RENDER_TARGETS:
 373                 /* XXX some r6xx are buggy and can only do 4 */
 374                 return 8;
 375
 376         case PIPE_CAP_MAX_VIEWPORTS:
 377                 return 16;
 378
 379         /* Timer queries, present when the clock frequency is non zero. */
 380         case PIPE_CAP_QUERY_TIME_ELAPSED:
 381                 return rscreen->b.info.r600_clock_crystal_freq != 0;
 382         case PIPE_CAP_QUERY_TIMESTAMP:
 383                 return rscreen->b.info.drm_minor >= 20 &&
 384                        rscreen->b.info.r600_clock_crystal_freq != 0;
 385
 386         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 387         case PIPE_CAP_MIN_TEXEL_OFFSET:
 388                 return -8;
 389
 390         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 391         case PIPE_CAP_MAX_TEXEL_OFFSET:
 392                 return 7;
 393
 394         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 395                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 396         case PIPE_CAP_ENDIANNESS:
 397                 return PIPE_ENDIAN_LITTLE;
 398
 399         case PIPE_CAP_VENDOR_ID:
 400                 return 0x1002;
 401         case PIPE_CAP_DEVICE_ID:
 402                 return rscreen->b.info.pci_id;
 403         case PIPE_CAP_ACCELERATED:
 404                 return 1;
 405         case PIPE_CAP_VIDEO_MEMORY:
 406                 return rscreen->b.info.vram_size >> 20;
 407         case PIPE_CAP_UMA:
 408                 return 0;
 409         case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 410                 return rscreen->b.chip_class >= R700;
 411         }
 412         return 0;
 413 }
 414
 415 static int r600_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
 416 {
 417         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 418
 419         switch(shader)
 420         {
 421         case PIPE_SHADER_FRAGMENT:
 422         case PIPE_SHADER_VERTEX:
 423         case PIPE_SHADER_COMPUTE:
 424                 break;
 425         case PIPE_SHADER_GEOMETRY:
 426                 if (rscreen->b.family >= CHIP_CEDAR)
 427                         break;
 428                 /* pre-evergreen geom shaders need newer kernel */
 429                 if (rscreen->b.info.drm_minor >= 37)
 430                         break;
 431                 return 0;
 432         default:
 433                 /* XXX: support tessellation on Evergreen */
 434                 return 0;
 435         }
 436
 437         switch (param) {
 438         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 439         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 440         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 441         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 442                 return 16384;
 443         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 444                 return 32;
 445         case PIPE_SHADER_CAP_MAX_INPUTS:
 446                 return shader == PIPE_SHADER_VERTEX ? 16 : 32;
 447         case PIPE_SHADER_CAP_MAX_OUTPUTS:
 448                 return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
 449         case PIPE_SHADER_CAP_MAX_TEMPS:
 450                 return 256; /* Max native temporaries. */
 451         case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 452                 if (shader == PIPE_SHADER_COMPUTE) {
 453                         uint64_t max_const_buffer_size;
 454                         pscreen->get_compute_param(pscreen,
 455                                 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
 456                                 &max_const_buffer_size);
 457                         return max_const_buffer_size;
 458
 459                 } else {
 460                         return R600_MAX_CONST_BUFFER_SIZE;
 461                 }
 462         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 463                 return R600_MAX_USER_CONST_BUFFERS;
 464         case PIPE_SHADER_CAP_MAX_PREDS:
 465                 return 0; /* nothing uses this */
 466         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 467                 return 1;
 468         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 469                 return 1;
 470         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 471         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 472         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 473         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 474                 return 1;
 475         case PIPE_SHADER_CAP_SUBROUTINES:
 476                 return 0;
 477         case PIPE_SHADER_CAP_INTEGERS:
 478                 return 1;
 479         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 480         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 481                 return 16;
 482         case PIPE_SHADER_CAP_PREFERRED_IR:
 483                 if (shader == PIPE_SHADER_COMPUTE) {
 484 #if HAVE_LLVM < 0x0306
 485                         return PIPE_SHADER_IR_LLVM;
 486 #else
 487                         return PIPE_SHADER_IR_NATIVE;
 488 #endif
 489                 } else {
 490                         return PIPE_SHADER_IR_TGSI;
 491                 }
 492         case PIPE_SHADER_CAP_DOUBLES:
 493                 return 0;
 494         case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 495         case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 496                 return 0;
 497         }
 498         return 0;
 499 }
 500
 501 static void r600_destroy_screen(struct pipe_screen* pscreen)
 502 {
 503         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 504
 505         if (rscreen == NULL)
 506                 return;
 507
 508         if (!rscreen->b.ws->unref(rscreen->b.ws))
 509                 return;
 510
 511         if (rscreen->global_pool) {
 512                 compute_memory_pool_delete(rscreen->global_pool);
 513         }
 514
 515         r600_destroy_common_screen(&rscreen->b);
 516 }
 517
 518 static struct pipe_resource *r600_resource_create(struct pipe_screen *screen,
 519                                                   const struct pipe_resource *templ)
 520 {
 521         if (templ->target == PIPE_BUFFER &&
 522             (templ->bind & PIPE_BIND_GLOBAL))
 523                 return r600_compute_global_buffer_create(screen, templ);
 524
 525         return r600_resource_create_common(screen, templ);
 526 }
 527
 528 struct pipe_screen *r600_screen_create(struct radeon_winsys *ws)
 529 {
 530         struct r600_screen *rscreen = CALLOC_STRUCT(r600_screen);
 531
 532         if (rscreen == NULL) {
 533                 return NULL;
 534         }
 535
 536         /* Set functions first. */
 537         rscreen->b.b.context_create = r600_create_context;
 538         rscreen->b.b.destroy = r600_destroy_screen;
 539         rscreen->b.b.get_param = r600_get_param;
 540         rscreen->b.b.get_shader_param = r600_get_shader_param;
 541         rscreen->b.b.resource_create = r600_resource_create;
 542
 543         if (!r600_common_screen_init(&rscreen->b, ws)) {
 544                 FREE(rscreen);
 545                 return NULL;
 546         }
 547
 548         if (rscreen->b.info.chip_class >= EVERGREEN) {
 549                 rscreen->b.b.is_format_supported = evergreen_is_format_supported;
 550         } else {
 551                 rscreen->b.b.is_format_supported = r600_is_format_supported;
 552         }
 553
 554         rscreen->b.debug_flags |= debug_get_flags_option("R600_DEBUG", r600_debug_options, 0);
 555         if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE))
 556                 rscreen->b.debug_flags |= DBG_COMPUTE;
 557         if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE))
 558                 rscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
 559         if (!debug_get_bool_option("R600_HYPERZ", TRUE))
 560                 rscreen->b.debug_flags |= DBG_NO_HYPERZ;
 561         if (debug_get_bool_option("R600_LLVM", FALSE))
 562                 rscreen->b.debug_flags |= DBG_LLVM;
 563
 564         if (rscreen->b.family == CHIP_UNKNOWN) {
 565                 fprintf(stderr, "r600: Unknown chipset 0x%04X\n", rscreen->b.info.pci_id);
 566                 FREE(rscreen);
 567                 return NULL;
 568         }
 569
 570         /* Figure out streamout kernel support. */
 571         switch (rscreen->b.chip_class) {
 572         case R600:
 573                 if (rscreen->b.family < CHIP_RS780) {
 574                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 575                 } else {
 576                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 23;
 577                 }
 578                 break;
 579         case R700:
 580                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 17;
 581                 break;
 582         case EVERGREEN:
 583         case CAYMAN:
 584                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 585                 break;
 586         default:
 587                 rscreen->b.has_streamout = FALSE;
 588                 break;
 589         }
 590
 591         /* MSAA support. */
 592         switch (rscreen->b.chip_class) {
 593         case R600:
 594         case R700:
 595                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 22;
 596                 rscreen->has_compressed_msaa_texturing = false;
 597                 break;
 598         case EVERGREEN:
 599                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 600                 rscreen->has_compressed_msaa_texturing = rscreen->b.info.drm_minor >= 24;
 601                 break;
 602         case CAYMAN:
 603                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 604                 rscreen->has_compressed_msaa_texturing = true;
 605                 break;
 606         default:
 607                 rscreen->has_msaa = FALSE;
 608                 rscreen->has_compressed_msaa_texturing = false;
 609         }
 610
 611         rscreen->b.has_cp_dma = rscreen->b.info.drm_minor >= 27 &&
 612                               !(rscreen->b.debug_flags & DBG_NO_CP_DMA);
 613
 614         rscreen->global_pool = compute_memory_pool_new(rscreen);
 615
 616         /* Create the auxiliary context. This must be done last. */
 617         rscreen->b.aux_context = rscreen->b.b.context_create(&rscreen->b.b, NULL);
 618
 619 #if 0 /* This is for testing whether aux_context and buffer clearing work correctly. */
 620         struct pipe_resource templ = {};
 621
 622         templ.width0 = 4;
 623         templ.height0 = 2048;
 624         templ.depth0 = 1;
 625         templ.array_size = 1;
 626         templ.target = PIPE_TEXTURE_2D;
 627         templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 628         templ.usage = PIPE_USAGE_DEFAULT;
 629
 630         struct r600_resource *res = r600_resource(rscreen->screen.resource_create(&rscreen->screen, &templ));
 631         unsigned char *map = ws->buffer_map(res->cs_buf, NULL, PIPE_TRANSFER_WRITE);
 632
 633         memset(map, 0, 256);
 634
 635         r600_screen_clear_buffer(rscreen, &res->b.b, 4, 4, 0xCC);
 636         r600_screen_clear_buffer(rscreen, &res->b.b, 8, 4, 0xDD);
 637         r600_screen_clear_buffer(rscreen, &res->b.b, 12, 4, 0xEE);
 638         r600_screen_clear_buffer(rscreen, &res->b.b, 20, 4, 0xFF);
 639         r600_screen_clear_buffer(rscreen, &res->b.b, 32, 20, 0x87);
 640
 641         ws->buffer_wait(res->buf, RADEON_USAGE_WRITE);
 642
 643         int i;
 644         for (i = 0; i < 256; i++) {
 645                 printf("%02X", map[i]);
 646                 if (i % 16 == 15)
 647                         printf("\n");
 648         }
 649 #endif
 650
 651         return &rscreen->b.b;
 652 }