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         unsigned sh;
  72
  73         r600_isa_destroy(rctx->isa);
  74
  75         r600_sb_context_destroy(rctx->sb_context);
  76
  77         pipe_resource_reference((struct pipe_resource**)&rctx->dummy_cmask, NULL);
  78         pipe_resource_reference((struct pipe_resource**)&rctx->dummy_fmask, NULL);
  79
  80         for (sh = 0; sh < PIPE_SHADER_TYPES; sh++) {
  81                 rctx->b.b.set_constant_buffer(&rctx->b.b, sh, R600_BUFFER_INFO_CONST_BUFFER, NULL);
  82                 free(rctx->driver_consts[sh].constants);
  83         }
  84
  85         if (rctx->fixed_func_tcs_shader)
  86                 rctx->b.b.delete_tcs_state(&rctx->b.b, rctx->fixed_func_tcs_shader);
  87
  88         if (rctx->dummy_pixel_shader) {
  89                 rctx->b.b.delete_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
  90         }
  91         if (rctx->custom_dsa_flush) {
  92                 rctx->b.b.delete_depth_stencil_alpha_state(&rctx->b.b, rctx->custom_dsa_flush);
  93         }
  94         if (rctx->custom_blend_resolve) {
  95                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_resolve);
  96         }
  97         if (rctx->custom_blend_decompress) {
  98                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_decompress);
  99         }
 100         if (rctx->custom_blend_fastclear) {
 101                 rctx->b.b.delete_blend_state(&rctx->b.b, rctx->custom_blend_fastclear);
 102         }
 103         util_unreference_framebuffer_state(&rctx->framebuffer.state);
 104
 105         if (rctx->blitter) {
 106                 util_blitter_destroy(rctx->blitter);
 107         }
 108         if (rctx->allocator_fetch_shader) {
 109                 u_suballocator_destroy(rctx->allocator_fetch_shader);
 110         }
 111
 112         r600_release_command_buffer(&rctx->start_cs_cmd);
 113
 114         FREE(rctx->start_compute_cs_cmd.buf);
 115
 116         r600_common_context_cleanup(&rctx->b);
 117         FREE(rctx);
 118 }
 119
 120 static struct pipe_context *r600_create_context(struct pipe_screen *screen,
 121                                                 void *priv, unsigned flags)
 122 {
 123         struct r600_context *rctx = CALLOC_STRUCT(r600_context);
 124         struct r600_screen* rscreen = (struct r600_screen *)screen;
 125         struct radeon_winsys *ws = rscreen->b.ws;
 126
 127         if (!rctx)
 128                 return NULL;
 129
 130         rctx->b.b.screen = screen;
 131         rctx->b.b.priv = priv;
 132         rctx->b.b.destroy = r600_destroy_context;
 133         rctx->b.set_atom_dirty = (void *)r600_set_atom_dirty;
 134
 135         if (!r600_common_context_init(&rctx->b, &rscreen->b))
 136                 goto fail;
 137
 138         rctx->screen = rscreen;
 139         rctx->keep_tiling_flags = rscreen->b.info.drm_minor >= 12;
 140
 141         r600_init_blit_functions(rctx);
 142
 143         if (rscreen->b.info.has_uvd) {
 144                 rctx->b.b.create_video_codec = r600_uvd_create_decoder;
 145                 rctx->b.b.create_video_buffer = r600_video_buffer_create;
 146         } else {
 147                 rctx->b.b.create_video_codec = vl_create_decoder;
 148                 rctx->b.b.create_video_buffer = vl_video_buffer_create;
 149         }
 150
 151         r600_init_common_state_functions(rctx);
 152
 153         switch (rctx->b.chip_class) {
 154         case R600:
 155         case R700:
 156                 r600_init_state_functions(rctx);
 157                 r600_init_atom_start_cs(rctx);
 158                 rctx->custom_dsa_flush = r600_create_db_flush_dsa(rctx);
 159                 rctx->custom_blend_resolve = rctx->b.chip_class == R700 ? r700_create_resolve_blend(rctx)
 160                                                                       : r600_create_resolve_blend(rctx);
 161                 rctx->custom_blend_decompress = r600_create_decompress_blend(rctx);
 162                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_RV610 ||
 163                                            rctx->b.family == CHIP_RV620 ||
 164                                            rctx->b.family == CHIP_RS780 ||
 165                                            rctx->b.family == CHIP_RS880 ||
 166                                            rctx->b.family == CHIP_RV710);
 167                 break;
 168         case EVERGREEN:
 169         case CAYMAN:
 170                 evergreen_init_state_functions(rctx);
 171                 evergreen_init_atom_start_cs(rctx);
 172                 evergreen_init_atom_start_compute_cs(rctx);
 173                 rctx->custom_dsa_flush = evergreen_create_db_flush_dsa(rctx);
 174                 rctx->custom_blend_resolve = evergreen_create_resolve_blend(rctx);
 175                 rctx->custom_blend_decompress = evergreen_create_decompress_blend(rctx);
 176                 rctx->custom_blend_fastclear = evergreen_create_fastclear_blend(rctx);
 177                 rctx->has_vertex_cache = !(rctx->b.family == CHIP_CEDAR ||
 178                                            rctx->b.family == CHIP_PALM ||
 179                                            rctx->b.family == CHIP_SUMO ||
 180                                            rctx->b.family == CHIP_SUMO2 ||
 181                                            rctx->b.family == CHIP_CAICOS ||
 182                                            rctx->b.family == CHIP_CAYMAN ||
 183                                            rctx->b.family == CHIP_ARUBA);
 184                 break;
 185         default:
 186                 R600_ERR("Unsupported chip class %d.\n", rctx->b.chip_class);
 187                 goto fail;
 188         }
 189
 190         rctx->b.gfx.cs = ws->cs_create(rctx->b.ctx, RING_GFX,
 191                                        r600_context_gfx_flush, rctx,
 192                                        rscreen->b.trace_bo ?
 193                                                rscreen->b.trace_bo->buf : NULL);
 194         rctx->b.gfx.flush = r600_context_gfx_flush;
 195
 196         rctx->allocator_fetch_shader = u_suballocator_create(&rctx->b.b, 64 * 1024, 256,
 197                                                              0, PIPE_USAGE_DEFAULT, FALSE);
 198         if (!rctx->allocator_fetch_shader)
 199                 goto fail;
 200
 201         rctx->isa = calloc(1, sizeof(struct r600_isa));
 202         if (!rctx->isa || r600_isa_init(rctx, rctx->isa))
 203                 goto fail;
 204
 205         if (rscreen->b.debug_flags & DBG_FORCE_DMA)
 206                 rctx->b.b.resource_copy_region = rctx->b.dma_copy;
 207
 208         rctx->blitter = util_blitter_create(&rctx->b.b);
 209         if (rctx->blitter == NULL)
 210                 goto fail;
 211         util_blitter_set_texture_multisample(rctx->blitter, rscreen->has_msaa);
 212         rctx->blitter->draw_rectangle = r600_draw_rectangle;
 213
 214         r600_begin_new_cs(rctx);
 215         r600_query_init_backend_mask(&rctx->b); /* this emits commands and must be last */
 216
 217         rctx->dummy_pixel_shader =
 218                 util_make_fragment_cloneinput_shader(&rctx->b.b, 0,
 219                                                      TGSI_SEMANTIC_GENERIC,
 220                                                      TGSI_INTERPOLATE_CONSTANT);
 221         rctx->b.b.bind_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
 222
 223         return &rctx->b.b;
 224
 225 fail:
 226         r600_destroy_context(&rctx->b.b);
 227         return NULL;
 228 }
 229
 230 /*
 231  * pipe_screen
 232  */
 233
 234 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 235 {
 236         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 237         enum radeon_family family = rscreen->b.family;
 238
 239         switch (param) {
 240         /* Supported features (boolean caps). */
 241         case PIPE_CAP_NPOT_TEXTURES:
 242         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 243         case PIPE_CAP_TWO_SIDED_STENCIL:
 244         case PIPE_CAP_ANISOTROPIC_FILTER:
 245         case PIPE_CAP_POINT_SPRITE:
 246         case PIPE_CAP_OCCLUSION_QUERY:
 247         case PIPE_CAP_TEXTURE_SHADOW_MAP:
 248         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 249         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 250         case PIPE_CAP_TEXTURE_SWIZZLE:
 251         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 252         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 253         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 254         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 255         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 256         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 257         case PIPE_CAP_SM3:
 258         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 259         case PIPE_CAP_PRIMITIVE_RESTART:
 260         case PIPE_CAP_CONDITIONAL_RENDER:
 261         case PIPE_CAP_TEXTURE_BARRIER:
 262         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 263         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 264         case PIPE_CAP_TGSI_INSTANCEID:
 265         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 266         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 267         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 268         case PIPE_CAP_USER_INDEX_BUFFERS:
 269         case PIPE_CAP_USER_CONSTANT_BUFFERS:
 270         case PIPE_CAP_START_INSTANCE:
 271         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 272         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 273         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 274         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 275         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 276         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 277         case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 278         case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 279         case PIPE_CAP_SAMPLE_SHADING:
 280         case PIPE_CAP_CLIP_HALFZ:
 281         case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 282         case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 283         case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 284         case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 285         case PIPE_CAP_TGSI_TXQS:
 286         case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 287         case PIPE_CAP_INVALIDATE_BUFFER:
 288         case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
 289                 return 1;
 290
 291         case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 292                 return rscreen->b.info.drm_major == 2 && rscreen->b.info.drm_minor >= 43;
 293
 294         case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 295                 return !R600_BIG_ENDIAN && rscreen->b.info.has_userptr;
 296
 297         case PIPE_CAP_COMPUTE:
 298                 return rscreen->b.chip_class > R700;
 299
 300         case PIPE_CAP_TGSI_TEXCOORD:
 301                 return 0;
 302
 303         case PIPE_CAP_FAKE_SW_MSAA:
 304                 return 0;
 305
 306         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 307                 return MIN2(rscreen->b.info.vram_size, 0xFFFFFFFF);
 308
 309         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 310                 return R600_MAP_BUFFER_ALIGNMENT;
 311
 312         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 313                 return 256;
 314
 315         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 316                 return 1;
 317
 318         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 319                 if (family >= CHIP_CEDAR)
 320                    return 410;
 321                 /* pre-evergreen geom shaders need newer kernel */
 322                 if (rscreen->b.info.drm_minor >= 37)
 323                    return 330;
 324                 return 140;
 325
 326         /* Supported except the original R600. */
 327         case PIPE_CAP_INDEP_BLEND_ENABLE:
 328         case PIPE_CAP_INDEP_BLEND_FUNC:
 329                 /* R600 doesn't support per-MRT blends */
 330                 return family == CHIP_R600 ? 0 : 1;
 331
 332         /* Supported on Evergreen. */
 333         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 334         case PIPE_CAP_CUBE_MAP_ARRAY:
 335         case PIPE_CAP_TEXTURE_GATHER_SM5:
 336         case PIPE_CAP_TEXTURE_QUERY_LOD:
 337         case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 338         case PIPE_CAP_SAMPLER_VIEW_TARGET:
 339                 return family >= CHIP_CEDAR ? 1 : 0;
 340         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 341                 return family >= CHIP_CEDAR ? 4 : 0;
 342         case PIPE_CAP_DRAW_INDIRECT:
 343                 /* kernel command checker support is also required */
 344                 return family >= CHIP_CEDAR && rscreen->b.info.drm_minor >= 41;
 345
 346         case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
 347                 return family >= CHIP_CEDAR ? 0 : 1;
 348
 349         /* Unsupported features. */
 350         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 351         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 352         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 353         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 354         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 355         case PIPE_CAP_USER_VERTEX_BUFFERS:
 356         case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 357         case PIPE_CAP_VERTEXID_NOBASE:
 358         case PIPE_CAP_DEPTH_BOUNDS_TEST:
 359         case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 360         case PIPE_CAP_SHAREABLE_SHADERS:
 361         case PIPE_CAP_CLEAR_TEXTURE:
 362         case PIPE_CAP_DRAW_PARAMETERS:
 363         case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 364         case PIPE_CAP_MULTI_DRAW_INDIRECT:
 365         case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 366         case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 367         case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 368         case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 369         case PIPE_CAP_GENERATE_MIPMAP:
 370         case PIPE_CAP_STRING_MARKER:
 371         case PIPE_CAP_QUERY_BUFFER_OBJECT:
 372                 return 0;
 373
 374         case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 375                 if (family >= CHIP_CEDAR)
 376                         return 30;
 377                 else
 378                         return 0;
 379         /* Stream output. */
 380         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 381                 return rscreen->b.has_streamout ? 4 : 0;
 382         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 383                 return rscreen->b.has_streamout ? 1 : 0;
 384         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 385         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 386                 return 32*4;
 387
 388         /* Geometry shader output. */
 389         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 390                 return 1024;
 391         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 392                 return 16384;
 393         case PIPE_CAP_MAX_VERTEX_STREAMS:
 394                 return family >= CHIP_CEDAR ? 4 : 1;
 395
 396         case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 397                 return 2047;
 398
 399         /* Texturing. */
 400         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 401         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 402                 if (family >= CHIP_CEDAR)
 403                         return 15;
 404                 else
 405                         return 14;
 406         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 407                 /* textures support 8192, but layered rendering supports 2048 */
 408                 return 12;
 409         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 410                 /* textures support 8192, but layered rendering supports 2048 */
 411                 return rscreen->b.info.drm_minor >= 9 ? 2048 : 0;
 412
 413         /* Render targets. */
 414         case PIPE_CAP_MAX_RENDER_TARGETS:
 415                 /* XXX some r6xx are buggy and can only do 4 */
 416                 return 8;
 417
 418         case PIPE_CAP_MAX_VIEWPORTS:
 419                 return R600_MAX_VIEWPORTS;
 420
 421         /* Timer queries, present when the clock frequency is non zero. */
 422         case PIPE_CAP_QUERY_TIME_ELAPSED:
 423                 return rscreen->b.info.clock_crystal_freq != 0;
 424         case PIPE_CAP_QUERY_TIMESTAMP:
 425                 return rscreen->b.info.drm_minor >= 20 &&
 426                        rscreen->b.info.clock_crystal_freq != 0;
 427
 428         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 429         case PIPE_CAP_MIN_TEXEL_OFFSET:
 430                 return -8;
 431
 432         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 433         case PIPE_CAP_MAX_TEXEL_OFFSET:
 434                 return 7;
 435
 436         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 437                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 438         case PIPE_CAP_ENDIANNESS:
 439                 return PIPE_ENDIAN_LITTLE;
 440
 441         case PIPE_CAP_VENDOR_ID:
 442                 return 0x1002;
 443         case PIPE_CAP_DEVICE_ID:
 444                 return rscreen->b.info.pci_id;
 445         case PIPE_CAP_ACCELERATED:
 446                 return 1;
 447         case PIPE_CAP_VIDEO_MEMORY:
 448                 return rscreen->b.info.vram_size >> 20;
 449         case PIPE_CAP_UMA:
 450                 return 0;
 451         case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 452                 return rscreen->b.chip_class >= R700;
 453         }
 454         return 0;
 455 }
 456
 457 static int r600_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
 458 {
 459         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 460
 461         switch(shader)
 462         {
 463         case PIPE_SHADER_FRAGMENT:
 464         case PIPE_SHADER_VERTEX:
 465         case PIPE_SHADER_COMPUTE:
 466                 break;
 467         case PIPE_SHADER_GEOMETRY:
 468                 if (rscreen->b.family >= CHIP_CEDAR)
 469                         break;
 470                 /* pre-evergreen geom shaders need newer kernel */
 471                 if (rscreen->b.info.drm_minor >= 37)
 472                         break;
 473                 return 0;
 474         case PIPE_SHADER_TESS_CTRL:
 475         case PIPE_SHADER_TESS_EVAL:
 476                 if (rscreen->b.family >= CHIP_CEDAR)
 477                         break;
 478         default:
 479                 return 0;
 480         }
 481
 482         switch (param) {
 483         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 484         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 485         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 486         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 487                 return 16384;
 488         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 489                 return 32;
 490         case PIPE_SHADER_CAP_MAX_INPUTS:
 491                 return shader == PIPE_SHADER_VERTEX ? 16 : 32;
 492         case PIPE_SHADER_CAP_MAX_OUTPUTS:
 493                 return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
 494         case PIPE_SHADER_CAP_MAX_TEMPS:
 495                 return 256; /* Max native temporaries. */
 496         case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 497                 if (shader == PIPE_SHADER_COMPUTE) {
 498                         uint64_t max_const_buffer_size;
 499                         pscreen->get_compute_param(pscreen,
 500                                 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
 501                                 &max_const_buffer_size);
 502                         return max_const_buffer_size;
 503
 504                 } else {
 505                         return R600_MAX_CONST_BUFFER_SIZE;
 506                 }
 507         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 508                 return R600_MAX_USER_CONST_BUFFERS;
 509         case PIPE_SHADER_CAP_MAX_PREDS:
 510                 return 0; /* nothing uses this */
 511         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 512                 return 1;
 513         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 514                 return 1;
 515         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 516         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 517         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 518         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 519                 return 1;
 520         case PIPE_SHADER_CAP_SUBROUTINES:
 521                 return 0;
 522         case PIPE_SHADER_CAP_INTEGERS:
 523         case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 524                 return 1;
 525         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 526         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 527                 return 16;
 528         case PIPE_SHADER_CAP_PREFERRED_IR:
 529                 if (shader == PIPE_SHADER_COMPUTE) {
 530 #if HAVE_LLVM < 0x0306
 531                         return PIPE_SHADER_IR_LLVM;
 532 #else
 533                         return PIPE_SHADER_IR_NATIVE;
 534 #endif
 535                 } else {
 536                         return PIPE_SHADER_IR_TGSI;
 537                 }
 538         case PIPE_SHADER_CAP_DOUBLES:
 539                 if (rscreen->b.family == CHIP_CYPRESS ||
 540                         rscreen->b.family == CHIP_CAYMAN || rscreen->b.family == CHIP_ARUBA)
 541                         return 1;
 542                 return 0;
 543         case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 544         case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 545         case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 546         case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 547                 return 0;
 548         case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 549                 /* due to a bug in the shader compiler, some loops hang
 550                  * if they are not unrolled, see:
 551                  *    https://bugs.freedesktop.org/show_bug.cgi?id=86720
 552                  */
 553                 return 255;
 554         }
 555         return 0;
 556 }
 557
 558 static void r600_destroy_screen(struct pipe_screen* pscreen)
 559 {
 560         struct r600_screen *rscreen = (struct r600_screen *)pscreen;
 561
 562         if (!rscreen)
 563                 return;
 564
 565         if (!rscreen->b.ws->unref(rscreen->b.ws))
 566                 return;
 567
 568         if (rscreen->global_pool) {
 569                 compute_memory_pool_delete(rscreen->global_pool);
 570         }
 571
 572         r600_destroy_common_screen(&rscreen->b);
 573 }
 574
 575 static struct pipe_resource *r600_resource_create(struct pipe_screen *screen,
 576                                                   const struct pipe_resource *templ)
 577 {
 578         if (templ->target == PIPE_BUFFER &&
 579             (templ->bind & PIPE_BIND_GLOBAL))
 580                 return r600_compute_global_buffer_create(screen, templ);
 581
 582         return r600_resource_create_common(screen, templ);
 583 }
 584
 585 struct pipe_screen *r600_screen_create(struct radeon_winsys *ws)
 586 {
 587         struct r600_screen *rscreen = CALLOC_STRUCT(r600_screen);
 588
 589         if (!rscreen) {
 590                 return NULL;
 591         }
 592
 593         /* Set functions first. */
 594         rscreen->b.b.context_create = r600_create_context;
 595         rscreen->b.b.destroy = r600_destroy_screen;
 596         rscreen->b.b.get_param = r600_get_param;
 597         rscreen->b.b.get_shader_param = r600_get_shader_param;
 598         rscreen->b.b.resource_create = r600_resource_create;
 599
 600         if (!r600_common_screen_init(&rscreen->b, ws)) {
 601                 FREE(rscreen);
 602                 return NULL;
 603         }
 604
 605         if (rscreen->b.info.chip_class >= EVERGREEN) {
 606                 rscreen->b.b.is_format_supported = evergreen_is_format_supported;
 607         } else {
 608                 rscreen->b.b.is_format_supported = r600_is_format_supported;
 609         }
 610
 611         rscreen->b.debug_flags |= debug_get_flags_option("R600_DEBUG", r600_debug_options, 0);
 612         if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE))
 613                 rscreen->b.debug_flags |= DBG_COMPUTE;
 614         if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE))
 615                 rscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS | DBG_TCS | DBG_TES;
 616         if (!debug_get_bool_option("R600_HYPERZ", TRUE))
 617                 rscreen->b.debug_flags |= DBG_NO_HYPERZ;
 618         if (debug_get_bool_option("R600_LLVM", FALSE))
 619                 rscreen->b.debug_flags |= DBG_LLVM;
 620
 621         if (rscreen->b.family == CHIP_UNKNOWN) {
 622                 fprintf(stderr, "r600: Unknown chipset 0x%04X\n", rscreen->b.info.pci_id);
 623                 FREE(rscreen);
 624                 return NULL;
 625         }
 626
 627         /* Figure out streamout kernel support. */
 628         switch (rscreen->b.chip_class) {
 629         case R600:
 630                 if (rscreen->b.family < CHIP_RS780) {
 631                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 632                 } else {
 633                         rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 23;
 634                 }
 635                 break;
 636         case R700:
 637                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 17;
 638                 break;
 639         case EVERGREEN:
 640         case CAYMAN:
 641                 rscreen->b.has_streamout = rscreen->b.info.drm_minor >= 14;
 642                 break;
 643         default:
 644                 rscreen->b.has_streamout = FALSE;
 645                 break;
 646         }
 647
 648         /* MSAA support. */
 649         switch (rscreen->b.chip_class) {
 650         case R600:
 651         case R700:
 652                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 22;
 653                 rscreen->has_compressed_msaa_texturing = false;
 654                 break;
 655         case EVERGREEN:
 656                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 657                 rscreen->has_compressed_msaa_texturing = rscreen->b.info.drm_minor >= 24;
 658                 break;
 659         case CAYMAN:
 660                 rscreen->has_msaa = rscreen->b.info.drm_minor >= 19;
 661                 rscreen->has_compressed_msaa_texturing = true;
 662                 break;
 663         default:
 664                 rscreen->has_msaa = FALSE;
 665                 rscreen->has_compressed_msaa_texturing = false;
 666         }
 667
 668         rscreen->b.has_cp_dma = rscreen->b.info.drm_minor >= 27 &&
 669                               !(rscreen->b.debug_flags & DBG_NO_CP_DMA);
 670
 671         rscreen->global_pool = compute_memory_pool_new(rscreen);
 672
 673         /* Create the auxiliary context. This must be done last. */
 674         rscreen->b.aux_context = rscreen->b.b.context_create(&rscreen->b.b, NULL, 0);
 675
 676 #if 0 /* This is for testing whether aux_context and buffer clearing work correctly. */
 677         struct pipe_resource templ = {};
 678
 679         templ.width0 = 4;
 680         templ.height0 = 2048;
 681         templ.depth0 = 1;
 682         templ.array_size = 1;
 683         templ.target = PIPE_TEXTURE_2D;
 684         templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 685         templ.usage = PIPE_USAGE_DEFAULT;
 686
 687         struct r600_resource *res = r600_resource(rscreen->screen.resource_create(&rscreen->screen, &templ));
 688         unsigned char *map = ws->buffer_map(res->buf, NULL, PIPE_TRANSFER_WRITE);
 689
 690         memset(map, 0, 256);
 691
 692         r600_screen_clear_buffer(rscreen, &res->b.b, 4, 4, 0xCC);
 693         r600_screen_clear_buffer(rscreen, &res->b.b, 8, 4, 0xDD);
 694         r600_screen_clear_buffer(rscreen, &res->b.b, 12, 4, 0xEE);
 695         r600_screen_clear_buffer(rscreen, &res->b.b, 20, 4, 0xFF);
 696         r600_screen_clear_buffer(rscreen, &res->b.b, 32, 20, 0x87);
 697
 698         ws->buffer_wait(res->buf, RADEON_USAGE_WRITE);
 699
 700         int i;
 701         for (i = 0; i < 256; i++) {
 702                 printf("%02X", map[i]);
 703                 if (i % 16 == 15)
 704                         printf("\n");
 705         }
 706 #endif
 707
 708         return &rscreen->b.b;
 709 }