src/gallium/drivers/radeonsi/si_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
  24 #include "si_pipe.h"
  25 #include "si_public.h"
  26 #include "sid.h"
  27
  28 #include "radeon/radeon_llvm_emit.h"
  29 #include "radeon/radeon_uvd.h"
  30 #include "util/u_memory.h"
  31 #include "vl/vl_decoder.h"
  32
  33 /*
  34  * pipe_context
  35  */
  36 static void si_destroy_context(struct pipe_context *context)
  37 {
  38         struct si_context *sctx = (struct si_context *)context;
  39         int i;
  40
  41         si_release_all_descriptors(sctx);
  42
  43         pipe_resource_reference(&sctx->esgs_ring, NULL);
  44         pipe_resource_reference(&sctx->gsvs_ring, NULL);
  45         pipe_resource_reference(&sctx->tf_ring, NULL);
  46         pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
  47         r600_resource_reference(&sctx->border_color_buffer, NULL);
  48         free(sctx->border_color_table);
  49         r600_resource_reference(&sctx->scratch_buffer, NULL);
  50         sctx->b.ws->fence_reference(&sctx->last_gfx_fence, NULL);
  51
  52         si_pm4_free_state(sctx, sctx->init_config, ~0);
  53         if (sctx->init_config_gs_rings)
  54                 si_pm4_free_state(sctx, sctx->init_config_gs_rings, ~0);
  55         for (i = 0; i < Elements(sctx->vgt_shader_config); i++)
  56                 si_pm4_delete_state(sctx, vgt_shader_config, sctx->vgt_shader_config[i]);
  57
  58         if (sctx->pstipple_sampler_state)
  59                 sctx->b.b.delete_sampler_state(&sctx->b.b, sctx->pstipple_sampler_state);
  60         if (sctx->fixed_func_tcs_shader.cso)
  61                 sctx->b.b.delete_tcs_state(&sctx->b.b, sctx->fixed_func_tcs_shader.cso);
  62         if (sctx->custom_dsa_flush)
  63                 sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush);
  64         if (sctx->custom_blend_resolve)
  65                 sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
  66         if (sctx->custom_blend_decompress)
  67                 sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
  68         if (sctx->custom_blend_fastclear)
  69                 sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_fastclear);
  70         util_unreference_framebuffer_state(&sctx->framebuffer.state);
  71
  72         if (sctx->blitter)
  73                 util_blitter_destroy(sctx->blitter);
  74
  75         r600_common_context_cleanup(&sctx->b);
  76
  77 #if HAVE_LLVM >= 0x0306
  78         LLVMDisposeTargetMachine(sctx->tm);
  79 #endif
  80
  81         r600_resource_reference(&sctx->trace_buf, NULL);
  82         r600_resource_reference(&sctx->last_trace_buf, NULL);
  83         free(sctx->last_ib);
  84         if (sctx->last_bo_list) {
  85                 for (i = 0; i < sctx->last_bo_count; i++)
  86                         pb_reference(&sctx->last_bo_list[i].buf, NULL);
  87                 free(sctx->last_bo_list);
  88         }
  89         FREE(sctx);
  90 }
  91
  92 static enum pipe_reset_status
  93 si_amdgpu_get_reset_status(struct pipe_context *ctx)
  94 {
  95         struct si_context *sctx = (struct si_context *)ctx;
  96
  97         return sctx->b.ws->ctx_query_reset_status(sctx->b.ctx);
  98 }
  99
 100 static struct pipe_context *si_create_context(struct pipe_screen *screen,
 101                                               void *priv, unsigned flags)
 102 {
 103         struct si_context *sctx = CALLOC_STRUCT(si_context);
 104         struct si_screen* sscreen = (struct si_screen *)screen;
 105         struct radeon_winsys *ws = sscreen->b.ws;
 106         LLVMTargetRef r600_target;
 107 #if HAVE_LLVM >= 0x0306
 108         const char *triple = "amdgcn--";
 109 #endif
 110         int shader, i;
 111
 112         if (!sctx)
 113                 return NULL;
 114
 115         if (sscreen->b.debug_flags & DBG_CHECK_VM)
 116                 flags |= PIPE_CONTEXT_DEBUG;
 117
 118         sctx->b.b.screen = screen; /* this must be set first */
 119         sctx->b.b.priv = priv;
 120         sctx->b.b.destroy = si_destroy_context;
 121         sctx->b.set_atom_dirty = (void *)si_set_atom_dirty;
 122         sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
 123         sctx->is_debug = (flags & PIPE_CONTEXT_DEBUG) != 0;
 124
 125         if (!r600_common_context_init(&sctx->b, &sscreen->b))
 126                 goto fail;
 127
 128         if (sscreen->b.info.drm_major == 3)
 129                 sctx->b.b.get_device_reset_status = si_amdgpu_get_reset_status;
 130
 131         si_init_blit_functions(sctx);
 132         si_init_compute_functions(sctx);
 133         si_init_cp_dma_functions(sctx);
 134         si_init_debug_functions(sctx);
 135
 136         if (sscreen->b.info.has_uvd) {
 137                 sctx->b.b.create_video_codec = si_uvd_create_decoder;
 138                 sctx->b.b.create_video_buffer = si_video_buffer_create;
 139         } else {
 140                 sctx->b.b.create_video_codec = vl_create_decoder;
 141                 sctx->b.b.create_video_buffer = vl_video_buffer_create;
 142         }
 143
 144         sctx->b.gfx.cs = ws->cs_create(sctx->b.ctx, RING_GFX, si_context_gfx_flush,
 145                                        sctx, sscreen->b.trace_bo ?
 146                                                sscreen->b.trace_bo->buf : NULL);
 147         sctx->b.gfx.flush = si_context_gfx_flush;
 148
 149         /* Border colors. */
 150         sctx->border_color_table = malloc(SI_MAX_BORDER_COLORS *
 151                                           sizeof(*sctx->border_color_table));
 152         if (!sctx->border_color_table)
 153                 goto fail;
 154
 155         sctx->border_color_buffer = (struct r600_resource*)
 156                 pipe_buffer_create(screen, PIPE_BIND_CUSTOM, PIPE_USAGE_DEFAULT,
 157                                    SI_MAX_BORDER_COLORS *
 158                                    sizeof(*sctx->border_color_table));
 159         if (!sctx->border_color_buffer)
 160                 goto fail;
 161
 162         sctx->border_color_map =
 163                 ws->buffer_map(sctx->border_color_buffer->buf,
 164                                NULL, PIPE_TRANSFER_WRITE);
 165         if (!sctx->border_color_map)
 166                 goto fail;
 167
 168         si_init_all_descriptors(sctx);
 169         si_init_state_functions(sctx);
 170         si_init_shader_functions(sctx);
 171
 172         if (sscreen->b.debug_flags & DBG_FORCE_DMA)
 173                 sctx->b.b.resource_copy_region = sctx->b.dma_copy;
 174
 175         sctx->blitter = util_blitter_create(&sctx->b.b);
 176         if (sctx->blitter == NULL)
 177                 goto fail;
 178         sctx->blitter->draw_rectangle = r600_draw_rectangle;
 179
 180         sctx->sample_mask.sample_mask = 0xffff;
 181
 182         /* these must be last */
 183         si_begin_new_cs(sctx);
 184         r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
 185
 186         /* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
 187          * with a NULL buffer). We need to use a dummy buffer instead. */
 188         if (sctx->b.chip_class == CIK) {
 189                 sctx->null_const_buf.buffer = pipe_buffer_create(screen, PIPE_BIND_CONSTANT_BUFFER,
 190                                                                  PIPE_USAGE_DEFAULT, 16);
 191                 if (!sctx->null_const_buf.buffer)
 192                         goto fail;
 193                 sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
 194
 195                 for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
 196                         for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
 197                                 sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
 198                                                               &sctx->null_const_buf);
 199                         }
 200                 }
 201
 202                 /* Clear the NULL constant buffer, because loads should return zeros. */
 203                 sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
 204                                      sctx->null_const_buf.buffer->width0, 0, false);
 205         }
 206
 207         /* XXX: This is the maximum value allowed.  I'm not sure how to compute
 208          * this for non-cs shaders.  Using the wrong value here can result in
 209          * GPU lockups, but the maximum value seems to always work.
 210          */
 211         sctx->scratch_waves = 32 * sscreen->b.info.num_good_compute_units;
 212
 213 #if HAVE_LLVM >= 0x0306
 214         /* Initialize LLVM TargetMachine */
 215         r600_target = radeon_llvm_get_r600_target(triple);
 216         sctx->tm = LLVMCreateTargetMachine(r600_target, triple,
 217                                            r600_get_llvm_processor_name(sscreen->b.family),
 218 #if HAVE_LLVM >= 0x0308
 219                                            sscreen->b.debug_flags & DBG_SI_SCHED ?
 220                                                 "+DumpCode,+vgpr-spilling,+si-scheduler" :
 221 #endif
 222                                                 "+DumpCode,+vgpr-spilling",
 223                                            LLVMCodeGenLevelDefault,
 224                                            LLVMRelocDefault,
 225                                            LLVMCodeModelDefault);
 226 #endif
 227
 228         return &sctx->b.b;
 229 fail:
 230         fprintf(stderr, "radeonsi: Failed to create a context.\n");
 231         si_destroy_context(&sctx->b.b);
 232         return NULL;
 233 }
 234
 235 /*
 236  * pipe_screen
 237  */
 238
 239 static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 240 {
 241         struct si_screen *sscreen = (struct si_screen *)pscreen;
 242
 243         switch (param) {
 244         /* Supported features (boolean caps). */
 245         case PIPE_CAP_TWO_SIDED_STENCIL:
 246         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 247         case PIPE_CAP_ANISOTROPIC_FILTER:
 248         case PIPE_CAP_POINT_SPRITE:
 249         case PIPE_CAP_OCCLUSION_QUERY:
 250         case PIPE_CAP_TEXTURE_SHADOW_MAP:
 251         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 252         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 253         case PIPE_CAP_TEXTURE_SWIZZLE:
 254         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 255         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 256         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 257         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 258         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 259         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 260         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 261         case PIPE_CAP_SM3:
 262         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 263         case PIPE_CAP_PRIMITIVE_RESTART:
 264         case PIPE_CAP_CONDITIONAL_RENDER:
 265         case PIPE_CAP_TEXTURE_BARRIER:
 266         case PIPE_CAP_INDEP_BLEND_ENABLE:
 267         case PIPE_CAP_INDEP_BLEND_FUNC:
 268         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 269         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 270         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 271         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 272         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 273         case PIPE_CAP_USER_INDEX_BUFFERS:
 274         case PIPE_CAP_USER_CONSTANT_BUFFERS:
 275         case PIPE_CAP_START_INSTANCE:
 276         case PIPE_CAP_NPOT_TEXTURES:
 277         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 278         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 279         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 280         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 281         case PIPE_CAP_TGSI_INSTANCEID:
 282         case PIPE_CAP_COMPUTE:
 283         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 284         case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
 285         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 286         case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
 287         case PIPE_CAP_CUBE_MAP_ARRAY:
 288         case PIPE_CAP_SAMPLE_SHADING:
 289         case PIPE_CAP_DRAW_INDIRECT:
 290         case PIPE_CAP_CLIP_HALFZ:
 291         case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
 292         case PIPE_CAP_POLYGON_OFFSET_CLAMP:
 293         case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 294         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 295         case PIPE_CAP_TGSI_TEXCOORD:
 296         case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 297         case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 298         case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
 299         case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
 300         case PIPE_CAP_SHAREABLE_SHADERS:
 301         case PIPE_CAP_DEPTH_BOUNDS_TEST:
 302         case PIPE_CAP_SAMPLER_VIEW_TARGET:
 303         case PIPE_CAP_TEXTURE_QUERY_LOD:
 304         case PIPE_CAP_TEXTURE_GATHER_SM5:
 305         case PIPE_CAP_TGSI_TXQS:
 306         case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
 307         case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
 308         case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
 309         case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
 310         case PIPE_CAP_INVALIDATE_BUFFER:
 311                 return 1;
 312
 313         case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
 314                 return !SI_BIG_ENDIAN && sscreen->b.info.has_userptr;
 315
 316         case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
 317                 return (sscreen->b.info.drm_major == 2 &&
 318                         sscreen->b.info.drm_minor >= 43) ||
 319                        sscreen->b.info.drm_major == 3;
 320
 321         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 322                 /* 2D tiling on CIK is supported since DRM 2.35.0 */
 323                 return sscreen->b.chip_class < CIK ||
 324                        (sscreen->b.info.drm_major == 2 &&
 325                         sscreen->b.info.drm_minor >= 35) ||
 326                        sscreen->b.info.drm_major == 3;
 327
 328         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 329                 return R600_MAP_BUFFER_ALIGNMENT;
 330
 331         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 332         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 333         case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
 334                 return 4;
 335
 336         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 337                 return HAVE_LLVM >= 0x0307 ? 410 : 330;
 338
 339         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 340                 return MIN2(sscreen->b.info.vram_size, 0xFFFFFFFF);
 341
 342         /* Unsupported features. */
 343         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 344         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 345         case PIPE_CAP_USER_VERTEX_BUFFERS:
 346         case PIPE_CAP_FAKE_SW_MSAA:
 347         case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
 348         case PIPE_CAP_VERTEXID_NOBASE:
 349         case PIPE_CAP_CLEAR_TEXTURE:
 350         case PIPE_CAP_DRAW_PARAMETERS:
 351         case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
 352         case PIPE_CAP_MULTI_DRAW_INDIRECT:
 353         case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
 354         case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
 355         case PIPE_CAP_GENERATE_MIPMAP:
 356         case PIPE_CAP_STRING_MARKER:
 357                 return 0;
 358
 359         case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
 360                 return 30;
 361
 362         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 363                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 364
 365         /* Stream output. */
 366         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 367                 return sscreen->b.has_streamout ? 4 : 0;
 368         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 369                 return sscreen->b.has_streamout ? 1 : 0;
 370         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 371         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 372                 return sscreen->b.has_streamout ? 32*4 : 0;
 373
 374         /* Geometry shader output. */
 375         case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
 376                 return 1024;
 377         case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 378                 return 4095;
 379         case PIPE_CAP_MAX_VERTEX_STREAMS:
 380                 return 4;
 381
 382         case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 383                 return 2048;
 384
 385         /* Texturing. */
 386         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 387         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 388                 return 15; /* 16384 */
 389         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 390                 /* textures support 8192, but layered rendering supports 2048 */
 391                 return 12;
 392         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 393                 /* textures support 8192, but layered rendering supports 2048 */
 394                 return 2048;
 395
 396         /* Render targets. */
 397         case PIPE_CAP_MAX_RENDER_TARGETS:
 398                 return 8;
 399
 400         case PIPE_CAP_MAX_VIEWPORTS:
 401                 return SI_MAX_VIEWPORTS;
 402
 403         /* Timer queries, present when the clock frequency is non zero. */
 404         case PIPE_CAP_QUERY_TIMESTAMP:
 405         case PIPE_CAP_QUERY_TIME_ELAPSED:
 406                 return sscreen->b.info.r600_clock_crystal_freq != 0;
 407
 408         case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
 409         case PIPE_CAP_MIN_TEXEL_OFFSET:
 410                 return -32;
 411
 412         case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
 413         case PIPE_CAP_MAX_TEXEL_OFFSET:
 414                 return 31;
 415
 416         case PIPE_CAP_ENDIANNESS:
 417                 return PIPE_ENDIAN_LITTLE;
 418
 419         case PIPE_CAP_VENDOR_ID:
 420                 return 0x1002;
 421         case PIPE_CAP_DEVICE_ID:
 422                 return sscreen->b.info.pci_id;
 423         case PIPE_CAP_ACCELERATED:
 424                 return 1;
 425         case PIPE_CAP_VIDEO_MEMORY:
 426                 return sscreen->b.info.vram_size >> 20;
 427         case PIPE_CAP_UMA:
 428                 return 0;
 429         }
 430         return 0;
 431 }
 432
 433 static int si_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
 434 {
 435         switch(shader)
 436         {
 437         case PIPE_SHADER_FRAGMENT:
 438         case PIPE_SHADER_VERTEX:
 439         case PIPE_SHADER_GEOMETRY:
 440                 break;
 441         case PIPE_SHADER_TESS_CTRL:
 442         case PIPE_SHADER_TESS_EVAL:
 443                 /* LLVM 3.6.2 is required for tessellation because of bug fixes there */
 444                 if (HAVE_LLVM < 0x0306 ||
 445                     (HAVE_LLVM == 0x0306 && MESA_LLVM_VERSION_PATCH < 2))
 446                         return 0;
 447                 break;
 448         case PIPE_SHADER_COMPUTE:
 449                 switch (param) {
 450                 case PIPE_SHADER_CAP_PREFERRED_IR:
 451 #if HAVE_LLVM < 0x0306
 452                         return PIPE_SHADER_IR_LLVM;
 453 #else
 454                         return PIPE_SHADER_IR_NATIVE;
 455 #endif
 456                 case PIPE_SHADER_CAP_DOUBLES:
 457                         return HAVE_LLVM >= 0x0307;
 458
 459                 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE: {
 460                         uint64_t max_const_buffer_size;
 461                         pscreen->get_compute_param(pscreen,
 462                                 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
 463                                 &max_const_buffer_size);
 464                         return max_const_buffer_size;
 465                 }
 466                 default:
 467                         /* If compute shaders don't require a special value
 468                          * for this cap, we can return the same value we
 469                          * do for other shader types. */
 470                         break;
 471                 }
 472                 break;
 473         default:
 474                 return 0;
 475         }
 476
 477         switch (param) {
 478         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 479         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 480         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 481         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 482                 return 16384;
 483         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 484                 return 32;
 485         case PIPE_SHADER_CAP_MAX_INPUTS:
 486                 return shader == PIPE_SHADER_VERTEX ? SI_NUM_VERTEX_BUFFERS : 32;
 487         case PIPE_SHADER_CAP_MAX_OUTPUTS:
 488                 return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
 489         case PIPE_SHADER_CAP_MAX_TEMPS:
 490                 return 256; /* Max native temporaries. */
 491         case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 492                 return 4096 * sizeof(float[4]); /* actually only memory limits this */
 493         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 494                 return SI_NUM_USER_CONST_BUFFERS;
 495         case PIPE_SHADER_CAP_MAX_PREDS:
 496                 return 0; /* FIXME */
 497         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 498                 return 1;
 499         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 500                 return 1;
 501         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 502                 /* Indirection of geometry shader input dimension is not
 503                  * handled yet
 504                  */
 505                 return shader != PIPE_SHADER_GEOMETRY;
 506         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 507         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 508         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 509                 return 1;
 510         case PIPE_SHADER_CAP_INTEGERS:
 511                 return 1;
 512         case PIPE_SHADER_CAP_SUBROUTINES:
 513                 return 0;
 514         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 515         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 516                 return 16;
 517         case PIPE_SHADER_CAP_PREFERRED_IR:
 518                 return PIPE_SHADER_IR_TGSI;
 519         case PIPE_SHADER_CAP_DOUBLES:
 520                 return HAVE_LLVM >= 0x0307;
 521         case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
 522         case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
 523                 return 0;
 524         case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
 525         case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
 526                 return 1;
 527         case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
 528                 return 32;
 529         case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
 530                 return 0;
 531         }
 532         return 0;
 533 }
 534
 535 static void si_destroy_screen(struct pipe_screen* pscreen)
 536 {
 537         struct si_screen *sscreen = (struct si_screen *)pscreen;
 538
 539         if (!sscreen)
 540                 return;
 541
 542         if (!sscreen->b.ws->unref(sscreen->b.ws))
 543                 return;
 544
 545         r600_destroy_common_screen(&sscreen->b);
 546 }
 547
 548 #define SI_TILE_MODE_COLOR_2D_8BPP  14
 549
 550 /* Initialize pipe config. This is especially important for GPUs
 551  * with 16 pipes and more where it's initialized incorrectly by
 552  * the TILING_CONFIG ioctl. */
 553 static bool si_initialize_pipe_config(struct si_screen *sscreen)
 554 {
 555         unsigned mode2d;
 556
 557         /* This is okay, because there can be no 2D tiling without
 558          * the tile mode array, so we won't need the pipe config.
 559          * Return "success".
 560          */
 561         if (!sscreen->b.info.si_tile_mode_array_valid)
 562                 return true;
 563
 564         /* The same index is used for the 2D mode on CIK too. */
 565         mode2d = sscreen->b.info.si_tile_mode_array[SI_TILE_MODE_COLOR_2D_8BPP];
 566
 567         switch (G_009910_PIPE_CONFIG(mode2d)) {
 568         case V_02803C_ADDR_SURF_P2:
 569                 sscreen->b.tiling_info.num_channels = 2;
 570                 break;
 571         case V_02803C_X_ADDR_SURF_P4_8X16:
 572         case V_02803C_X_ADDR_SURF_P4_16X16:
 573         case V_02803C_X_ADDR_SURF_P4_16X32:
 574         case V_02803C_X_ADDR_SURF_P4_32X32:
 575                 sscreen->b.tiling_info.num_channels = 4;
 576                 break;
 577         case V_02803C_X_ADDR_SURF_P8_16X16_8X16:
 578         case V_02803C_X_ADDR_SURF_P8_16X32_8X16:
 579         case V_02803C_X_ADDR_SURF_P8_32X32_8X16:
 580         case V_02803C_X_ADDR_SURF_P8_16X32_16X16:
 581         case V_02803C_X_ADDR_SURF_P8_32X32_16X16:
 582         case V_02803C_X_ADDR_SURF_P8_32X32_16X32:
 583         case V_02803C_X_ADDR_SURF_P8_32X64_32X32:
 584                 sscreen->b.tiling_info.num_channels = 8;
 585                 break;
 586         case V_02803C_X_ADDR_SURF_P16_32X32_8X16:
 587         case V_02803C_X_ADDR_SURF_P16_32X32_16X16:
 588                 sscreen->b.tiling_info.num_channels = 16;
 589                 break;
 590         default:
 591                 assert(0);
 592                 fprintf(stderr, "radeonsi: Unknown pipe config %i.\n",
 593                         G_009910_PIPE_CONFIG(mode2d));
 594                 return false;
 595         }
 596         return true;
 597 }
 598
 599 static bool si_init_gs_info(struct si_screen *sscreen)
 600 {
 601         switch (sscreen->b.family) {
 602         case CHIP_OLAND:
 603         case CHIP_HAINAN:
 604         case CHIP_KAVERI:
 605         case CHIP_KABINI:
 606         case CHIP_MULLINS:
 607         case CHIP_ICELAND:
 608         case CHIP_CARRIZO:
 609         case CHIP_STONEY:
 610                 sscreen->gs_table_depth = 16;
 611                 return true;
 612         case CHIP_TAHITI:
 613         case CHIP_PITCAIRN:
 614         case CHIP_VERDE:
 615         case CHIP_BONAIRE:
 616         case CHIP_HAWAII:
 617         case CHIP_TONGA:
 618         case CHIP_FIJI:
 619                 sscreen->gs_table_depth = 32;
 620                 return true;
 621         default:
 622                 return false;
 623         }
 624 }
 625
 626 struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
 627 {
 628         struct si_screen *sscreen = CALLOC_STRUCT(si_screen);
 629
 630         if (!sscreen) {
 631                 return NULL;
 632         }
 633
 634         /* Set functions first. */
 635         sscreen->b.b.context_create = si_create_context;
 636         sscreen->b.b.destroy = si_destroy_screen;
 637         sscreen->b.b.get_param = si_get_param;
 638         sscreen->b.b.get_shader_param = si_get_shader_param;
 639         sscreen->b.b.is_format_supported = si_is_format_supported;
 640         sscreen->b.b.resource_create = r600_resource_create_common;
 641
 642         if (!r600_common_screen_init(&sscreen->b, ws) ||
 643             !si_initialize_pipe_config(sscreen) ||
 644             !si_init_gs_info(sscreen)) {
 645                 FREE(sscreen);
 646                 return NULL;
 647         }
 648
 649         if (!debug_get_bool_option("RADEON_DISABLE_PERFCOUNTERS", FALSE))
 650                 si_init_perfcounters(sscreen);
 651
 652         sscreen->b.has_cp_dma = true;
 653         sscreen->b.has_streamout = true;
 654
 655         if (debug_get_bool_option("RADEON_DUMP_SHADERS", FALSE))
 656                 sscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
 657
 658         /* Create the auxiliary context. This must be done last. */
 659         sscreen->b.aux_context = sscreen->b.b.context_create(&sscreen->b.b, NULL, 0);
 660
 661         return &sscreen->b.b;
 662 }