src/gallium/drivers/r600/r600_pipe.h

   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  * Authors:
  24  *      Jerome Glisse
  25  */
  26 #ifndef R600_PIPE_H
  27 #define R600_PIPE_H
  28
  29 #include "util/u_slab.h"
  30 #include "r600.h"
  31 #include "r600_llvm.h"
  32 #include "r600_public.h"
  33 #include "r600_shader.h"
  34 #include "r600_resource.h"
  35 #include "evergreen_compute.h"
  36
  37 #define R600_MAX_CONST_BUFFERS 2
  38 #define R600_MAX_CONST_BUFFER_SIZE 4096
  39
  40 #ifdef PIPE_ARCH_BIG_ENDIAN
  41 #define R600_BIG_ENDIAN 1
  42 #else
  43 #define R600_BIG_ENDIAN 0
  44 #endif
  45
  46 enum r600_atom_flags {
  47         /* When set, atoms are added at the beginning of the dirty list
  48          * instead of the end. */
  49         EMIT_EARLY = (1 << 0)
  50 };
  51
  52 /* This encapsulates a state or an operation which can emitted into the GPU
  53  * command stream. It's not limited to states only, it can be used for anything
  54  * that wants to write commands into the CS (e.g. cache flushes). */
  55 struct r600_atom {
  56         void (*emit)(struct r600_context *ctx, struct r600_atom *state);
  57
  58         unsigned                num_dw;
  59         enum r600_atom_flags    flags;
  60         bool                    dirty;
  61
  62         struct list_head        head;
  63 };
  64
  65 /* This is an atom containing GPU commands that never change.
  66  * This is supposed to be copied directly into the CS. */
  67 struct r600_command_buffer {
  68         struct r600_atom atom;
  69         uint32_t *buf;
  70         unsigned max_num_dw;
  71         unsigned pkt_flags;
  72 };
  73
  74 struct r600_surface_sync_cmd {
  75         struct r600_atom atom;
  76         unsigned flush_flags; /* CP_COHER_CNTL */
  77 };
  78
  79 struct r600_db_misc_state {
  80         struct r600_atom atom;
  81         bool occlusion_query_enabled;
  82         bool flush_depthstencil_through_cb;
  83         bool copy_depth, copy_stencil;
  84 };
  85
  86 struct r600_cb_misc_state {
  87         struct r600_atom atom;
  88         unsigned cb_color_control; /* this comes from blend state */
  89         unsigned blend_colormask; /* 8*4 bits for 8 RGBA colorbuffers */
  90         unsigned nr_cbufs;
  91         unsigned nr_ps_color_outputs;
  92         bool multiwrite;
  93         bool dual_src_blend;
  94 };
  95
  96 struct r600_alphatest_state {
  97         struct r600_atom atom;
  98         unsigned sx_alpha_test_control; /* this comes from dsa state */
  99         unsigned sx_alpha_ref; /* this comes from dsa state */
 100         bool bypass;
 101         bool cb0_export_16bpc; /* from set_framebuffer_state */
 102 };
 103
 104 struct r600_cs_shader_state {
 105         struct r600_atom atom;
 106         struct r600_pipe_compute *shader;
 107 };
 108
 109 enum r600_pipe_state_id {
 110         R600_PIPE_STATE_BLEND = 0,
 111         R600_PIPE_STATE_BLEND_COLOR,
 112         R600_PIPE_STATE_CONFIG,
 113         R600_PIPE_STATE_SEAMLESS_CUBEMAP,
 114         R600_PIPE_STATE_CLIP,
 115         R600_PIPE_STATE_SCISSOR,
 116         R600_PIPE_STATE_VIEWPORT,
 117         R600_PIPE_STATE_RASTERIZER,
 118         R600_PIPE_STATE_VGT,
 119         R600_PIPE_STATE_FRAMEBUFFER,
 120         R600_PIPE_STATE_DSA,
 121         R600_PIPE_STATE_STENCIL_REF,
 122         R600_PIPE_STATE_PS_SHADER,
 123         R600_PIPE_STATE_VS_SHADER,
 124         R600_PIPE_STATE_CONSTANT,
 125         R600_PIPE_STATE_SAMPLER,
 126         R600_PIPE_STATE_RESOURCE,
 127         R600_PIPE_STATE_POLYGON_OFFSET,
 128         R600_PIPE_STATE_FETCH_SHADER,
 129         R600_PIPE_STATE_SPI,
 130         R600_PIPE_NSTATES
 131 };
 132
 133 struct compute_memory_pool;
 134 void compute_memory_pool_delete(struct compute_memory_pool* pool);
 135 struct compute_memory_pool* compute_memory_pool_new(
 136         struct r600_screen *rscreen);
 137
 138 struct r600_pipe_fences {
 139         struct r600_resource            *bo;
 140         unsigned                        *data;
 141         unsigned                        next_index;
 142         /* linked list of preallocated blocks */
 143         struct list_head                blocks;
 144         /* linked list of freed fences */
 145         struct list_head                pool;
 146         pipe_mutex                      mutex;
 147 };
 148
 149 struct r600_screen {
 150         struct pipe_screen              screen;
 151         struct radeon_winsys            *ws;
 152         unsigned                        family;
 153         enum chip_class                 chip_class;
 154         struct radeon_info              info;
 155         bool                            has_streamout;
 156         struct r600_tiling_info         tiling_info;
 157         struct r600_pipe_fences         fences;
 158
 159         /*for compute global memory binding, we allocate stuff here, instead of
 160          * buffers.
 161          * XXX: Not sure if this is the best place for global_pool.  Also,
 162          * it's not thread safe, so it won't work with multiple contexts. */
 163         struct compute_memory_pool *global_pool;
 164 };
 165
 166 struct r600_pipe_sampler_view {
 167         struct pipe_sampler_view        base;
 168         struct r600_resource            *tex_resource;
 169         uint32_t                        tex_resource_words[8];
 170 };
 171
 172 struct r600_pipe_rasterizer {
 173         struct r600_pipe_state          rstate;
 174         boolean                         flatshade;
 175         boolean                         two_side;
 176         unsigned                        sprite_coord_enable;
 177         unsigned                        clip_plane_enable;
 178         unsigned                        pa_sc_line_stipple;
 179         unsigned                        pa_cl_clip_cntl;
 180         float                           offset_units;
 181         float                           offset_scale;
 182         bool                            scissor_enable;
 183 };
 184
 185 struct r600_pipe_blend {
 186         struct r600_pipe_state          rstate;
 187         unsigned                        cb_target_mask;
 188         unsigned                        cb_color_control;
 189         bool                            dual_src_blend;
 190 };
 191
 192 struct r600_pipe_dsa {
 193         struct r600_pipe_state          rstate;
 194         unsigned                        alpha_ref;
 195         ubyte                           valuemask[2];
 196         ubyte                           writemask[2];
 197         unsigned                        sx_alpha_test_control;
 198 };
 199
 200 struct r600_vertex_element
 201 {
 202         unsigned                        count;
 203         struct pipe_vertex_element      elements[PIPE_MAX_ATTRIBS];
 204         struct r600_resource            *fetch_shader;
 205         unsigned                        fs_size;
 206         struct r600_pipe_state          rstate;
 207 };
 208
 209 struct r600_pipe_shader;
 210
 211 struct r600_pipe_shader_selector {
 212         struct r600_pipe_shader *current;
 213
 214         struct tgsi_token       *tokens;
 215         struct pipe_stream_output_info  so;
 216
 217         unsigned        num_shaders;
 218
 219         /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
 220         unsigned        type;
 221
 222         unsigned        nr_ps_max_color_exports;
 223 };
 224
 225 struct r600_pipe_shader {
 226         struct r600_pipe_shader_selector *selector;
 227         struct r600_pipe_shader *next_variant;
 228         struct r600_shader              shader;
 229         struct r600_pipe_state          rstate;
 230         struct r600_resource            *bo;
 231         struct r600_resource            *bo_fetch;
 232         struct r600_vertex_element      vertex_elements;
 233         unsigned        sprite_coord_enable;
 234         unsigned        flatshade;
 235         unsigned        pa_cl_vs_out_cntl;
 236         unsigned        nr_ps_color_outputs;
 237         unsigned        key;
 238         unsigned                db_shader_control;
 239         unsigned                ps_depth_export;
 240 };
 241
 242 struct r600_pipe_sampler_state {
 243         uint32_t                        tex_sampler_words[3];
 244         uint32_t                        border_color[4];
 245         bool                            border_color_use;
 246         bool                            seamless_cube_map;
 247 };
 248
 249 /* needed for blitter save */
 250 #define NUM_TEX_UNITS 16
 251
 252 struct r600_seamless_cube_map {
 253         struct r600_atom                atom;
 254         bool                            enabled;
 255 };
 256
 257 struct r600_samplerview_state {
 258         struct r600_atom                atom;
 259         struct r600_pipe_sampler_view   *views[NUM_TEX_UNITS];
 260         uint32_t                        enabled_mask;
 261         uint32_t                        dirty_mask;
 262         uint32_t                        depth_texture_mask; /* which textures are depth */
 263 };
 264
 265 struct r600_textures_info {
 266         struct r600_samplerview_state   views;
 267         struct r600_atom                atom_sampler;
 268         struct r600_pipe_sampler_state  *samplers[NUM_TEX_UNITS];
 269         unsigned                        n_samplers;
 270         bool                            is_array_sampler[NUM_TEX_UNITS];
 271 };
 272
 273 struct r600_fence {
 274         struct pipe_reference           reference;
 275         unsigned                        index; /* in the shared bo */
 276         struct r600_resource            *sleep_bo;
 277         struct list_head                head;
 278 };
 279
 280 #define FENCE_BLOCK_SIZE 16
 281
 282 struct r600_fence_block {
 283         struct r600_fence               fences[FENCE_BLOCK_SIZE];
 284         struct list_head                head;
 285 };
 286
 287 #define R600_CONSTANT_ARRAY_SIZE 256
 288 #define R600_RESOURCE_ARRAY_SIZE 160
 289
 290 struct r600_stencil_ref
 291 {
 292         ubyte ref_value[2];
 293         ubyte valuemask[2];
 294         ubyte writemask[2];
 295 };
 296
 297 struct r600_constbuf_state
 298 {
 299         struct r600_atom                atom;
 300         struct pipe_constant_buffer     cb[PIPE_MAX_CONSTANT_BUFFERS];
 301         uint32_t                        enabled_mask;
 302         uint32_t                        dirty_mask;
 303 };
 304
 305 struct r600_vertexbuf_state
 306 {
 307         struct r600_atom                atom;
 308         struct pipe_vertex_buffer       vb[PIPE_MAX_ATTRIBS];
 309         uint32_t                        enabled_mask; /* non-NULL buffers */
 310         uint32_t                        dirty_mask;
 311 };
 312
 313 struct r600_context {
 314         struct pipe_context             context;
 315         struct blitter_context          *blitter;
 316         enum radeon_family              family;
 317         enum chip_class                 chip_class;
 318         boolean                         has_vertex_cache;
 319         unsigned                        r6xx_num_clause_temp_gprs;
 320         void                            *custom_dsa_flush;
 321         struct r600_screen              *screen;
 322         struct radeon_winsys            *ws;
 323         struct r600_pipe_state          *states[R600_PIPE_NSTATES];
 324         struct r600_vertex_element      *vertex_elements;
 325         struct pipe_framebuffer_state   framebuffer;
 326         unsigned                        compute_cb_target_mask;
 327         unsigned                        db_shader_control;
 328         unsigned                        pa_sc_line_stipple;
 329         unsigned                        pa_cl_clip_cntl;
 330         /* for saving when using blitter */
 331         struct pipe_stencil_ref         stencil_ref;
 332         struct pipe_viewport_state      viewport;
 333         struct pipe_clip_state          clip;
 334         struct r600_pipe_shader_selector        *ps_shader;
 335         struct r600_pipe_shader_selector        *vs_shader;
 336         struct r600_pipe_rasterizer     *rasterizer;
 337         struct r600_pipe_state          vgt;
 338         struct r600_pipe_state          spi;
 339         struct pipe_query               *current_render_cond;
 340         unsigned                        current_render_cond_mode;
 341         struct pipe_query               *saved_render_cond;
 342         unsigned                        saved_render_cond_mode;
 343         /* shader information */
 344         boolean                         two_side;
 345         boolean                         spi_dirty;
 346         unsigned                        sprite_coord_enable;
 347         boolean                         flatshade;
 348         boolean                         export_16bpc;
 349         unsigned                        nr_cbufs;
 350
 351         struct u_upload_mgr             *uploader;
 352         struct util_slab_mempool        pool_transfers;
 353
 354         unsigned default_ps_gprs, default_vs_gprs;
 355
 356         /* States based on r600_atom. */
 357         struct list_head                dirty_states;
 358         struct r600_command_buffer      start_cs_cmd; /* invariant state mostly */
 359         /** Compute specific registers initializations.  The start_cs_cmd atom
 360          *  must be emitted before start_compute_cs_cmd. */
 361         struct r600_command_buffer      start_compute_cs_cmd;
 362         struct r600_surface_sync_cmd    surface_sync_cmd;
 363         struct r600_atom                r6xx_flush_and_inv_cmd;
 364         struct r600_alphatest_state     alphatest_state;
 365         struct r600_cb_misc_state       cb_misc_state;
 366         struct r600_db_misc_state       db_misc_state;
 367         /** Vertex buffers for fetch shaders */
 368         struct r600_vertexbuf_state     vertex_buffer_state;
 369         /** Vertex buffers for compute shaders */
 370         struct r600_vertexbuf_state     cs_vertex_buffer_state;
 371         struct r600_constbuf_state      vs_constbuf_state;
 372         struct r600_constbuf_state      ps_constbuf_state;
 373         struct r600_textures_info       vs_samplers;
 374         struct r600_textures_info       ps_samplers;
 375         struct r600_seamless_cube_map   seamless_cube_map;
 376         struct r600_cs_shader_state     cs_shader_state;
 377
 378         struct radeon_winsys_cs *cs;
 379
 380         struct r600_range       *range;
 381         unsigned                nblocks;
 382         struct r600_block       **blocks;
 383         struct list_head        dirty;
 384         struct list_head        enable_list;
 385         unsigned                pm4_dirty_cdwords;
 386         unsigned                ctx_pm4_ndwords;
 387
 388         /* The list of active queries. Only one query of each type can be active. */
 389         int                     num_occlusion_queries;
 390
 391         /* Manage queries in two separate groups:
 392          * The timer ones and the others (streamout, occlusion).
 393          *
 394          * We do this because we should only suspend non-timer queries for u_blitter,
 395          * and later if the non-timer queries are suspended, the context flush should
 396          * only suspend and resume the timer queries. */
 397         struct list_head        active_timer_queries;
 398         unsigned                num_cs_dw_timer_queries_suspend;
 399         struct list_head        active_nontimer_queries;
 400         unsigned                num_cs_dw_nontimer_queries_suspend;
 401
 402         unsigned                num_cs_dw_streamout_end;
 403
 404         unsigned                backend_mask;
 405         unsigned                max_db; /* for OQ */
 406         unsigned                flags;
 407         boolean                 predicate_drawing;
 408
 409         unsigned                num_so_targets;
 410         struct r600_so_target   *so_targets[PIPE_MAX_SO_BUFFERS];
 411         boolean                 streamout_start;
 412         unsigned                streamout_append_bitmask;
 413
 414         /* There is no scissor enable bit on r6xx, so we must use a workaround.
 415          * These track the current scissor state. */
 416         bool                    scissor_enable;
 417         struct pipe_scissor_state scissor_state;
 418
 419         /* With rasterizer discard, there doesn't have to be a pixel shader.
 420          * In that case, we bind this one: */
 421         void                    *dummy_pixel_shader;
 422
 423         boolean                 dual_src_blend;
 424
 425         /* Index buffer. */
 426         struct pipe_index_buffer index_buffer;
 427 };
 428
 429 static INLINE void r600_emit_atom(struct r600_context *rctx, struct r600_atom *atom)
 430 {
 431         atom->emit(rctx, atom);
 432         atom->dirty = false;
 433         if (atom->head.next && atom->head.prev)
 434                 LIST_DELINIT(&atom->head);
 435 }
 436
 437 static INLINE void r600_atom_dirty(struct r600_context *rctx, struct r600_atom *state)
 438 {
 439         if (!state->dirty) {
 440                 if (state->flags & EMIT_EARLY) {
 441                         LIST_ADD(&state->head, &rctx->dirty_states);
 442                 } else {
 443                         LIST_ADDTAIL(&state->head, &rctx->dirty_states);
 444                 }
 445                 state->dirty = true;
 446         }
 447 }
 448
 449 /* evergreen_state.c */
 450 void evergreen_init_state_functions(struct r600_context *rctx);
 451 void evergreen_init_atom_start_cs(struct r600_context *rctx);
 452 void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 453 void evergreen_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 454 void evergreen_fetch_shader(struct pipe_context *ctx, struct r600_vertex_element *ve);
 455 void *evergreen_create_db_flush_dsa(struct r600_context *rctx);
 456 void evergreen_polygon_offset_update(struct r600_context *rctx);
 457 boolean evergreen_is_format_supported(struct pipe_screen *screen,
 458                                       enum pipe_format format,
 459                                       enum pipe_texture_target target,
 460                                       unsigned sample_count,
 461                                       unsigned usage);
 462 void evergreen_init_color_surface(struct r600_context *rctx,
 463                                   struct r600_surface *surf);
 464 void evergreen_update_dual_export_state(struct r600_context * rctx);
 465
 466 /* r600_blit.c */
 467 void r600_copy_buffer(struct pipe_context *ctx, struct
 468                       pipe_resource *dst, unsigned dstx,
 469                       struct pipe_resource *src, const struct pipe_box *src_box);
 470 void r600_init_blit_functions(struct r600_context *rctx);
 471 void r600_blit_uncompress_depth(struct pipe_context *ctx,
 472                 struct r600_resource_texture *texture,
 473                 struct r600_resource_texture *staging,
 474                 unsigned first_level, unsigned last_level,
 475                 unsigned first_layer, unsigned last_layer);
 476 void r600_flush_depth_textures(struct r600_context *rctx,
 477                                struct r600_samplerview_state *textures);
 478 /* r600_buffer.c */
 479 bool r600_init_resource(struct r600_screen *rscreen,
 480                         struct r600_resource *res,
 481                         unsigned size, unsigned alignment,
 482                         unsigned bind, unsigned usage);
 483 struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 484                                          const struct pipe_resource *templ);
 485
 486 /* r600_pipe.c */
 487 void r600_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
 488                 unsigned flags);
 489
 490 /* r600_query.c */
 491 void r600_init_query_functions(struct r600_context *rctx);
 492 void r600_suspend_nontimer_queries(struct r600_context *ctx);
 493 void r600_resume_nontimer_queries(struct r600_context *ctx);
 494 void r600_suspend_timer_queries(struct r600_context *ctx);
 495 void r600_resume_timer_queries(struct r600_context *ctx);
 496
 497 /* r600_resource.c */
 498 void r600_init_context_resource_functions(struct r600_context *r600);
 499
 500 /* r600_shader.c */
 501 int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 502 #ifdef HAVE_OPENCL
 503 int r600_compute_shader_create(struct pipe_context * ctx,
 504         LLVMModuleRef mod,  struct r600_bytecode * bytecode);
 505 #endif
 506 void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 507
 508 /* r600_state.c */
 509 void r600_set_scissor_state(struct r600_context *rctx,
 510                             const struct pipe_scissor_state *state);
 511 void r600_init_state_functions(struct r600_context *rctx);
 512 void r600_init_atom_start_cs(struct r600_context *rctx);
 513 void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 514 void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 515 void r600_fetch_shader(struct pipe_context *ctx, struct r600_vertex_element *ve);
 516 void *r600_create_db_flush_dsa(struct r600_context *rctx);
 517 void r600_polygon_offset_update(struct r600_context *rctx);
 518 void r600_adjust_gprs(struct r600_context *rctx);
 519 boolean r600_is_format_supported(struct pipe_screen *screen,
 520                                  enum pipe_format format,
 521                                  enum pipe_texture_target target,
 522                                  unsigned sample_count,
 523                                  unsigned usage);
 524 void r600_update_dual_export_state(struct r600_context * rctx);
 525
 526 /* r600_texture.c */
 527 void r600_init_screen_texture_functions(struct pipe_screen *screen);
 528 void r600_init_surface_functions(struct r600_context *r600);
 529 uint32_t r600_translate_texformat(struct pipe_screen *screen, enum pipe_format format,
 530                                   const unsigned char *swizzle_view,
 531                                   uint32_t *word4_p, uint32_t *yuv_format_p);
 532 unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
 533                                         unsigned level, unsigned layer);
 534
 535 /* r600_translate.c */
 536 void r600_translate_index_buffer(struct r600_context *r600,
 537                                  struct pipe_index_buffer *ib,
 538                                  unsigned count);
 539
 540 /* r600_state_common.c */
 541 void r600_init_atom(struct r600_atom *atom,
 542                     void (*emit)(struct r600_context *ctx, struct r600_atom *state),
 543                     unsigned num_dw, enum r600_atom_flags flags);
 544 void r600_init_common_atoms(struct r600_context *rctx);
 545 unsigned r600_get_cb_flush_flags(struct r600_context *rctx);
 546 void r600_texture_barrier(struct pipe_context *ctx);
 547 void r600_set_index_buffer(struct pipe_context *ctx,
 548                            const struct pipe_index_buffer *ib);
 549 void r600_vertex_buffers_dirty(struct r600_context *rctx);
 550 void r600_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
 551                              const struct pipe_vertex_buffer *input);
 552 void r600_sampler_views_dirty(struct r600_context *rctx,
 553                               struct r600_samplerview_state *state);
 554 void r600_set_sampler_views(struct r600_context *rctx,
 555                             struct r600_textures_info *dst,
 556                             unsigned count,
 557                             struct pipe_sampler_view **views);
 558 void r600_bind_vs_samplers(struct pipe_context *ctx, unsigned count, void **states);
 559 void r600_bind_ps_samplers(struct pipe_context *ctx, unsigned count, void **states);
 560 void *r600_create_vertex_elements(struct pipe_context *ctx,
 561                                   unsigned count,
 562                                   const struct pipe_vertex_element *elements);
 563 void r600_delete_vertex_element(struct pipe_context *ctx, void *state);
 564 void r600_bind_blend_state(struct pipe_context *ctx, void *state);
 565 void r600_set_blend_color(struct pipe_context *ctx,
 566                           const struct pipe_blend_color *state);
 567 void r600_bind_dsa_state(struct pipe_context *ctx, void *state);
 568 void r600_set_max_scissor(struct r600_context *rctx);
 569 void r600_bind_rs_state(struct pipe_context *ctx, void *state);
 570 void r600_delete_rs_state(struct pipe_context *ctx, void *state);
 571 void r600_sampler_view_destroy(struct pipe_context *ctx,
 572                                struct pipe_sampler_view *state);
 573 void r600_delete_sampler(struct pipe_context *ctx, void *state);
 574 void r600_delete_state(struct pipe_context *ctx, void *state);
 575 void r600_bind_vertex_elements(struct pipe_context *ctx, void *state);
 576 void *r600_create_shader_state_ps(struct pipe_context *ctx,
 577                    const struct pipe_shader_state *state);
 578 void *r600_create_shader_state_vs(struct pipe_context *ctx,
 579                    const struct pipe_shader_state *state);
 580 void r600_bind_ps_shader(struct pipe_context *ctx, void *state);
 581 void r600_bind_vs_shader(struct pipe_context *ctx, void *state);
 582 void r600_delete_ps_shader(struct pipe_context *ctx, void *state);
 583 void r600_delete_vs_shader(struct pipe_context *ctx, void *state);
 584 void r600_constant_buffers_dirty(struct r600_context *rctx, struct r600_constbuf_state *state);
 585 void r600_set_constant_buffer(struct pipe_context *ctx, uint shader, uint index,
 586                               struct pipe_constant_buffer *cb);
 587 struct pipe_stream_output_target *
 588 r600_create_so_target(struct pipe_context *ctx,
 589                       struct pipe_resource *buffer,
 590                       unsigned buffer_offset,
 591                       unsigned buffer_size);
 592 void r600_so_target_destroy(struct pipe_context *ctx,
 593                             struct pipe_stream_output_target *target);
 594 void r600_set_so_targets(struct pipe_context *ctx,
 595                          unsigned num_targets,
 596                          struct pipe_stream_output_target **targets,
 597                          unsigned append_bitmask);
 598 void r600_set_pipe_stencil_ref(struct pipe_context *ctx,
 599                                const struct pipe_stencil_ref *state);
 600 void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info);
 601 uint32_t r600_translate_stencil_op(int s_op);
 602 uint32_t r600_translate_fill(uint32_t func);
 603 unsigned r600_tex_wrap(unsigned wrap);
 604 unsigned r600_tex_filter(unsigned filter);
 605 unsigned r600_tex_mipfilter(unsigned filter);
 606 unsigned r600_tex_compare(unsigned compare);
 607
 608 /*
 609  * Helpers for building command buffers
 610  */
 611
 612 #define PKT3_SET_CONFIG_REG     0x68
 613 #define PKT3_SET_CONTEXT_REG    0x69
 614 #define PKT3_SET_CTL_CONST      0x6F
 615 #define PKT3_SET_LOOP_CONST                    0x6C
 616
 617 #define R600_CONFIG_REG_OFFSET  0x08000
 618 #define R600_CONTEXT_REG_OFFSET 0x28000
 619 #define R600_CTL_CONST_OFFSET   0x3CFF0
 620 #define R600_LOOP_CONST_OFFSET                 0X0003E200
 621 #define EG_LOOP_CONST_OFFSET               0x0003A200
 622
 623 #define PKT_TYPE_S(x)                   (((x) & 0x3) << 30)
 624 #define PKT_COUNT_S(x)                  (((x) & 0x3FFF) << 16)
 625 #define PKT3_IT_OPCODE_S(x)             (((x) & 0xFF) << 8)
 626 #define PKT3_PREDICATE(x)               (((x) >> 0) & 0x1)
 627 #define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT_COUNT_S(count) | PKT3_IT_OPCODE_S(op) | PKT3_PREDICATE(predicate))
 628
 629 #define RADEON_CP_PACKET3_COMPUTE_MODE 0x00000002
 630
 631 /*Evergreen Compute packet3*/
 632 #define PKT3C(op, count, predicate) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count) | PKT3_PREDICATE(predicate) | RADEON_CP_PACKET3_COMPUTE_MODE)
 633
 634 static INLINE void r600_store_value(struct r600_command_buffer *cb, unsigned value)
 635 {
 636         cb->buf[cb->atom.num_dw++] = value;
 637 }
 638
 639 static INLINE void r600_store_config_reg_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
 640 {
 641         assert(reg < R600_CONTEXT_REG_OFFSET);
 642         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
 643         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CONFIG_REG, num, 0);
 644         cb->buf[cb->atom.num_dw++] = (reg - R600_CONFIG_REG_OFFSET) >> 2;
 645 }
 646
 647 /**
 648  * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
 649  * shaders.
 650  */
 651 static INLINE void r600_store_context_reg_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
 652 {
 653         assert(reg >= R600_CONTEXT_REG_OFFSET && reg < R600_CTL_CONST_OFFSET);
 654         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
 655         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CONTEXT_REG, num, 0) | cb->pkt_flags;
 656         cb->buf[cb->atom.num_dw++] = (reg - R600_CONTEXT_REG_OFFSET) >> 2;
 657 }
 658
 659 /**
 660  * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
 661  * shaders.
 662  */
 663 static INLINE void r600_store_ctl_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
 664 {
 665         assert(reg >= R600_CTL_CONST_OFFSET);
 666         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
 667         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CTL_CONST, num, 0) | cb->pkt_flags;
 668         cb->buf[cb->atom.num_dw++] = (reg - R600_CTL_CONST_OFFSET) >> 2;
 669 }
 670
 671 static INLINE void r600_store_loop_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
 672 {
 673         assert(reg >= R600_LOOP_CONST_OFFSET);
 674         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
 675         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_LOOP_CONST, num, 0);
 676         cb->buf[cb->atom.num_dw++] = (reg - R600_LOOP_CONST_OFFSET) >> 2;
 677 }
 678
 679 /**
 680  * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
 681  * shaders.
 682  */
 683 static INLINE void eg_store_loop_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
 684 {
 685         assert(reg >= EG_LOOP_CONST_OFFSET);
 686         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
 687         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_LOOP_CONST, num, 0) | cb->pkt_flags;
 688         cb->buf[cb->atom.num_dw++] = (reg - EG_LOOP_CONST_OFFSET) >> 2;
 689 }
 690
 691 static INLINE void r600_store_config_reg(struct r600_command_buffer *cb, unsigned reg, unsigned value)
 692 {
 693         r600_store_config_reg_seq(cb, reg, 1);
 694         r600_store_value(cb, value);
 695 }
 696
 697 static INLINE void r600_store_context_reg(struct r600_command_buffer *cb, unsigned reg, unsigned value)
 698 {
 699         r600_store_context_reg_seq(cb, reg, 1);
 700         r600_store_value(cb, value);
 701 }
 702
 703 static INLINE void r600_store_ctl_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
 704 {
 705         r600_store_ctl_const_seq(cb, reg, 1);
 706         r600_store_value(cb, value);
 707 }
 708
 709 static INLINE void r600_store_loop_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
 710 {
 711         r600_store_loop_const_seq(cb, reg, 1);
 712         r600_store_value(cb, value);
 713 }
 714
 715 static INLINE void eg_store_loop_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
 716 {
 717         eg_store_loop_const_seq(cb, reg, 1);
 718         r600_store_value(cb, value);
 719 }
 720
 721 void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw, enum r600_atom_flags flags);
 722 void r600_release_command_buffer(struct r600_command_buffer *cb);
 723
 724 /*
 725  * Helpers for emitting state into a command stream directly.
 726  */
 727
 728 static INLINE unsigned r600_context_bo_reloc(struct r600_context *ctx, struct r600_resource *rbo,
 729                                              enum radeon_bo_usage usage)
 730 {
 731         assert(usage);
 732         return ctx->ws->cs_add_reloc(ctx->cs, rbo->cs_buf, usage, rbo->domains) * 4;
 733 }
 734
 735 static INLINE void r600_write_value(struct radeon_winsys_cs *cs, unsigned value)
 736 {
 737         cs->buf[cs->cdw++] = value;
 738 }
 739
 740 static INLINE void r600_write_array(struct radeon_winsys_cs *cs, unsigned num, unsigned *ptr)
 741 {
 742         assert(cs->cdw+num <= RADEON_MAX_CMDBUF_DWORDS);
 743         memcpy(&cs->buf[cs->cdw], ptr, num * sizeof(ptr[0]));
 744         cs->cdw += num;
 745 }
 746
 747 static INLINE void r600_write_config_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
 748 {
 749         assert(reg < R600_CONTEXT_REG_OFFSET);
 750         assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
 751         cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONFIG_REG, num, 0);
 752         cs->buf[cs->cdw++] = (reg - R600_CONFIG_REG_OFFSET) >> 2;
 753 }
 754
 755 static INLINE void r600_write_context_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
 756 {
 757         assert(reg >= R600_CONTEXT_REG_OFFSET && reg < R600_CTL_CONST_OFFSET);
 758         assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
 759         cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, num, 0);
 760         cs->buf[cs->cdw++] = (reg - R600_CONTEXT_REG_OFFSET) >> 2;
 761 }
 762
 763 static INLINE void r600_write_compute_context_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
 764 {
 765         r600_write_context_reg_seq(cs, reg, num);
 766         /* Set the compute bit on the packet header */
 767         cs->buf[cs->cdw - 2] |= RADEON_CP_PACKET3_COMPUTE_MODE;
 768 }
 769
 770 static INLINE void r600_write_ctl_const_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
 771 {
 772         assert(reg >= R600_CTL_CONST_OFFSET);
 773         assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
 774         cs->buf[cs->cdw++] = PKT3(PKT3_SET_CTL_CONST, num, 0);
 775         cs->buf[cs->cdw++] = (reg - R600_CTL_CONST_OFFSET) >> 2;
 776 }
 777
 778 static INLINE void r600_write_config_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
 779 {
 780         r600_write_config_reg_seq(cs, reg, 1);
 781         r600_write_value(cs, value);
 782 }
 783
 784 static INLINE void r600_write_context_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
 785 {
 786         r600_write_context_reg_seq(cs, reg, 1);
 787         r600_write_value(cs, value);
 788 }
 789
 790 static INLINE void r600_write_compute_context_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
 791 {
 792         r600_write_compute_context_reg_seq(cs, reg, 1);
 793         r600_write_value(cs, value);
 794 }
 795
 796 static INLINE void r600_write_ctl_const(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
 797 {
 798         r600_write_ctl_const_seq(cs, reg, 1);
 799         r600_write_value(cs, value);
 800 }
 801
 802 /*
 803  * common helpers
 804  */
 805 static INLINE uint32_t S_FIXED(float value, uint32_t frac_bits)
 806 {
 807         return value * (1 << frac_bits);
 808 }
 809 #define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
 810
 811 static inline unsigned r600_tex_aniso_filter(unsigned filter)
 812 {
 813         if (filter <= 1)   return 0;
 814         if (filter <= 2)   return 1;
 815         if (filter <= 4)   return 2;
 816         if (filter <= 8)   return 3;
 817          /* else */        return 4;
 818 }
 819
 820 /* 12.4 fixed-point */
 821 static INLINE unsigned r600_pack_float_12p4(float x)
 822 {
 823         return x <= 0    ? 0 :
 824                x >= 4096 ? 0xffff : x * 16;
 825 }
 826
 827 static INLINE uint64_t r600_resource_va(struct pipe_screen *screen, struct pipe_resource *resource)
 828 {
 829         struct r600_screen *rscreen = (struct r600_screen*)screen;
 830         struct r600_resource *rresource = (struct r600_resource*)resource;
 831
 832         return rscreen->ws->buffer_get_virtual_address(rresource->cs_buf);
 833 }
 834
 835 #endif