src/mesa/drivers/dri/i965/brw_state.h

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a copy of this software and associated documentation files (the
   8  "Software"), to deal in the Software without restriction, including
   9  without limitation the rights to use, copy, modify, merge, publish,
  10  distribute, sublicense, and/or sell copies of the Software, and to
  11  permit persons to whom the Software is furnished to do so, subject to
  12  the following conditions:
  13
  14  The above copyright notice and this permission notice (including the
  15  next paragraph) shall be included in all copies or substantial
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keithw@vmware.com>
  30   */
  31
  32
  33 #ifndef BRW_STATE_H
  34 #define BRW_STATE_H
  35
  36 #include "brw_context.h"
  37
  38 #ifdef __cplusplus
  39 extern "C" {
  40 #endif
  41
  42 enum intel_msaa_layout;
  43
  44 extern const struct brw_tracked_state brw_blend_constant_color;
  45 extern const struct brw_tracked_state brw_cc_vp;
  46 extern const struct brw_tracked_state brw_cc_unit;
  47 extern const struct brw_tracked_state brw_clip_unit;
  48 extern const struct brw_tracked_state brw_vs_pull_constants;
  49 extern const struct brw_tracked_state brw_tcs_pull_constants;
  50 extern const struct brw_tracked_state brw_tes_pull_constants;
  51 extern const struct brw_tracked_state brw_gs_pull_constants;
  52 extern const struct brw_tracked_state brw_wm_pull_constants;
  53 extern const struct brw_tracked_state brw_cs_pull_constants;
  54 extern const struct brw_tracked_state brw_constant_buffer;
  55 extern const struct brw_tracked_state brw_curbe_offsets;
  56 extern const struct brw_tracked_state brw_invariant_state;
  57 extern const struct brw_tracked_state brw_fs_samplers;
  58 extern const struct brw_tracked_state brw_gs_unit;
  59 extern const struct brw_tracked_state brw_binding_table_pointers;
  60 extern const struct brw_tracked_state brw_depthbuffer;
  61 extern const struct brw_tracked_state brw_recalculate_urb_fence;
  62 extern const struct brw_tracked_state brw_sf_unit;
  63 extern const struct brw_tracked_state brw_sf_vp;
  64 extern const struct brw_tracked_state brw_vs_samplers;
  65 extern const struct brw_tracked_state brw_tcs_samplers;
  66 extern const struct brw_tracked_state brw_tes_samplers;
  67 extern const struct brw_tracked_state brw_gs_samplers;
  68 extern const struct brw_tracked_state brw_cs_samplers;
  69 extern const struct brw_tracked_state brw_cs_texture_surfaces;
  70 extern const struct brw_tracked_state brw_vs_ubo_surfaces;
  71 extern const struct brw_tracked_state brw_vs_abo_surfaces;
  72 extern const struct brw_tracked_state brw_vs_image_surfaces;
  73 extern const struct brw_tracked_state brw_tcs_ubo_surfaces;
  74 extern const struct brw_tracked_state brw_tcs_abo_surfaces;
  75 extern const struct brw_tracked_state brw_tcs_image_surfaces;
  76 extern const struct brw_tracked_state brw_tes_ubo_surfaces;
  77 extern const struct brw_tracked_state brw_tes_abo_surfaces;
  78 extern const struct brw_tracked_state brw_tes_image_surfaces;
  79 extern const struct brw_tracked_state brw_gs_ubo_surfaces;
  80 extern const struct brw_tracked_state brw_gs_abo_surfaces;
  81 extern const struct brw_tracked_state brw_gs_image_surfaces;
  82 extern const struct brw_tracked_state brw_renderbuffer_surfaces;
  83 extern const struct brw_tracked_state brw_renderbuffer_read_surfaces;
  84 extern const struct brw_tracked_state brw_texture_surfaces;
  85 extern const struct brw_tracked_state brw_wm_binding_table;
  86 extern const struct brw_tracked_state brw_gs_binding_table;
  87 extern const struct brw_tracked_state brw_tes_binding_table;
  88 extern const struct brw_tracked_state brw_tcs_binding_table;
  89 extern const struct brw_tracked_state brw_vs_binding_table;
  90 extern const struct brw_tracked_state brw_wm_ubo_surfaces;
  91 extern const struct brw_tracked_state brw_wm_abo_surfaces;
  92 extern const struct brw_tracked_state brw_wm_image_surfaces;
  93 extern const struct brw_tracked_state brw_cs_ubo_surfaces;
  94 extern const struct brw_tracked_state brw_cs_abo_surfaces;
  95 extern const struct brw_tracked_state brw_cs_image_surfaces;
  96 extern const struct brw_tracked_state brw_wm_unit;
  97
  98 extern const struct brw_tracked_state brw_psp_urb_cbs;
  99
 100 extern const struct brw_tracked_state brw_indices;
 101 extern const struct brw_tracked_state brw_index_buffer;
 102 extern const struct brw_tracked_state gen7_cs_push_constants;
 103 extern const struct brw_tracked_state gen6_binding_table_pointers;
 104 extern const struct brw_tracked_state gen6_gs_binding_table;
 105 extern const struct brw_tracked_state gen6_renderbuffer_surfaces;
 106 extern const struct brw_tracked_state gen6_sampler_state;
 107 extern const struct brw_tracked_state gen6_sol_surface;
 108 extern const struct brw_tracked_state gen6_sf_vp;
 109 extern const struct brw_tracked_state gen6_urb;
 110 extern const struct brw_tracked_state gen7_depthbuffer;
 111 extern const struct brw_tracked_state gen7_l3_state;
 112 extern const struct brw_tracked_state gen7_push_constant_space;
 113 extern const struct brw_tracked_state gen7_urb;
 114 extern const struct brw_tracked_state gen8_index_buffer;
 115 extern const struct brw_tracked_state gen8_pma_fix;
 116 extern const struct brw_tracked_state gen8_vf_topology;
 117 extern const struct brw_tracked_state brw_cs_work_groups_surface;
 118
 119 static inline bool
 120 brw_state_dirty(const struct brw_context *brw,
 121                 GLuint mesa_flags, uint64_t brw_flags)
 122 {
 123    return ((brw->NewGLState & mesa_flags) |
 124            (brw->ctx.NewDriverState & brw_flags)) != 0;
 125 }
 126
 127 /* brw_binding_tables.c */
 128 void brw_upload_binding_table(struct brw_context *brw,
 129                               uint32_t packet_name,
 130                               const struct brw_stage_prog_data *prog_data,
 131                               struct brw_stage_state *stage_state);
 132
 133 /* brw_misc_state.c */
 134 void brw_upload_invariant_state(struct brw_context *brw);
 135 uint32_t
 136 brw_depthbuffer_format(struct brw_context *brw);
 137
 138 void brw_upload_state_base_address(struct brw_context *brw);
 139
 140 /* gen8_depth_state.c */
 141 void gen8_write_pma_stall_bits(struct brw_context *brw,
 142                                uint32_t pma_stall_bits);
 143
 144 /***********************************************************************
 145  * brw_state.c
 146  */
 147 void brw_upload_render_state(struct brw_context *brw);
 148 void brw_render_state_finished(struct brw_context *brw);
 149 void brw_upload_compute_state(struct brw_context *brw);
 150 void brw_compute_state_finished(struct brw_context *brw);
 151 void brw_init_state(struct brw_context *brw);
 152 void brw_destroy_state(struct brw_context *brw);
 153 void brw_emit_select_pipeline(struct brw_context *brw,
 154                               enum brw_pipeline pipeline);
 155
 156 static inline void
 157 brw_select_pipeline(struct brw_context *brw, enum brw_pipeline pipeline)
 158 {
 159    if (unlikely(brw->last_pipeline != pipeline)) {
 160       assert(pipeline < BRW_NUM_PIPELINES);
 161       brw_emit_select_pipeline(brw, pipeline);
 162       brw->last_pipeline = pipeline;
 163    }
 164 }
 165
 166 /***********************************************************************
 167  * brw_program_cache.c
 168  */
 169
 170 void brw_upload_cache(struct brw_cache *cache,
 171                       enum brw_cache_id cache_id,
 172                       const void *key,
 173                       GLuint key_sz,
 174                       const void *data,
 175                       GLuint data_sz,
 176                       const void *aux,
 177                       GLuint aux_sz,
 178                       uint32_t *out_offset, void *out_aux);
 179
 180 bool brw_search_cache(struct brw_cache *cache,
 181                       enum brw_cache_id cache_id,
 182                       const void *key,
 183                       GLuint key_size,
 184                       uint32_t *inout_offset, void *inout_aux);
 185
 186 const void *brw_find_previous_compile(struct brw_cache *cache,
 187                                       enum brw_cache_id cache_id,
 188                                       unsigned program_string_id);
 189
 190 void brw_program_cache_check_size(struct brw_context *brw);
 191
 192 void brw_init_caches( struct brw_context *brw );
 193 void brw_destroy_caches( struct brw_context *brw );
 194
 195 void brw_print_program_cache(struct brw_context *brw);
 196
 197 /***********************************************************************
 198  * brw_state_batch.c
 199  */
 200 #define BRW_BATCH_STRUCT(brw, s) \
 201    intel_batchbuffer_data(brw, (s), sizeof(*(s)), RENDER_RING)
 202
 203 void *brw_state_batch(struct brw_context *brw,
 204                       int size, int alignment, uint32_t *out_offset);
 205 uint32_t brw_state_batch_size(struct brw_context *brw, uint32_t offset);
 206
 207 /* brw_wm_surface_state.c */
 208 void gen4_init_vtable_surface_functions(struct brw_context *brw);
 209 uint32_t brw_get_surface_tiling_bits(uint32_t tiling);
 210 uint32_t brw_get_surface_num_multisamples(unsigned num_samples);
 211
 212 uint32_t brw_isl_format_for_mesa_format(mesa_format mesa_format);
 213
 214 GLuint translate_tex_target(GLenum target);
 215
 216 GLuint translate_tex_format(struct brw_context *brw,
 217                             mesa_format mesa_format,
 218                             GLenum srgb_decode);
 219
 220 int brw_get_texture_swizzle(const struct gl_context *ctx,
 221                             const struct gl_texture_object *t);
 222
 223 void brw_emit_buffer_surface_state(struct brw_context *brw,
 224                                    uint32_t *out_offset,
 225                                    struct brw_bo *bo,
 226                                    unsigned buffer_offset,
 227                                    unsigned surface_format,
 228                                    unsigned buffer_size,
 229                                    unsigned pitch,
 230                                    bool rw);
 231
 232 void brw_update_texture_surface(struct gl_context *ctx,
 233                                 unsigned unit, uint32_t *surf_offset,
 234                                 bool for_gather, uint32_t plane);
 235
 236 uint32_t brw_update_renderbuffer_surface(struct brw_context *brw,
 237                                          struct gl_renderbuffer *rb,
 238                                          uint32_t flags, unsigned unit,
 239                                          uint32_t surf_index);
 240
 241 void brw_update_renderbuffer_surfaces(struct brw_context *brw,
 242                                       const struct gl_framebuffer *fb,
 243                                       uint32_t render_target_start,
 244                                       uint32_t *surf_offset);
 245
 246 /* gen7_wm_surface_state.c */
 247 void gen7_check_surface_setup(uint32_t *surf, bool is_render_target);
 248 void gen7_init_vtable_surface_functions(struct brw_context *brw);
 249
 250 /* gen8_surface_state.c */
 251
 252 void gen8_init_vtable_surface_functions(struct brw_context *brw);
 253
 254 /* brw_sampler_state.c */
 255 void brw_emit_sampler_state(struct brw_context *brw,
 256                             uint32_t *sampler_state,
 257                             uint32_t batch_offset_for_sampler_state,
 258                             unsigned min_filter,
 259                             unsigned mag_filter,
 260                             unsigned mip_filter,
 261                             unsigned max_anisotropy,
 262                             unsigned address_rounding,
 263                             unsigned wrap_s,
 264                             unsigned wrap_t,
 265                             unsigned wrap_r,
 266                             unsigned base_level,
 267                             unsigned min_lod,
 268                             unsigned max_lod,
 269                             int lod_bias,
 270                             unsigned shadow_function,
 271                             bool non_normalized_coordinates,
 272                             uint32_t border_color_offset);
 273
 274 /* gen6_surface_state.c */
 275 void gen6_init_vtable_surface_functions(struct brw_context *brw);
 276
 277 /* brw_vs_surface_state.c */
 278 void
 279 brw_upload_pull_constants(struct brw_context *brw,
 280                           GLbitfield64 brw_new_constbuf,
 281                           const struct gl_program *prog,
 282                           struct brw_stage_state *stage_state,
 283                           const struct brw_stage_prog_data *prog_data);
 284
 285 /* gen7_vs_state.c */
 286 void
 287 gen7_upload_constant_state(struct brw_context *brw,
 288                            const struct brw_stage_state *stage_state,
 289                            bool active, unsigned opcode);
 290
 291 /* brw_clip.c */
 292 void brw_upload_clip_prog(struct brw_context *brw);
 293
 294 /* brw_sf.c */
 295 void brw_upload_sf_prog(struct brw_context *brw);
 296
 297 bool brw_is_drawing_points(const struct brw_context *brw);
 298 bool brw_is_drawing_lines(const struct brw_context *brw);
 299
 300 /* gen7_l3_state.c */
 301 void
 302 gen7_restore_default_l3_config(struct brw_context *brw);
 303
 304 static inline bool
 305 use_state_point_size(const struct brw_context *brw)
 306 {
 307    const struct gl_context *ctx = &brw->ctx;
 308
 309    /* Section 14.4 (Points) of the OpenGL 4.5 specification says:
 310     *
 311     *    "If program point size mode is enabled, the derived point size is
 312     *     taken from the (potentially clipped) shader built-in gl_PointSize
 313     *     written by:
 314     *
 315     *        * the geometry shader, if active;
 316     *        * the tessellation evaluation shader, if active and no
 317     *          geometry shader is active;
 318     *        * the vertex shader, otherwise
 319     *
 320     *    and clamped to the implementation-dependent point size range.  If
 321     *    the value written to gl_PointSize is less than or equal to zero,
 322     *    or if no value was written to gl_PointSize, results are undefined.
 323     *    If program point size mode is disabled, the derived point size is
 324     *    specified with the command
 325     *
 326     *       void PointSize(float size);
 327     *
 328     *    size specifies the requested size of a point.  The default value
 329     *    is 1.0."
 330     *
 331     * The rules for GLES come from the ES 3.2, OES_geometry_point_size, and
 332     * OES_tessellation_point_size specifications.  To summarize: if the last
 333     * stage before rasterization is a GS or TES, then use gl_PointSize from
 334     * the shader if written.  Otherwise, use 1.0.  If the last stage is a
 335     * vertex shader, use gl_PointSize, or it is undefined.
 336     *
 337     * We can combine these rules into a single condition for both APIs.
 338     * Using the state point size when the last shader stage doesn't write
 339     * gl_PointSize satisfies GL's requirements, as it's undefined.  Because
 340     * ES doesn't have a PointSize() command, the state point size will
 341     * remain 1.0, satisfying the ES default value in the GS/TES case, and
 342     * the VS case (1.0 works for "undefined").  Mesa sets the program point
 343     * mode flag to always-enabled in ES, so we can safely check that, and
 344     * it'll be ignored for ES.
 345     *
 346     * _NEW_PROGRAM | _NEW_POINT
 347     * BRW_NEW_VUE_MAP_GEOM_OUT
 348     */
 349    return (!ctx->VertexProgram.PointSizeEnabled && !ctx->Point._Attenuated) ||
 350           (brw->vue_map_geom_out.slots_valid & VARYING_BIT_PSIZ) == 0;
 351 }
 352
 353 void brw_copy_pipeline_atoms(struct brw_context *brw,
 354                              enum brw_pipeline pipeline,
 355                              const struct brw_tracked_state **atoms,
 356                              int num_atoms);
 357 void gen4_init_atoms(struct brw_context *brw);
 358 void gen45_init_atoms(struct brw_context *brw);
 359 void gen5_init_atoms(struct brw_context *brw);
 360 void gen6_init_atoms(struct brw_context *brw);
 361 void gen7_init_atoms(struct brw_context *brw);
 362 void gen75_init_atoms(struct brw_context *brw);
 363 void gen8_init_atoms(struct brw_context *brw);
 364 void gen9_init_atoms(struct brw_context *brw);
 365
 366 void upload_gs_state_for_tf(struct brw_context *brw);
 367
 368 /* Memory Object Control State:
 369  * Specifying zero for L3 means "uncached in L3", at least on Haswell
 370  * and Baytrail, since there are no PTE flags for setting L3 cacheability.
 371  * On Ivybridge, the PTEs do have a cache-in-L3 bit, so setting MOCS to 0
 372  * may still respect that.
 373  */
 374 #define GEN7_MOCS_L3                    1
 375
 376 /* Ivybridge only: cache in LLC.
 377  * Specifying zero here means to use the PTE values set by the kernel;
 378  * non-zero overrides the PTE values.
 379  */
 380 #define IVB_MOCS_LLC                    (1 << 1)
 381
 382 /* Baytrail only: snoop in CPU cache */
 383 #define BYT_MOCS_SNOOP                  (1 << 1)
 384
 385 /* Haswell only: LLC/eLLC controls (write-back or uncached).
 386  * Specifying zero here means to use the PTE values set by the kernel,
 387  * which is useful since it offers additional control (write-through
 388  * cacheing and age).  Non-zero overrides the PTE values.
 389  */
 390 #define HSW_MOCS_UC_LLC_UC_ELLC         (1 << 1)
 391 #define HSW_MOCS_WB_LLC_WB_ELLC         (2 << 1)
 392 #define HSW_MOCS_UC_LLC_WB_ELLC         (3 << 1)
 393
 394 /* Broadwell: these defines always use all available caches (L3, LLC, eLLC),
 395  * and let you force write-back (WB) or write-through (WT) caching, or leave
 396  * it up to the page table entry (PTE) specified by the kernel.
 397  */
 398 #define BDW_MOCS_WB  0x78
 399 #define BDW_MOCS_WT  0x58
 400 #define BDW_MOCS_PTE 0x18
 401
 402 /* Skylake: MOCS is now an index into an array of 62 different caching
 403  * configurations programmed by the kernel.
 404  */
 405 /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */
 406 #define SKL_MOCS_WB  (2 << 1)
 407 /* TC=LLC/eLLC, LeCC=PTE, LRUM=3, L3CC=WB */
 408 #define SKL_MOCS_PTE (1 << 1)
 409
 410 #ifdef __cplusplus
 411 }
 412 #endif
 413
 414 #endif