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_clip_unit;
  46 extern const struct brw_tracked_state brw_vs_pull_constants;
  47 extern const struct brw_tracked_state brw_tcs_pull_constants;
  48 extern const struct brw_tracked_state brw_tes_pull_constants;
  49 extern const struct brw_tracked_state brw_gs_pull_constants;
  50 extern const struct brw_tracked_state brw_wm_pull_constants;
  51 extern const struct brw_tracked_state brw_cs_pull_constants;
  52 extern const struct brw_tracked_state brw_constant_buffer;
  53 extern const struct brw_tracked_state brw_curbe_offsets;
  54 extern const struct brw_tracked_state brw_invariant_state;
  55 extern const struct brw_tracked_state brw_binding_table_pointers;
  56 extern const struct brw_tracked_state brw_depthbuffer;
  57 extern const struct brw_tracked_state brw_recalculate_urb_fence;
  58 extern const struct brw_tracked_state brw_sf_vp;
  59 extern const struct brw_tracked_state brw_cs_texture_surfaces;
  60 extern const struct brw_tracked_state brw_vs_ubo_surfaces;
  61 extern const struct brw_tracked_state brw_vs_abo_surfaces;
  62 extern const struct brw_tracked_state brw_vs_image_surfaces;
  63 extern const struct brw_tracked_state brw_tcs_ubo_surfaces;
  64 extern const struct brw_tracked_state brw_tcs_abo_surfaces;
  65 extern const struct brw_tracked_state brw_tcs_image_surfaces;
  66 extern const struct brw_tracked_state brw_tes_ubo_surfaces;
  67 extern const struct brw_tracked_state brw_tes_abo_surfaces;
  68 extern const struct brw_tracked_state brw_tes_image_surfaces;
  69 extern const struct brw_tracked_state brw_gs_ubo_surfaces;
  70 extern const struct brw_tracked_state brw_gs_abo_surfaces;
  71 extern const struct brw_tracked_state brw_gs_image_surfaces;
  72 extern const struct brw_tracked_state brw_renderbuffer_surfaces;
  73 extern const struct brw_tracked_state brw_renderbuffer_read_surfaces;
  74 extern const struct brw_tracked_state brw_texture_surfaces;
  75 extern const struct brw_tracked_state brw_wm_binding_table;
  76 extern const struct brw_tracked_state brw_gs_binding_table;
  77 extern const struct brw_tracked_state brw_tes_binding_table;
  78 extern const struct brw_tracked_state brw_tcs_binding_table;
  79 extern const struct brw_tracked_state brw_vs_binding_table;
  80 extern const struct brw_tracked_state brw_wm_ubo_surfaces;
  81 extern const struct brw_tracked_state brw_wm_abo_surfaces;
  82 extern const struct brw_tracked_state brw_wm_image_surfaces;
  83 extern const struct brw_tracked_state brw_cs_ubo_surfaces;
  84 extern const struct brw_tracked_state brw_cs_abo_surfaces;
  85 extern const struct brw_tracked_state brw_cs_image_surfaces;
  86
  87 extern const struct brw_tracked_state brw_psp_urb_cbs;
  88
  89 extern const struct brw_tracked_state brw_indices;
  90 extern const struct brw_tracked_state brw_index_buffer;
  91 extern const struct brw_tracked_state gen7_cs_push_constants;
  92 extern const struct brw_tracked_state gen6_binding_table_pointers;
  93 extern const struct brw_tracked_state gen6_gs_binding_table;
  94 extern const struct brw_tracked_state gen6_renderbuffer_surfaces;
  95 extern const struct brw_tracked_state gen6_sampler_state;
  96 extern const struct brw_tracked_state gen6_sol_surface;
  97 extern const struct brw_tracked_state gen6_sf_vp;
  98 extern const struct brw_tracked_state gen6_urb;
  99 extern const struct brw_tracked_state gen7_depthbuffer;
 100 extern const struct brw_tracked_state gen7_l3_state;
 101 extern const struct brw_tracked_state gen7_push_constant_space;
 102 extern const struct brw_tracked_state gen7_urb;
 103 extern const struct brw_tracked_state gen8_pma_fix;
 104 extern const struct brw_tracked_state brw_cs_work_groups_surface;
 105
 106 static inline bool
 107 brw_state_dirty(const struct brw_context *brw,
 108                 GLuint mesa_flags, uint64_t brw_flags)
 109 {
 110    return ((brw->NewGLState & mesa_flags) |
 111            (brw->ctx.NewDriverState & brw_flags)) != 0;
 112 }
 113
 114 /* brw_binding_tables.c */
 115 void brw_upload_binding_table(struct brw_context *brw,
 116                               uint32_t packet_name,
 117                               const struct brw_stage_prog_data *prog_data,
 118                               struct brw_stage_state *stage_state);
 119
 120 /* brw_misc_state.c */
 121 void brw_upload_invariant_state(struct brw_context *brw);
 122 uint32_t
 123 brw_depthbuffer_format(struct brw_context *brw);
 124
 125 uint32_t
 126 brw_convert_depth_value(mesa_format format, float value);
 127
 128 void brw_upload_state_base_address(struct brw_context *brw);
 129
 130 /* gen8_depth_state.c */
 131 void gen8_write_pma_stall_bits(struct brw_context *brw,
 132                                uint32_t pma_stall_bits);
 133
 134 /* brw_disk_cache.c */
 135 bool brw_disk_cache_upload_program(struct brw_context *brw,
 136                                    gl_shader_stage stage);
 137 void brw_disk_cache_write_program(struct brw_context *brw);
 138
 139 /***********************************************************************
 140  * brw_state.c
 141  */
 142 void brw_upload_render_state(struct brw_context *brw);
 143 void brw_render_state_finished(struct brw_context *brw);
 144 void brw_upload_compute_state(struct brw_context *brw);
 145 void brw_compute_state_finished(struct brw_context *brw);
 146 void brw_init_state(struct brw_context *brw);
 147 void brw_destroy_state(struct brw_context *brw);
 148 void brw_emit_select_pipeline(struct brw_context *brw,
 149                               enum brw_pipeline pipeline);
 150
 151 static inline void
 152 brw_select_pipeline(struct brw_context *brw, enum brw_pipeline pipeline)
 153 {
 154    if (unlikely(brw->last_pipeline != pipeline)) {
 155       assert(pipeline < BRW_NUM_PIPELINES);
 156       brw_emit_select_pipeline(brw, pipeline);
 157       brw->last_pipeline = pipeline;
 158    }
 159 }
 160
 161 /***********************************************************************
 162  * brw_program_cache.c
 163  */
 164
 165 void brw_upload_cache(struct brw_cache *cache,
 166                       enum brw_cache_id cache_id,
 167                       const void *key,
 168                       GLuint key_sz,
 169                       const void *data,
 170                       GLuint data_sz,
 171                       const void *aux,
 172                       GLuint aux_sz,
 173                       uint32_t *out_offset, void *out_aux);
 174
 175 bool brw_search_cache(struct brw_cache *cache,
 176                       enum brw_cache_id cache_id,
 177                       const void *key,
 178                       GLuint key_size,
 179                       uint32_t *inout_offset, void *inout_aux);
 180
 181 const void *brw_find_previous_compile(struct brw_cache *cache,
 182                                       enum brw_cache_id cache_id,
 183                                       unsigned program_string_id);
 184
 185 void brw_program_cache_check_size(struct brw_context *brw);
 186
 187 void brw_init_caches( struct brw_context *brw );
 188 void brw_destroy_caches( struct brw_context *brw );
 189
 190 void brw_print_program_cache(struct brw_context *brw);
 191
 192 /* intel_batchbuffer.c */
 193 void brw_require_statebuffer_space(struct brw_context *brw, int size);
 194 void *brw_state_batch(struct brw_context *brw,
 195                       int size, int alignment, uint32_t *out_offset);
 196 uint32_t brw_state_batch_size(struct brw_context *brw, uint32_t offset);
 197
 198 /* brw_wm_surface_state.c */
 199 uint32_t brw_get_surface_tiling_bits(uint32_t tiling);
 200 uint32_t brw_get_surface_num_multisamples(unsigned num_samples);
 201 enum isl_format brw_isl_format_for_mesa_format(mesa_format mesa_format);
 202
 203 GLuint translate_tex_target(GLenum target);
 204
 205 enum isl_format translate_tex_format(struct brw_context *brw,
 206                                      mesa_format mesa_format,
 207                                      GLenum srgb_decode);
 208
 209 int brw_get_texture_swizzle(const struct gl_context *ctx,
 210                             const struct gl_texture_object *t);
 211
 212 void brw_emit_buffer_surface_state(struct brw_context *brw,
 213                                    uint32_t *out_offset,
 214                                    struct brw_bo *bo,
 215                                    unsigned buffer_offset,
 216                                    unsigned surface_format,
 217                                    unsigned buffer_size,
 218                                    unsigned pitch,
 219                                    unsigned reloc_flags);
 220
 221 /* brw_sampler_state.c */
 222 void brw_emit_sampler_state(struct brw_context *brw,
 223                             uint32_t *sampler_state,
 224                             uint32_t batch_offset_for_sampler_state,
 225                             unsigned min_filter,
 226                             unsigned mag_filter,
 227                             unsigned mip_filter,
 228                             unsigned max_anisotropy,
 229                             unsigned address_rounding,
 230                             unsigned wrap_s,
 231                             unsigned wrap_t,
 232                             unsigned wrap_r,
 233                             unsigned base_level,
 234                             unsigned min_lod,
 235                             unsigned max_lod,
 236                             int lod_bias,
 237                             unsigned shadow_function,
 238                             bool non_normalized_coordinates,
 239                             uint32_t border_color_offset);
 240
 241 /* gen6_constant_state.c */
 242 void
 243 brw_populate_constant_data(struct brw_context *brw,
 244                            const struct gl_program *prog,
 245                            const struct brw_stage_state *stage_state,
 246                            void *dst,
 247                            const uint32_t *param,
 248                            unsigned nr_params);
 249 void
 250 brw_upload_pull_constants(struct brw_context *brw,
 251                           GLbitfield64 brw_new_constbuf,
 252                           const struct gl_program *prog,
 253                           struct brw_stage_state *stage_state,
 254                           const struct brw_stage_prog_data *prog_data);
 255 void
 256 brw_upload_cs_push_constants(struct brw_context *brw,
 257                              const struct gl_program *prog,
 258                              const struct brw_cs_prog_data *cs_prog_data,
 259                              struct brw_stage_state *stage_state);
 260
 261 /* gen7_vs_state.c */
 262 void
 263 gen7_upload_constant_state(struct brw_context *brw,
 264                            const struct brw_stage_state *stage_state,
 265                            bool active, unsigned opcode);
 266
 267 /* brw_clip.c */
 268 void brw_upload_clip_prog(struct brw_context *brw);
 269
 270 /* brw_sf.c */
 271 void brw_upload_sf_prog(struct brw_context *brw);
 272
 273 bool brw_is_drawing_points(const struct brw_context *brw);
 274 bool brw_is_drawing_lines(const struct brw_context *brw);
 275
 276 /* gen7_l3_state.c */
 277 void
 278 gen7_restore_default_l3_config(struct brw_context *brw);
 279
 280 static inline bool
 281 use_state_point_size(const struct brw_context *brw)
 282 {
 283    const struct gl_context *ctx = &brw->ctx;
 284
 285    /* Section 14.4 (Points) of the OpenGL 4.5 specification says:
 286     *
 287     *    "If program point size mode is enabled, the derived point size is
 288     *     taken from the (potentially clipped) shader built-in gl_PointSize
 289     *     written by:
 290     *
 291     *        * the geometry shader, if active;
 292     *        * the tessellation evaluation shader, if active and no
 293     *          geometry shader is active;
 294     *        * the vertex shader, otherwise
 295     *
 296     *    and clamped to the implementation-dependent point size range.  If
 297     *    the value written to gl_PointSize is less than or equal to zero,
 298     *    or if no value was written to gl_PointSize, results are undefined.
 299     *    If program point size mode is disabled, the derived point size is
 300     *    specified with the command
 301     *
 302     *       void PointSize(float size);
 303     *
 304     *    size specifies the requested size of a point.  The default value
 305     *    is 1.0."
 306     *
 307     * The rules for GLES come from the ES 3.2, OES_geometry_point_size, and
 308     * OES_tessellation_point_size specifications.  To summarize: if the last
 309     * stage before rasterization is a GS or TES, then use gl_PointSize from
 310     * the shader if written.  Otherwise, use 1.0.  If the last stage is a
 311     * vertex shader, use gl_PointSize, or it is undefined.
 312     *
 313     * We can combine these rules into a single condition for both APIs.
 314     * Using the state point size when the last shader stage doesn't write
 315     * gl_PointSize satisfies GL's requirements, as it's undefined.  Because
 316     * ES doesn't have a PointSize() command, the state point size will
 317     * remain 1.0, satisfying the ES default value in the GS/TES case, and
 318     * the VS case (1.0 works for "undefined").  Mesa sets the program point
 319     * mode flag to always-enabled in ES, so we can safely check that, and
 320     * it'll be ignored for ES.
 321     *
 322     * _NEW_PROGRAM | _NEW_POINT
 323     * BRW_NEW_VUE_MAP_GEOM_OUT
 324     */
 325    return (!ctx->VertexProgram.PointSizeEnabled && !ctx->Point._Attenuated) ||
 326           (brw->vue_map_geom_out.slots_valid & VARYING_BIT_PSIZ) == 0;
 327 }
 328
 329 void brw_copy_pipeline_atoms(struct brw_context *brw,
 330                              enum brw_pipeline pipeline,
 331                              const struct brw_tracked_state **atoms,
 332                              int num_atoms);
 333 void gen4_init_atoms(struct brw_context *brw);
 334 void gen45_init_atoms(struct brw_context *brw);
 335 void gen5_init_atoms(struct brw_context *brw);
 336 void gen6_init_atoms(struct brw_context *brw);
 337 void gen7_init_atoms(struct brw_context *brw);
 338 void gen75_init_atoms(struct brw_context *brw);
 339 void gen8_init_atoms(struct brw_context *brw);
 340 void gen9_init_atoms(struct brw_context *brw);
 341 void gen10_init_atoms(struct brw_context *brw);
 342
 343 /* Memory Object Control State:
 344  * Specifying zero for L3 means "uncached in L3", at least on Haswell
 345  * and Baytrail, since there are no PTE flags for setting L3 cacheability.
 346  * On Ivybridge, the PTEs do have a cache-in-L3 bit, so setting MOCS to 0
 347  * may still respect that.
 348  */
 349 #define GEN7_MOCS_L3                    1
 350
 351 /* Ivybridge only: cache in LLC.
 352  * Specifying zero here means to use the PTE values set by the kernel;
 353  * non-zero overrides the PTE values.
 354  */
 355 #define IVB_MOCS_LLC                    (1 << 1)
 356
 357 /* Baytrail only: snoop in CPU cache */
 358 #define BYT_MOCS_SNOOP                  (1 << 1)
 359
 360 /* Haswell only: LLC/eLLC controls (write-back or uncached).
 361  * Specifying zero here means to use the PTE values set by the kernel,
 362  * which is useful since it offers additional control (write-through
 363  * cacheing and age).  Non-zero overrides the PTE values.
 364  */
 365 #define HSW_MOCS_UC_LLC_UC_ELLC         (1 << 1)
 366 #define HSW_MOCS_WB_LLC_WB_ELLC         (2 << 1)
 367 #define HSW_MOCS_UC_LLC_WB_ELLC         (3 << 1)
 368
 369 /* Broadwell: these defines always use all available caches (L3, LLC, eLLC),
 370  * and let you force write-back (WB) or write-through (WT) caching, or leave
 371  * it up to the page table entry (PTE) specified by the kernel.
 372  */
 373 #define BDW_MOCS_WB  0x78
 374 #define BDW_MOCS_WT  0x58
 375 #define BDW_MOCS_PTE 0x18
 376
 377 /* Skylake: MOCS is now an index into an array of 62 different caching
 378  * configurations programmed by the kernel.
 379  */
 380 /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */
 381 #define SKL_MOCS_WB  (2 << 1)
 382 /* TC=LLC/eLLC, LeCC=PTE, LRUM=3, L3CC=WB */
 383 #define SKL_MOCS_PTE (1 << 1)
 384
 385 /* Cannonlake: MOCS is now an index into an array of 62 different caching
 386  * configurations programmed by the kernel.
 387  */
 388 /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */
 389 #define CNL_MOCS_WB  (2 << 1)
 390 /* TC=LLC/eLLC, LeCC=PTE, LRUM=3, L3CC=WB */
 391 #define CNL_MOCS_PTE (1 << 1)
 392
 393 #ifdef __cplusplus
 394 }
 395 #endif
 396
 397 #endif