src/gallium/drivers/radeonsi/si_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 SI_PIPE_H
  27 #define SI_PIPE_H
  28
  29 #include "si_state.h"
  30
  31 #include <llvm-c/TargetMachine.h>
  32
  33 #ifdef PIPE_ARCH_BIG_ENDIAN
  34 #define SI_BIG_ENDIAN 1
  35 #else
  36 #define SI_BIG_ENDIAN 0
  37 #endif
  38
  39 /* The base vertex and primitive restart can be any number, but we must pick
  40  * one which will mean "unknown" for the purpose of state tracking and
  41  * the number shouldn't be a commonly-used one. */
  42 #define SI_BASE_VERTEX_UNKNOWN INT_MIN
  43 #define SI_RESTART_INDEX_UNKNOWN INT_MIN
  44 #define SI_NUM_SMOOTH_AA_SAMPLES 8
  45 #define SI_GS_PER_ES 128
  46
  47 /* Instruction cache. */
  48 #define SI_CONTEXT_INV_ICACHE           (R600_CONTEXT_PRIVATE_FLAG << 0)
  49 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
  50 #define SI_CONTEXT_INV_SMEM_L1          (R600_CONTEXT_PRIVATE_FLAG << 1)
  51 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
  52 #define SI_CONTEXT_INV_VMEM_L1          (R600_CONTEXT_PRIVATE_FLAG << 2)
  53 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
  54 #define SI_CONTEXT_INV_GLOBAL_L2        (R600_CONTEXT_PRIVATE_FLAG << 3)
  55 /* Framebuffer caches. */
  56 #define SI_CONTEXT_FLUSH_AND_INV_CB_META (R600_CONTEXT_PRIVATE_FLAG << 4)
  57 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 5)
  58 #define SI_CONTEXT_FLUSH_AND_INV_DB     (R600_CONTEXT_PRIVATE_FLAG << 6)
  59 #define SI_CONTEXT_FLUSH_AND_INV_CB     (R600_CONTEXT_PRIVATE_FLAG << 7)
  60 /* Engine synchronization. */
  61 #define SI_CONTEXT_VS_PARTIAL_FLUSH     (R600_CONTEXT_PRIVATE_FLAG << 8)
  62 #define SI_CONTEXT_PS_PARTIAL_FLUSH     (R600_CONTEXT_PRIVATE_FLAG << 9)
  63 #define SI_CONTEXT_CS_PARTIAL_FLUSH     (R600_CONTEXT_PRIVATE_FLAG << 10)
  64 #define SI_CONTEXT_VGT_FLUSH            (R600_CONTEXT_PRIVATE_FLAG << 11)
  65 #define SI_CONTEXT_VGT_STREAMOUT_SYNC   (R600_CONTEXT_PRIVATE_FLAG << 12)
  66 /* Compute only. */
  67 #define SI_CONTEXT_FLUSH_WITH_INV_L2    (R600_CONTEXT_PRIVATE_FLAG << 13) /* TODO: merge with TC? */
  68 #define SI_CONTEXT_FLAG_COMPUTE         (R600_CONTEXT_PRIVATE_FLAG << 14)
  69
  70 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
  71                                               SI_CONTEXT_FLUSH_AND_INV_CB_META | \
  72                                               SI_CONTEXT_FLUSH_AND_INV_DB | \
  73                                               SI_CONTEXT_FLUSH_AND_INV_DB_META)
  74
  75 #define SI_ENCODE_TRACE_POINT(id)       (0xcafe0000 | ((id) & 0xffff))
  76 #define SI_IS_TRACE_POINT(x)            (((x) & 0xcafe0000) == 0xcafe0000)
  77 #define SI_GET_TRACE_POINT_ID(x)        ((x) & 0xffff)
  78
  79 #define SI_MAX_BORDER_COLORS    4096
  80
  81 struct si_compute;
  82 struct hash_table;
  83 struct u_suballocator;
  84
  85 struct si_screen {
  86         struct r600_common_screen       b;
  87         unsigned                        gs_table_depth;
  88
  89         /* Whether shaders are monolithic (1-part) or separate (3-part). */
  90         bool                            use_monolithic_shaders;
  91
  92         pipe_mutex                      shader_parts_mutex;
  93         struct si_shader_part           *vs_prologs;
  94         struct si_shader_part           *vs_epilogs;
  95         struct si_shader_part           *tcs_epilogs;
  96         struct si_shader_part           *ps_prologs;
  97         struct si_shader_part           *ps_epilogs;
  98
  99         /* Shader cache in memory.
 100          *
 101          * Design & limitations:
 102          * - The shader cache is per screen (= per process), never saved to
 103          *   disk, and skips redundant shader compilations from TGSI to bytecode.
 104          * - It can only be used with one-variant-per-shader support, in which
 105          *   case only the main (typically middle) part of shaders is cached.
 106          * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
 107          *   variants of VS and TES are cached, so LS and ES aren't.
 108          * - GS and CS aren't cached, but it's certainly possible to cache
 109          *   those as well.
 110          */
 111         pipe_mutex                      shader_cache_mutex;
 112         struct hash_table               *shader_cache;
 113 };
 114
 115 struct si_blend_color {
 116         struct r600_atom                atom;
 117         struct pipe_blend_color         state;
 118 };
 119
 120 struct si_sampler_view {
 121         struct pipe_sampler_view        base;
 122         struct list_head                list;
 123         /* [0..7] = image descriptor
 124          * [4..7] = buffer descriptor */
 125         uint32_t                        state[8];
 126         uint32_t                        fmask_state[8];
 127         bool is_stencil_sampler;
 128 };
 129
 130 struct si_sampler_state {
 131         uint32_t                        val[4];
 132 };
 133
 134 struct si_cs_shader_state {
 135         struct si_compute               *program;
 136 };
 137
 138 struct si_textures_info {
 139         struct si_sampler_views         views;
 140         uint64_t                        depth_texture_mask; /* which textures are depth */
 141         uint64_t                        compressed_colortex_mask;
 142 };
 143
 144 struct si_images_info {
 145         struct si_descriptors           desc;
 146         struct pipe_image_view          views[SI_NUM_IMAGES];
 147         uint32_t                        compressed_colortex_mask;
 148 };
 149
 150 struct si_framebuffer {
 151         struct r600_atom                atom;
 152         struct pipe_framebuffer_state   state;
 153         unsigned                        nr_samples;
 154         unsigned                        log_samples;
 155         unsigned                        cb0_is_integer;
 156         unsigned                        compressed_cb_mask;
 157         unsigned                        spi_shader_col_format;
 158         unsigned                        spi_shader_col_format_alpha;
 159         unsigned                        spi_shader_col_format_blend;
 160         unsigned                        spi_shader_col_format_blend_alpha;
 161         unsigned                        color_is_int8; /* bitmask */
 162         unsigned                        dirty_cbufs;
 163         bool                            dirty_zsbuf;
 164 };
 165
 166 struct si_clip_state {
 167         struct r600_atom                atom;
 168         struct pipe_clip_state          state;
 169 };
 170
 171 struct si_sample_mask {
 172         struct r600_atom        atom;
 173         uint16_t                sample_mask;
 174 };
 175
 176 /* A shader state consists of the shader selector, which is a constant state
 177  * object shared by multiple contexts and shouldn't be modified, and
 178  * the current shader variant selected for this context.
 179  */
 180 struct si_shader_ctx_state {
 181         struct si_shader_selector       *cso;
 182         struct si_shader                *current;
 183 };
 184
 185 struct si_context {
 186         struct r600_common_context      b;
 187         struct blitter_context          *blitter;
 188         void                            *custom_dsa_flush;
 189         void                            *custom_blend_resolve;
 190         void                            *custom_blend_decompress;
 191         void                            *custom_blend_fastclear;
 192         void                            *custom_blend_dcc_decompress;
 193         void                            *pstipple_sampler_state;
 194         struct si_screen                *screen;
 195
 196         struct radeon_winsys_cs         *ce_ib;
 197         struct radeon_winsys_cs         *ce_preamble_ib;
 198         bool                            ce_need_synchronization;
 199         struct u_suballocator           *ce_suballocator;
 200
 201         struct pipe_fence_handle        *last_gfx_fence;
 202         struct si_shader_ctx_state      fixed_func_tcs_shader;
 203         LLVMTargetMachineRef            tm;
 204         bool                            gfx_flush_in_progress;
 205
 206         /* Atoms (direct states). */
 207         union si_state_atoms            atoms;
 208         unsigned                        dirty_atoms; /* mask */
 209         /* PM4 states (precomputed immutable states) */
 210         union si_state                  queued;
 211         union si_state                  emitted;
 212
 213         /* Atom declarations. */
 214         struct r600_atom                cache_flush;
 215         struct si_framebuffer           framebuffer;
 216         struct r600_atom                msaa_sample_locs;
 217         struct r600_atom                db_render_state;
 218         struct r600_atom                msaa_config;
 219         struct si_sample_mask           sample_mask;
 220         struct r600_atom                cb_render_state;
 221         struct si_blend_color           blend_color;
 222         struct r600_atom                clip_regs;
 223         struct si_clip_state            clip_state;
 224         struct si_shader_data           shader_userdata;
 225         struct si_stencil_ref           stencil_ref;
 226         struct r600_atom                spi_map;
 227
 228         /* Precomputed states. */
 229         struct si_pm4_state             *init_config;
 230         struct si_pm4_state             *init_config_gs_rings;
 231         bool                            init_config_has_vgt_flush;
 232         struct si_pm4_state             *vgt_shader_config[4];
 233
 234         /* shaders */
 235         struct si_shader_ctx_state      ps_shader;
 236         struct si_shader_ctx_state      gs_shader;
 237         struct si_shader_ctx_state      vs_shader;
 238         struct si_shader_ctx_state      tcs_shader;
 239         struct si_shader_ctx_state      tes_shader;
 240         struct si_cs_shader_state       cs_shader_state;
 241
 242         /* shader information */
 243         struct si_vertex_element        *vertex_elements;
 244         unsigned                        sprite_coord_enable;
 245         bool                            flatshade;
 246
 247         /* shader descriptors */
 248         struct si_descriptors           vertex_buffers;
 249         struct si_buffer_resources      const_buffers[SI_NUM_SHADERS];
 250         struct si_buffer_resources      rw_buffers[SI_NUM_SHADERS];
 251         struct si_buffer_resources      shader_buffers[SI_NUM_SHADERS];
 252         struct si_textures_info         samplers[SI_NUM_SHADERS];
 253         struct si_images_info           images[SI_NUM_SHADERS];
 254
 255         /* other shader resources */
 256         struct pipe_constant_buffer     null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
 257         struct pipe_resource            *esgs_ring;
 258         struct pipe_resource            *gsvs_ring;
 259         struct pipe_resource            *tf_ring;
 260         union pipe_color_union          *border_color_table; /* in CPU memory, any endian */
 261         struct r600_resource            *border_color_buffer;
 262         union pipe_color_union          *border_color_map; /* in VRAM (slow access), little endian */
 263         unsigned                        border_color_count;
 264
 265         /* Vertex and index buffers. */
 266         bool                            vertex_buffers_dirty;
 267         struct pipe_index_buffer        index_buffer;
 268         struct pipe_vertex_buffer       vertex_buffer[SI_NUM_VERTEX_BUFFERS];
 269
 270         /* MSAA config state. */
 271         int                             ps_iter_samples;
 272         bool                            smoothing_enabled;
 273
 274         /* DB render state. */
 275         bool                    dbcb_depth_copy_enabled;
 276         bool                    dbcb_stencil_copy_enabled;
 277         unsigned                dbcb_copy_sample;
 278         bool                    db_flush_depth_inplace;
 279         bool                    db_flush_stencil_inplace;
 280         bool                    db_depth_clear;
 281         bool                    db_depth_disable_expclear;
 282         bool                    db_stencil_clear;
 283         bool                    db_stencil_disable_expclear;
 284         unsigned                ps_db_shader_control;
 285         bool                    occlusion_queries_disabled;
 286
 287         /* Emitted draw state. */
 288         int                     last_base_vertex;
 289         int                     last_start_instance;
 290         int                     last_sh_base_reg;
 291         int                     last_primitive_restart_en;
 292         int                     last_restart_index;
 293         int                     last_gs_out_prim;
 294         int                     last_prim;
 295         int                     last_multi_vgt_param;
 296         int                     last_ls_hs_config;
 297         int                     last_rast_prim;
 298         unsigned                last_sc_line_stipple;
 299         int                     current_rast_prim; /* primitive type after TES, GS */
 300         unsigned                last_gsvs_itemsize;
 301
 302         /* Scratch buffer */
 303         struct r600_resource    *scratch_buffer;
 304         boolean                 emit_scratch_reloc;
 305         unsigned                scratch_waves;
 306         unsigned                spi_tmpring_size;
 307
 308         /* Emitted derived tessellation state. */
 309         struct si_shader        *last_ls; /* local shader (VS) */
 310         struct si_shader_selector *last_tcs;
 311         int                     last_num_tcs_input_cp;
 312         int                     last_tes_sh_base;
 313
 314         /* Debug state. */
 315         bool                    is_debug;
 316         uint32_t                *last_ib;
 317         unsigned                last_ib_dw_size;
 318         struct r600_resource    *last_trace_buf;
 319         struct r600_resource    *trace_buf;
 320         unsigned                trace_id;
 321         uint64_t                dmesg_timestamp;
 322         unsigned                last_bo_count;
 323         struct radeon_bo_list_item *last_bo_list;
 324 };
 325
 326 /* cik_sdma.c */
 327 void cik_sdma_copy(struct pipe_context *ctx,
 328                    struct pipe_resource *dst,
 329                    unsigned dst_level,
 330                    unsigned dstx, unsigned dsty, unsigned dstz,
 331                    struct pipe_resource *src,
 332                    unsigned src_level,
 333                    const struct pipe_box *src_box);
 334
 335 /* si_blit.c */
 336 void si_init_blit_functions(struct si_context *sctx);
 337 void si_decompress_textures(struct si_context *sctx);
 338 void si_resource_copy_region(struct pipe_context *ctx,
 339                              struct pipe_resource *dst,
 340                              unsigned dst_level,
 341                              unsigned dstx, unsigned dsty, unsigned dstz,
 342                              struct pipe_resource *src,
 343                              unsigned src_level,
 344                              const struct pipe_box *src_box);
 345
 346 /* si_cp_dma.c */
 347 void si_copy_buffer(struct si_context *sctx,
 348                     struct pipe_resource *dst, struct pipe_resource *src,
 349                     uint64_t dst_offset, uint64_t src_offset, unsigned size,
 350                     bool is_framebuffer);
 351 void si_init_cp_dma_functions(struct si_context *sctx);
 352
 353 /* si_debug.c */
 354 void si_init_debug_functions(struct si_context *sctx);
 355 void si_check_vm_faults(struct si_context *sctx);
 356 bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary);
 357
 358 /* si_dma.c */
 359 void si_dma_copy(struct pipe_context *ctx,
 360                  struct pipe_resource *dst,
 361                  unsigned dst_level,
 362                  unsigned dstx, unsigned dsty, unsigned dstz,
 363                  struct pipe_resource *src,
 364                  unsigned src_level,
 365                  const struct pipe_box *src_box);
 366
 367 /* si_hw_context.c */
 368 void si_context_gfx_flush(void *context, unsigned flags,
 369                           struct pipe_fence_handle **fence);
 370 void si_begin_new_cs(struct si_context *ctx);
 371 void si_need_cs_space(struct si_context *ctx);
 372
 373 /* si_compute.c */
 374 void si_init_compute_functions(struct si_context *sctx);
 375
 376 /* si_perfcounters.c */
 377 void si_init_perfcounters(struct si_screen *screen);
 378
 379 /* si_uvd.c */
 380 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
 381                                                const struct pipe_video_codec *templ);
 382
 383 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
 384                                                  const struct pipe_video_buffer *tmpl);
 385
 386 /*
 387  * common helpers
 388  */
 389
 390 static inline struct r600_resource *
 391 si_resource_create_custom(struct pipe_screen *screen,
 392                           unsigned usage, unsigned size)
 393 {
 394         assert(size);
 395         return r600_resource(pipe_buffer_create(screen,
 396                 PIPE_BIND_CUSTOM, usage, size));
 397 }
 398
 399 static inline void
 400 si_invalidate_draw_sh_constants(struct si_context *sctx)
 401 {
 402         sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
 403         sctx->last_start_instance = -1; /* reset to an unknown value */
 404         sctx->last_sh_base_reg = -1; /* reset to an unknown value */
 405 }
 406
 407 static inline void
 408 si_set_atom_dirty(struct si_context *sctx,
 409                   struct r600_atom *atom, bool dirty)
 410 {
 411         unsigned bit = 1 << (atom->id - 1);
 412
 413         if (dirty)
 414                 sctx->dirty_atoms |= bit;
 415         else
 416                 sctx->dirty_atoms &= ~bit;
 417 }
 418
 419 static inline void
 420 si_mark_atom_dirty(struct si_context *sctx,
 421                    struct r600_atom *atom)
 422 {
 423         si_set_atom_dirty(sctx, atom, true);
 424 }
 425
 426 #endif