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
  46 #define SI_TRACE_CS_DWORDS              7
  47
  48 #define SI_MAX_DRAW_CS_DWORDS \
  49         (/*scratch:*/ 3 + /*derived prim state:*/ 3 + \
  50          /*draw regs:*/ 18 + /*draw packets:*/ 31 +\
  51          /*derived tess state:*/ 19)
  52
  53 /* Instruction cache. */
  54 #define SI_CONTEXT_INV_ICACHE           (R600_CONTEXT_PRIVATE_FLAG << 0)
  55 /* Cache used by scalar memory (SMEM) instructions. They also use TC
  56  * as a second level cache, which isn't flushed by this.
  57  * Other names: constant cache, data cache, DCACHE */
  58 #define SI_CONTEXT_INV_KCACHE           (R600_CONTEXT_PRIVATE_FLAG << 1)
  59 /* Caches used by vector memory (VMEM) instructions.
  60  * L1 can optionally be bypassed (GLC=1) and can only be used by shaders.
  61  * L2 is used by shaders and can be used by other blocks (CP, sDMA). */
  62 #define SI_CONTEXT_INV_TC_L1            (R600_CONTEXT_PRIVATE_FLAG << 2)
  63 #define SI_CONTEXT_INV_TC_L2            (R600_CONTEXT_PRIVATE_FLAG << 3)
  64 /* Framebuffer caches. */
  65 #define SI_CONTEXT_FLUSH_AND_INV_CB_META (R600_CONTEXT_PRIVATE_FLAG << 4)
  66 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 5)
  67 #define SI_CONTEXT_FLUSH_AND_INV_DB     (R600_CONTEXT_PRIVATE_FLAG << 6)
  68 #define SI_CONTEXT_FLUSH_AND_INV_CB     (R600_CONTEXT_PRIVATE_FLAG << 7)
  69 /* Engine synchronization. */
  70 #define SI_CONTEXT_VS_PARTIAL_FLUSH     (R600_CONTEXT_PRIVATE_FLAG << 8)
  71 #define SI_CONTEXT_PS_PARTIAL_FLUSH     (R600_CONTEXT_PRIVATE_FLAG << 9)
  72 #define SI_CONTEXT_CS_PARTIAL_FLUSH     (R600_CONTEXT_PRIVATE_FLAG << 10)
  73 #define SI_CONTEXT_VGT_FLUSH            (R600_CONTEXT_PRIVATE_FLAG << 11)
  74 #define SI_CONTEXT_VGT_STREAMOUT_SYNC   (R600_CONTEXT_PRIVATE_FLAG << 12)
  75 /* Compute only. */
  76 #define SI_CONTEXT_FLUSH_WITH_INV_L2    (R600_CONTEXT_PRIVATE_FLAG << 13) /* TODO: merge with TC? */
  77 #define SI_CONTEXT_FLAG_COMPUTE         (R600_CONTEXT_PRIVATE_FLAG << 14)
  78
  79 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
  80                                               SI_CONTEXT_FLUSH_AND_INV_CB_META | \
  81                                               SI_CONTEXT_FLUSH_AND_INV_DB | \
  82                                               SI_CONTEXT_FLUSH_AND_INV_DB_META)
  83
  84 #define SI_ENCODE_TRACE_POINT(id)       (0xcafe0000 | ((id) & 0xffff))
  85 #define SI_IS_TRACE_POINT(x)            (((x) & 0xcafe0000) == 0xcafe0000)
  86 #define SI_GET_TRACE_POINT_ID(x)        ((x) & 0xffff)
  87
  88 struct si_compute;
  89
  90 struct si_screen {
  91         struct r600_common_screen       b;
  92 };
  93
  94 struct si_sampler_view {
  95         struct pipe_sampler_view        base;
  96         struct list_head                list;
  97         struct r600_resource            *resource;
  98         /* [0..7] = image descriptor
  99          * [4..7] = buffer descriptor */
 100         uint32_t                        state[8];
 101         uint32_t                        fmask_state[8];
 102 };
 103
 104 struct si_sampler_state {
 105         uint32_t                        val[4];
 106         uint32_t                        border_color[4];
 107 };
 108
 109 struct si_cs_shader_state {
 110         struct si_compute               *program;
 111 };
 112
 113 struct si_textures_info {
 114         struct si_sampler_views         views;
 115         struct si_sampler_states        states;
 116         uint32_t                        depth_texture_mask; /* which textures are depth */
 117         uint32_t                        compressed_colortex_mask;
 118 };
 119
 120 struct si_framebuffer {
 121         struct r600_atom                atom;
 122         struct pipe_framebuffer_state   state;
 123         unsigned                        nr_samples;
 124         unsigned                        log_samples;
 125         unsigned                        cb0_is_integer;
 126         unsigned                        compressed_cb_mask;
 127         unsigned                        export_16bpc;
 128 };
 129
 130 #define SI_NUM_ATOMS(sctx) (sizeof((sctx)->atoms)/sizeof((sctx)->atoms.array[0]))
 131
 132 struct si_context {
 133         struct r600_common_context      b;
 134         struct blitter_context          *blitter;
 135         void                            *custom_dsa_flush;
 136         void                            *custom_blend_resolve;
 137         void                            *custom_blend_decompress;
 138         void                            *custom_blend_fastclear;
 139         void                            *pstipple_sampler_state;
 140         struct si_screen                *screen;
 141         struct si_pm4_state             *init_config;
 142         struct pipe_fence_handle        *last_gfx_fence;
 143         struct si_shader_selector       *fixed_func_tcs_shader;
 144
 145         union {
 146                 struct {
 147                         /* The order matters. */
 148                         struct r600_atom *cache_flush;
 149                         struct r600_atom *streamout_begin;
 150                         struct r600_atom *streamout_enable; /* must be after streamout_begin */
 151                         struct r600_atom *framebuffer;
 152                         struct r600_atom *msaa_sample_locs;
 153                         struct r600_atom *db_render_state;
 154                         struct r600_atom *msaa_config;
 155                         struct r600_atom *clip_regs;
 156                         struct r600_atom *shader_userdata;
 157                 } s;
 158                 struct r600_atom *array[0];
 159         } atoms;
 160
 161         struct si_framebuffer           framebuffer;
 162         struct si_vertex_element        *vertex_elements;
 163         /* for saving when using blitter */
 164         struct pipe_stencil_ref         stencil_ref;
 165         /* shaders */
 166         struct si_shader_selector       *ps_shader;
 167         struct si_shader_selector       *gs_shader;
 168         struct si_shader_selector       *vs_shader;
 169         struct si_shader_selector       *tcs_shader;
 170         struct si_shader_selector       *tes_shader;
 171         struct si_cs_shader_state       cs_shader_state;
 172         struct si_shader_data           shader_userdata;
 173         /* shader information */
 174         unsigned                        sprite_coord_enable;
 175         bool                            flatshade;
 176         struct si_descriptors           vertex_buffers;
 177         struct si_buffer_resources      const_buffers[SI_NUM_SHADERS];
 178         struct si_buffer_resources      rw_buffers[SI_NUM_SHADERS];
 179         struct si_textures_info         samplers[SI_NUM_SHADERS];
 180         struct r600_resource            *scratch_buffer;
 181         struct r600_resource            *border_color_table;
 182         unsigned                        border_color_offset;
 183
 184         struct r600_atom                clip_regs;
 185         struct r600_atom                msaa_sample_locs;
 186         struct r600_atom                msaa_config;
 187         int                             ps_iter_samples;
 188         bool                            smoothing_enabled;
 189
 190         /* Vertex and index buffers. */
 191         bool                    vertex_buffers_dirty;
 192         struct pipe_index_buffer index_buffer;
 193         struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
 194
 195         /* With rasterizer discard, there doesn't have to be a pixel shader.
 196          * In that case, we bind this one: */
 197         void                    *dummy_pixel_shader;
 198         struct r600_atom        cache_flush;
 199         struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
 200
 201         /* VGT states. */
 202         struct si_pm4_state     *vgt_shader_config[4];
 203         struct si_pm4_state     *gs_rings;
 204         struct pipe_resource    *esgs_ring;
 205         struct pipe_resource    *gsvs_ring;
 206         struct si_pm4_state     *tf_state;
 207         struct pipe_resource    *tf_ring;
 208
 209         LLVMTargetMachineRef            tm;
 210
 211         /* SI state handling */
 212         union si_state  queued;
 213         union si_state  emitted;
 214
 215         /* DB render state. */
 216         struct r600_atom        db_render_state;
 217         bool                    dbcb_depth_copy_enabled;
 218         bool                    dbcb_stencil_copy_enabled;
 219         unsigned                dbcb_copy_sample;
 220         bool                    db_inplace_flush_enabled;
 221         bool                    db_depth_clear;
 222         bool                    db_depth_disable_expclear;
 223         unsigned                ps_db_shader_control;
 224
 225         /* Emitted draw state. */
 226         int                     last_base_vertex;
 227         int                     last_start_instance;
 228         int                     last_sh_base_reg;
 229         int                     last_primitive_restart_en;
 230         int                     last_restart_index;
 231         int                     last_gs_out_prim;
 232         int                     last_prim;
 233         int                     last_multi_vgt_param;
 234         int                     last_ls_hs_config;
 235         int                     last_rast_prim;
 236         unsigned                last_sc_line_stipple;
 237         int                     current_rast_prim; /* primitive type after TES, GS */
 238
 239         /* Scratch buffer */
 240         boolean                 emit_scratch_reloc;
 241         unsigned                scratch_waves;
 242         unsigned                spi_tmpring_size;
 243
 244         /* Emitted derived tessellation state. */
 245         struct si_shader        *last_ls; /* local shader (VS) */
 246         struct si_shader_selector *last_tcs;
 247         int                     last_num_tcs_input_cp;
 248         int                     last_tes_sh_base;
 249
 250         /* Debug state. */
 251         bool                    is_debug;
 252         uint32_t                *last_ib;
 253         unsigned                last_ib_dw_size;
 254         struct r600_resource    *last_trace_buf;
 255         struct r600_resource    *trace_buf;
 256         unsigned                trace_id;
 257 };
 258
 259 /* cik_sdma.c */
 260 void cik_sdma_copy(struct pipe_context *ctx,
 261                    struct pipe_resource *dst,
 262                    unsigned dst_level,
 263                    unsigned dstx, unsigned dsty, unsigned dstz,
 264                    struct pipe_resource *src,
 265                    unsigned src_level,
 266                    const struct pipe_box *src_box);
 267
 268 /* si_blit.c */
 269 void si_init_blit_functions(struct si_context *sctx);
 270 void si_flush_depth_textures(struct si_context *sctx,
 271                              struct si_textures_info *textures);
 272 void si_decompress_color_textures(struct si_context *sctx,
 273                                   struct si_textures_info *textures);
 274 void si_resource_copy_region(struct pipe_context *ctx,
 275                              struct pipe_resource *dst,
 276                              unsigned dst_level,
 277                              unsigned dstx, unsigned dsty, unsigned dstz,
 278                              struct pipe_resource *src,
 279                              unsigned src_level,
 280                              const struct pipe_box *src_box);
 281
 282 /* si_cp_dma.c */
 283 void si_copy_buffer(struct si_context *sctx,
 284                     struct pipe_resource *dst, struct pipe_resource *src,
 285                     uint64_t dst_offset, uint64_t src_offset, unsigned size,
 286                     bool is_framebuffer);
 287 void si_init_cp_dma_functions(struct si_context *sctx);
 288
 289 /* si_debug.c */
 290 void si_init_debug_functions(struct si_context *sctx);
 291
 292 /* si_dma.c */
 293 void si_dma_copy(struct pipe_context *ctx,
 294                  struct pipe_resource *dst,
 295                  unsigned dst_level,
 296                  unsigned dstx, unsigned dsty, unsigned dstz,
 297                  struct pipe_resource *src,
 298                  unsigned src_level,
 299                  const struct pipe_box *src_box);
 300
 301 /* si_hw_context.c */
 302 void si_context_gfx_flush(void *context, unsigned flags,
 303                           struct pipe_fence_handle **fence);
 304 void si_begin_new_cs(struct si_context *ctx);
 305 void si_need_cs_space(struct si_context *ctx, unsigned num_dw, boolean count_draw_in);
 306
 307 /* si_compute.c */
 308 void si_init_compute_functions(struct si_context *sctx);
 309
 310 /* si_uvd.c */
 311 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
 312                                                const struct pipe_video_codec *templ);
 313
 314 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
 315                                                  const struct pipe_video_buffer *tmpl);
 316
 317 /*
 318  * common helpers
 319  */
 320
 321 static inline struct r600_resource *
 322 si_resource_create_custom(struct pipe_screen *screen,
 323                           unsigned usage, unsigned size)
 324 {
 325         assert(size);
 326         return r600_resource(pipe_buffer_create(screen,
 327                 PIPE_BIND_CUSTOM, usage, size));
 328 }
 329
 330 static inline void
 331 si_invalidate_draw_sh_constants(struct si_context *sctx)
 332 {
 333         sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
 334         sctx->last_start_instance = -1; /* reset to an unknown value */
 335         sctx->last_sh_base_reg = -1; /* reset to an unknown value */
 336 }
 337
 338 static inline void
 339 si_set_atom_dirty(struct si_context *sctx,
 340                   struct r600_atom *atom, bool dirty)
 341 {
 342         atom->dirty = dirty;
 343 }
 344
 345 static inline void
 346 si_mark_atom_dirty(struct si_context *sctx,
 347                    struct r600_atom *atom)
 348 {
 349         si_set_atom_dirty(sctx, atom, true);
 350 }
 351
 352 #endif