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