#define SI_RESTART_INDEX_UNKNOWN INT_MIN
#define SI_NUM_SMOOTH_AA_SAMPLES 8
#define SI_GS_PER_ES 128
+/* Alignment for optimal CP DMA performance. */
+#define SI_CPDMA_ALIGNMENT 32
/* Instruction cache. */
#define SI_CONTEXT_INV_ICACHE (R600_CONTEXT_PRIVATE_FLAG << 0)
/* Write dirty L2 lines back to memory (shader and CP DMA stores), but don't
* invalidate L2. SI-CIK can't do it, so they will do complete invalidation. */
#define SI_CONTEXT_WRITEBACK_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 4)
+/* gaps */
/* Framebuffer caches. */
-#define SI_CONTEXT_FLUSH_AND_INV_CB_META (R600_CONTEXT_PRIVATE_FLAG << 5)
-#define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 6)
#define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 7)
#define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 8)
/* Engine synchronization. */
#define SI_CONTEXT_VGT_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 12)
#define SI_CONTEXT_VGT_STREAMOUT_SYNC (R600_CONTEXT_PRIVATE_FLAG << 13)
-#define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
- SI_CONTEXT_FLUSH_AND_INV_CB_META | \
- SI_CONTEXT_FLUSH_AND_INV_DB | \
- SI_CONTEXT_FLUSH_AND_INV_DB_META)
-
#define SI_MAX_BORDER_COLORS 4096
struct si_compute;
bool has_distributed_tess;
bool has_draw_indirect_multi;
bool has_ds_bpermute;
+ bool has_msaa_sample_loc_bug;
/* Whether shaders are monolithic (1-part) or separate (3-part). */
bool use_monolithic_shaders;
bool record_llvm_ir;
- pipe_mutex shader_parts_mutex;
+ mtx_t shader_parts_mutex;
struct si_shader_part *vs_prologs;
- struct si_shader_part *vs_epilogs;
struct si_shader_part *tcs_epilogs;
struct si_shader_part *gs_prologs;
struct si_shader_part *ps_prologs;
* - GS and CS aren't cached, but it's certainly possible to cache
* those as well.
*/
- pipe_mutex shader_cache_mutex;
+ mtx_t shader_cache_mutex;
struct hash_table *shader_cache;
/* Shader compiler queue for multithreaded compilation. */
struct util_queue shader_compiler_queue;
LLVMTargetMachineRef tm[4]; /* used by the queue only */
+
+ struct util_queue shader_compiler_queue_low_priority;
+ LLVMTargetMachineRef tm_low_priority[4];
};
struct si_blend_color {
* [4..7] = buffer descriptor */
uint32_t state[8];
uint32_t fmask_state[8];
- const struct radeon_surf_level *base_level_info;
+ const struct legacy_surf_level *base_level_info;
unsigned base_level;
unsigned block_width;
bool is_stencil_sampler;
+ bool dcc_incompatible;
};
#define SI_SAMPLER_STATE_MAGIC 0x34f1c35a
unsigned spi_shader_col_format_alpha;
unsigned spi_shader_col_format_blend;
unsigned spi_shader_col_format_blend_alpha;
- unsigned color_is_int8; /* bitmask */
+ unsigned color_is_int8;
+ unsigned color_is_int10;
unsigned dirty_cbufs;
bool dirty_zsbuf;
bool any_dst_linear;
unsigned count_from_stream_output:1;
unsigned line_stipple_enabled:1;
unsigned uses_tess:1;
- unsigned tcs_tes_uses_prim_id:1;
+ unsigned tess_uses_prim_id:1;
unsigned uses_gs:1;
unsigned _pad:32 - SI_NUM_VGT_PARAM_KEY_BITS;
} u;
struct radeon_winsys_cs *ce_ib;
struct radeon_winsys_cs *ce_preamble_ib;
+ struct r600_resource *ce_ram_saved_buffer;
+ unsigned ce_ram_saved_offset;
+ unsigned total_ce_ram_allocated;
bool ce_need_synchronization;
struct u_suballocator *ce_suballocator;
struct r600_atom msaa_config;
struct si_sample_mask sample_mask;
struct r600_atom cb_render_state;
+ unsigned last_cb_target_mask;
struct si_blend_color blend_color;
struct r600_atom clip_regs;
struct si_clip_state clip_state;
unsigned shader_pointers_dirty;
unsigned compressed_tex_shader_mask;
struct si_buffer_resources rw_buffers;
- struct si_buffer_resources const_buffers[SI_NUM_SHADERS];
- struct si_buffer_resources shader_buffers[SI_NUM_SHADERS];
+ struct si_buffer_resources const_and_shader_buffers[SI_NUM_SHADERS];
struct si_textures_info samplers[SI_NUM_SHADERS];
struct si_images_info images[SI_NUM_SHADERS];
/* Vertex and index buffers. */
bool vertex_buffers_dirty;
bool vertex_buffer_pointer_dirty;
- struct pipe_index_buffer index_buffer;
struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
/* MSAA config state. */
int last_multi_vgt_param;
int last_rast_prim;
unsigned last_sc_line_stipple;
- int current_rast_prim; /* primitive type after TES, GS */
+ unsigned current_vs_state;
+ unsigned last_vs_state;
+ enum pipe_prim_type current_rast_prim; /* primitive type after TES, GS */
bool gs_tri_strip_adj_fix;
/* Scratch buffer */
+ struct r600_atom scratch_state;
struct r600_resource *scratch_buffer;
- bool emit_scratch_reloc;
unsigned scratch_waves;
unsigned spi_tmpring_size;
struct r600_resource *compute_scratch_buffer;
/* Emitted derived tessellation state. */
- struct si_shader *last_ls; /* local shader (VS) */
+ /* Local shader (VS), or HS if LS-HS are merged. */
+ struct si_shader *last_ls;
struct si_shader_selector *last_tcs;
int last_num_tcs_input_cp;
int last_tes_sh_base;
void si_init_debug_functions(struct si_context *sctx);
void si_check_vm_faults(struct r600_common_context *ctx,
struct radeon_saved_cs *saved, enum ring_type ring);
-bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary);
+bool si_replace_shader(unsigned num, struct ac_shader_binary *binary);
/* si_dma.c */
void si_init_dma_functions(struct si_context *sctx);
return sctx->vs_shader.current;
}
-static inline bool si_vs_exports_prim_id(struct si_shader *shader)
+static inline unsigned
+si_optimal_tcc_alignment(struct si_context *sctx, unsigned upload_size)
{
- if (shader->selector->type == PIPE_SHADER_VERTEX)
- return shader->key.part.vs.epilog.export_prim_id;
- else if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
- return shader->key.part.tes.epilog.export_prim_id;
- else
- return false;
+ unsigned alignment, tcc_cache_line_size;
+
+ /* If the upload size is less than the cache line size (e.g. 16, 32),
+ * the whole thing will fit into a cache line if we align it to its size.
+ * The idea is that multiple small uploads can share a cache line.
+ * If the upload size is greater, align it to the cache line size.
+ */
+ alignment = util_next_power_of_two(upload_size);
+ tcc_cache_line_size = sctx->screen->b.info.tcc_cache_line_size;
+ return MIN2(alignment, tcc_cache_line_size);
}
#endif