/* 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 */
+/* Writeback & invalidate the L2 metadata cache. It can only be coupled with
+ * a CB or DB flush. */
+#define SI_CONTEXT_INV_L2_METADATA (R600_CONTEXT_PRIVATE_FLAG << 5)
/* Framebuffer caches. */
-#define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 7)
+#define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 6)
+#define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 7)
#define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 8)
/* Engine synchronization. */
#define SI_CONTEXT_VS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 9)
#define SI_PREFETCH_PS (1 << 6)
#define SI_MAX_BORDER_COLORS 4096
+#define SI_MAX_VIEWPORTS 16
#define SIX_BITS 0x3F
struct si_compute;
bool has_clear_state;
bool has_distributed_tess;
bool has_draw_indirect_multi;
- bool has_ds_bpermute;
+ bool has_out_of_order_rast;
+ bool assume_no_z_fights;
+ bool commutative_blend_add;
+ bool clear_db_cache_before_clear;
bool has_msaa_sample_loc_bug;
+ bool dpbb_allowed;
+ bool dfsm_allowed;
bool llvm_has_working_vgpr_indexing;
/* Whether shaders are monolithic (1-part) or separate (3-part). */
ubyte base_level;
ubyte block_width;
bool is_stencil_sampler;
+ bool is_integer;
bool dcc_incompatible;
};
unsigned magic;
#endif
uint32_t val[4];
+ uint32_t integer_val[4];
+ uint32_t upgraded_depth_val[4];
};
struct si_cs_shader_state {
bool uses_scratch;
};
-struct si_textures_info {
- struct si_sampler_views views;
+struct si_samplers {
+ struct pipe_sampler_view *views[SI_NUM_SAMPLERS];
+ struct si_sampler_state *sampler_states[SI_NUM_SAMPLERS];
+
+ /* The i-th bit is set if that element is enabled (non-NULL resource). */
+ unsigned enabled_mask;
uint32_t needs_depth_decompress_mask;
uint32_t needs_color_decompress_mask;
};
-struct si_images_info {
+struct si_images {
struct pipe_image_view views[SI_NUM_IMAGES];
uint32_t needs_color_decompress_mask;
unsigned enabled_mask;
ubyte dirty_cbufs;
bool dirty_zsbuf;
bool any_dst_linear;
+ bool CB_has_shader_readable_metadata;
+ bool DB_has_shader_readable_metadata;
+};
+
+struct si_signed_scissor {
+ int minx;
+ int miny;
+ int maxx;
+ int maxy;
+};
+
+struct si_scissors {
+ struct r600_atom atom;
+ unsigned dirty_mask;
+ struct pipe_scissor_state states[SI_MAX_VIEWPORTS];
+};
+
+struct si_viewports {
+ struct r600_atom atom;
+ unsigned dirty_mask;
+ unsigned depth_range_dirty_mask;
+ struct pipe_viewport_state states[SI_MAX_VIEWPORTS];
+ struct si_signed_scissor as_scissor[SI_MAX_VIEWPORTS];
};
struct si_clip_state {
uint16_t sample_mask;
};
+struct si_streamout_target {
+ struct pipe_stream_output_target b;
+
+ /* The buffer where BUFFER_FILLED_SIZE is stored. */
+ struct r600_resource *buf_filled_size;
+ unsigned buf_filled_size_offset;
+ bool buf_filled_size_valid;
+
+ unsigned stride_in_dw;
+};
+
+struct si_streamout {
+ struct r600_atom begin_atom;
+ bool begin_emitted;
+
+ unsigned enabled_mask;
+ unsigned num_targets;
+ struct si_streamout_target *targets[PIPE_MAX_SO_BUFFERS];
+
+ unsigned append_bitmask;
+ bool suspended;
+
+ /* External state which comes from the vertex shader,
+ * it must be set explicitly when binding a shader. */
+ uint16_t *stride_in_dw;
+ unsigned enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
+
+ /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
+ unsigned hw_enabled_mask;
+
+ /* The state of VGT_STRMOUT_(CONFIG|EN). */
+ struct r600_atom enable_atom;
+ bool streamout_enabled;
+ bool prims_gen_query_enabled;
+ int num_prims_gen_queries;
+};
+
/* A shader state consists of the shader selector, which is a constant state
* object shared by multiple contexts and shouldn't be modified, and
* the current shader variant selected for this context.
void *custom_blend_fmask_decompress;
void *custom_blend_eliminate_fastclear;
void *custom_blend_dcc_decompress;
+ void *vs_blit_pos;
+ void *vs_blit_pos_layered;
+ void *vs_blit_color;
+ void *vs_blit_color_layered;
+ void *vs_blit_texcoord;
struct si_screen *screen;
LLVMTargetMachineRef tm; /* only non-threaded compilation */
struct si_shader_ctx_state fixed_func_tcs_shader;
struct si_framebuffer framebuffer;
struct si_sample_locs msaa_sample_locs;
struct r600_atom db_render_state;
+ struct r600_atom dpbb_state;
struct r600_atom msaa_config;
struct si_sample_mask sample_mask;
struct r600_atom cb_render_state;
struct si_shader_data shader_pointers;
struct si_stencil_ref stencil_ref;
struct r600_atom spi_map;
+ struct si_scissors scissors;
+ struct si_streamout streamout;
+ struct si_viewports viewports;
/* Precomputed states. */
struct si_pm4_state *init_config;
unsigned shader_needs_decompress_mask;
struct si_buffer_resources rw_buffers;
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];
+ struct si_samplers samplers[SI_NUM_SHADERS];
+ struct si_images images[SI_NUM_SHADERS];
/* other shader resources */
struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
struct r600_resource *border_color_buffer;
union pipe_color_union *border_color_map; /* in VRAM (slow access), little endian */
unsigned border_color_count;
+ unsigned num_vs_blit_sgprs;
+ uint32_t vs_blit_sh_data[SI_VS_BLIT_SGPRS_POS_TEXCOORD];
/* Vertex and index buffers. */
bool vertex_buffers_dirty;
/* Emitted draw state. */
bool gs_tri_strip_adj_fix:1;
+ bool ls_vgpr_fix:1;
int last_index_size;
int last_base_vertex;
int last_start_instance;
bool need_check_render_feedback;
bool decompression_enabled;
+ bool vs_writes_viewport_index;
+ bool vs_disables_clipping_viewport;
+
/* Precomputed IA_MULTI_VGT_PARAM */
union si_vgt_param_key ia_multi_vgt_param_key;
unsigned ia_multi_vgt_param[SI_NUM_VGT_PARAM_STATES];
/* Bindless state */
bool uses_bindless_samplers;
bool uses_bindless_images;
+
+ /* MSAA sample locations.
+ * The first index is the sample index.
+ * The second index is the coordinate: X, Y. */
+ float sample_locations_1x[1][2];
+ float sample_locations_2x[2][2];
+ float sample_locations_4x[4][2];
+ float sample_locations_8x[8][2];
+ float sample_locations_16x[16][2];
};
/* cik_sdma.c */
/* si_blit.c */
void si_init_blit_functions(struct si_context *sctx);
-void si_decompress_graphics_textures(struct si_context *sctx);
-void si_decompress_compute_textures(struct si_context *sctx);
+void si_decompress_textures(struct si_context *sctx, unsigned shader_mask);
void si_resource_copy_region(struct pipe_context *ctx,
struct pipe_resource *dst,
unsigned dst_level,
struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
const struct pipe_video_buffer *tmpl);
+/* si_viewport.c */
+void si_update_vs_viewport_state(struct si_context *ctx);
+void si_init_viewport_functions(struct si_context *ctx);
+
+
/*
* common helpers
*/
return vs->current ? vs->current : NULL;
}
+static inline bool si_get_strmout_en(struct si_context *sctx)
+{
+ return sctx->streamout.streamout_enabled ||
+ sctx->streamout.prims_gen_query_enabled;
+}
+
static inline unsigned
si_optimal_tcc_alignment(struct si_context *sctx, unsigned upload_size)
{
}
static inline void
-si_make_CB_shader_coherent(struct si_context *sctx, unsigned num_samples)
+si_make_CB_shader_coherent(struct si_context *sctx, unsigned num_samples,
+ bool shaders_read_metadata)
{
sctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_CB |
SI_CONTEXT_INV_VMEM_L1;
- /* Single-sample color is coherent with shaders on GFX9. */
- if (sctx->b.chip_class <= VI || num_samples >= 2)
+ if (sctx->b.chip_class >= GFX9) {
+ /* Single-sample color is coherent with shaders on GFX9, but
+ * L2 metadata must be flushed if shaders read metadata.
+ * (DCC, CMASK).
+ */
+ if (num_samples >= 2)
+ sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
+ else if (shaders_read_metadata)
+ sctx->b.flags |= SI_CONTEXT_INV_L2_METADATA;
+ } else {
+ /* SI-CI-VI */
sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
+ }
}
static inline void
si_make_DB_shader_coherent(struct si_context *sctx, unsigned num_samples,
- bool include_stencil)
+ bool include_stencil, bool shaders_read_metadata)
{
sctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_DB |
SI_CONTEXT_INV_VMEM_L1;
- /* Single-sample depth (not stencil) is coherent with shaders on GFX9. */
- if (sctx->b.chip_class <= VI || num_samples >= 2 || include_stencil)
+ if (sctx->b.chip_class >= GFX9) {
+ /* Single-sample depth (not stencil) is coherent with shaders
+ * on GFX9, but L2 metadata must be flushed if shaders read
+ * metadata.
+ */
+ if (num_samples >= 2 || include_stencil)
+ sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
+ else if (shaders_read_metadata)
+ sctx->b.flags |= SI_CONTEXT_INV_L2_METADATA;
+ } else {
+ /* SI-CI-VI */
sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
+ }
}
#endif