#define SI_BASE_VERTEX_UNKNOWN INT_MIN
#define SI_RESTART_INDEX_UNKNOWN INT_MIN
#define SI_NUM_SMOOTH_AA_SAMPLES 8
+#define SI_MAX_POINT_SIZE 2048
#define SI_GS_PER_ES 128
/* Alignment for optimal CP DMA performance. */
#define SI_CPDMA_ALIGNMENT 32
+/* Tunables for compute-based clear_buffer and copy_buffer: */
+#define SI_COMPUTE_CLEAR_DW_PER_THREAD 4
+#define SI_COMPUTE_COPY_DW_PER_THREAD 4
+#define SI_COMPUTE_DST_CACHE_POLICY L2_STREAM
+
/* Pipeline & streamout query controls. */
#define SI_CONTEXT_START_PIPELINE_STATS (1 << 0)
#define SI_CONTEXT_STOP_PIPELINE_STATS (1 << 1)
#define SI_RESOURCE_FLAG_TRANSFER (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
#define SI_RESOURCE_FLAG_FLUSHED_DEPTH (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
-#define SI_RESOURCE_FLAG_FORCE_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
+#define SI_RESOURCE_FLAG_FORCE_MSAA_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
#define SI_RESOURCE_FLAG_DISABLE_DCC (PIPE_RESOURCE_FLAG_DRV_PRIV << 3)
#define SI_RESOURCE_FLAG_UNMAPPABLE (PIPE_RESOURCE_FLAG_DRV_PRIV << 4)
#define SI_RESOURCE_FLAG_READ_ONLY (PIPE_RESOURCE_FLAG_DRV_PRIV << 5)
#define SI_RESOURCE_FLAG_32BIT (PIPE_RESOURCE_FLAG_DRV_PRIV << 6)
+#define SI_RESOURCE_FLAG_SO_FILLED_SIZE (PIPE_RESOURCE_FLAG_DRV_PRIV << 7)
/* Debug flags. */
enum {
DBG_TEST_VMFAULT_SHADER,
DBG_TEST_DMA_PERF,
DBG_TEST_GDS,
+ DBG_TEST_GDS_MM,
+ DBG_TEST_GDS_OA_MM,
};
#define DBG_ALL_SHADERS (((1 << (DBG_CS + 1)) - 1))
#define DBG(name) (1ull << DBG_##name)
+enum si_cache_policy {
+ L2_BYPASS,
+ L2_STREAM, /* same as SLC=1 */
+ L2_LRU, /* same as SLC=0 */
+};
+
+enum si_coherency {
+ SI_COHERENCY_NONE, /* no cache flushes needed */
+ SI_COHERENCY_SHADER,
+ SI_COHERENCY_CB_META,
+ SI_COHERENCY_CP,
+};
+
struct si_compute;
struct hash_table;
struct u_suballocator;
uint64_t debug_flags;
char renderer_string[183];
+ unsigned pa_sc_raster_config;
+ unsigned pa_sc_raster_config_1;
+ unsigned se_tile_repeat;
unsigned gs_table_depth;
unsigned tess_offchip_block_dw_size;
unsigned tess_offchip_ring_size;
bool clear_db_cache_before_clear;
bool has_msaa_sample_loc_bug;
bool has_ls_vgpr_init_bug;
+ bool has_dcc_constant_encode;
bool dpbb_allowed;
bool dfsm_allowed;
bool llvm_has_working_vgpr_indexing;
ubyte color_is_int8;
ubyte color_is_int10;
ubyte dirty_cbufs;
+ ubyte dcc_overwrite_combiner_watermark;
bool dirty_zsbuf;
bool any_dst_linear;
bool CB_has_shader_readable_metadata;
bool DB_has_shader_readable_metadata;
};
+enum si_quant_mode {
+ /* This is the list we want to support. */
+ SI_QUANT_MODE_16_8_FIXED_POINT_1_256TH,
+ SI_QUANT_MODE_14_10_FIXED_POINT_1_1024TH,
+ SI_QUANT_MODE_12_12_FIXED_POINT_1_4096TH,
+};
+
struct si_signed_scissor {
int minx;
int miny;
int maxx;
int maxy;
+ enum si_quant_mode quant_mode;
};
struct si_scissors {
void *vs_blit_color;
void *vs_blit_color_layered;
void *vs_blit_texcoord;
+ void *cs_clear_buffer;
+ void *cs_copy_buffer;
struct si_screen *screen;
struct pipe_debug_callback debug;
struct ac_llvm_compiler compiler; /* only non-threaded compilation */
/* 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];
+ struct {
+ float x1[1][2];
+ float x2[2][2];
+ float x4[4][2];
+ float x8[8][2];
+ float x16[16][2];
+ } sample_positions;
+ struct pipe_resource *sample_pos_buffer;
/* Misc stats. */
unsigned num_draw_calls;
unsigned num_resident_handles;
uint64_t num_alloc_tex_transfer_bytes;
unsigned last_tex_ps_draw_ratio; /* for query */
+ unsigned context_roll_counter;
/* Queries. */
/* Maintain the list of active queries for pausing between IBs. */
unsigned level, unsigned clear_value);
void si_init_clear_functions(struct si_context *sctx);
+/* si_compute_blit.c */
+unsigned si_get_flush_flags(struct si_context *sctx, enum si_coherency coher,
+ enum si_cache_policy cache_policy);
+void si_clear_buffer(struct si_context *sctx, struct pipe_resource *dst,
+ uint64_t offset, uint64_t size, uint32_t *clear_value,
+ uint32_t clear_value_size, enum si_coherency coher);
+void si_copy_buffer(struct si_context *sctx,
+ struct pipe_resource *dst, struct pipe_resource *src,
+ uint64_t dst_offset, uint64_t src_offset, unsigned size);
+void si_init_compute_blit_functions(struct si_context *sctx);
+
/* si_cp_dma.c */
#define SI_CPDMA_SKIP_CHECK_CS_SPACE (1 << 0) /* don't call need_cs_space */
#define SI_CPDMA_SKIP_SYNC_AFTER (1 << 1) /* don't wait for DMA after the copy */
SI_CPDMA_SKIP_GFX_SYNC | \
SI_CPDMA_SKIP_BO_LIST_UPDATE)
-enum si_cache_policy {
- L2_BYPASS,
- L2_STREAM, /* same as SLC=1 */
- L2_LRU, /* same as SLC=0 */
-};
-
-enum si_coherency {
- SI_COHERENCY_NONE, /* no cache flushes needed */
- SI_COHERENCY_SHADER,
- SI_COHERENCY_CB_META,
-};
-
void si_cp_dma_wait_for_idle(struct si_context *sctx);
-void si_cp_dma_clear_buffer(struct si_context *sctx, struct pipe_resource *dst,
- uint64_t offset, uint64_t size, unsigned value,
- enum si_coherency coher,
- enum si_cache_policy cache_policy);
-void si_clear_buffer(struct si_context *sctx, struct pipe_resource *dst,
- uint64_t offset, uint64_t size, unsigned value,
- enum si_coherency coher);
+void si_cp_dma_clear_buffer(struct si_context *sctx, struct radeon_cmdbuf *cs,
+ struct pipe_resource *dst, uint64_t offset,
+ uint64_t size, unsigned value, unsigned user_flags,
+ enum si_coherency coher, enum si_cache_policy cache_policy);
void si_cp_dma_copy_buffer(struct si_context *sctx,
struct pipe_resource *dst, struct pipe_resource *src,
uint64_t dst_offset, uint64_t src_offset, unsigned size,
unsigned user_flags, enum si_coherency coher,
enum si_cache_policy cache_policy);
-void si_copy_buffer(struct si_context *sctx,
- struct pipe_resource *dst, struct pipe_resource *src,
- uint64_t dst_offset, uint64_t src_offset, unsigned size);
void cik_prefetch_TC_L2_async(struct si_context *sctx, struct pipe_resource *buf,
uint64_t offset, unsigned size);
void cik_emit_prefetch_L2(struct si_context *sctx, bool vertex_stage_only);
void si_test_gds(struct si_context *sctx);
-void si_init_cp_dma_functions(struct si_context *sctx);
/* si_debug.c */
void si_save_cs(struct radeon_winsys *ws, struct radeon_cmdbuf *cs,
projects / mesa.git / blobdiff