#include "util/u_idalloc.h"
#include "util/u_threaded_context.h"
-#ifdef PIPE_ARCH_BIG_ENDIAN
+#if UTIL_ARCH_BIG_ENDIAN
#define SI_BIG_ENDIAN 1
#else
#define SI_BIG_ENDIAN 0
#define SI_RESOURCE_FLAG_CLEAR (PIPE_RESOURCE_FLAG_DRV_PRIV << 7)
/* For const_uploader, upload data via GTT and copy to VRAM on context flush via SDMA. */
#define SI_RESOURCE_FLAG_UPLOAD_FLUSH_EXPLICIT_VIA_SDMA (PIPE_RESOURCE_FLAG_DRV_PRIV << 8)
+/* Set a micro tile mode: */
+#define SI_RESOURCE_FLAG_FORCE_MICRO_TILE_MODE (PIPE_RESOURCE_FLAG_DRV_PRIV << 9)
+#define SI_RESOURCE_FLAG_MICRO_TILE_MODE_SHIFT (util_logbase2(PIPE_RESOURCE_FLAG_DRV_PRIV) + 10)
+#define SI_RESOURCE_FLAG_MICRO_TILE_MODE_SET(x) (((x) & 0x3) << SI_RESOURCE_FLAG_MICRO_TILE_MODE_SHIFT)
+#define SI_RESOURCE_FLAG_MICRO_TILE_MODE_GET(x) (((x) >> SI_RESOURCE_FLAG_MICRO_TILE_MODE_SHIFT) & 0x3)
enum si_clear_code
{
DBG_TES = PIPE_SHADER_TESS_EVAL,
DBG_CS = PIPE_SHADER_COMPUTE,
DBG_NO_IR,
- DBG_NO_TGSI,
+ DBG_NO_NIR,
DBG_NO_ASM,
DBG_PREOPT_IR,
DBG_VM,
/* Driver options: */
- DBG_FORCE_DMA,
- DBG_NO_ASYNC_DMA,
+ DBG_FORCE_SDMA,
+ DBG_NO_SDMA,
+ DBG_NO_SDMA_CLEARS,
+ DBG_NO_SDMA_COPY_IMAGE,
DBG_NO_WC,
DBG_CHECK_VM,
DBG_RESERVE_VMID,
uint16_t stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
enum pipe_format db_render_format:16;
uint8_t stencil_clear_value;
- bool fmask_is_not_identity:1;
+ bool fmask_is_identity:1;
bool tc_compatible_htile:1;
bool htile_stencil_disabled:1;
bool depth_cleared:1; /* if it was cleared at least once */
* for a possible future enablement.
*/
bool separate_dcc_dirty:1;
+ bool displayable_dcc_dirty:1;
+
/* Statistics gathering for the DCC enablement heuristic. */
bool dcc_gather_statistics:1;
/* Counter that should be non-zero if the texture is bound to a
uint32_t *state,
uint32_t *fmask_state);
+ unsigned num_vbos_in_user_sgprs;
unsigned pa_sc_raster_config;
unsigned pa_sc_raster_config_1;
unsigned se_tile_repeat;
*
* Design & limitations:
* - The shader cache is per screen (= per process), never saved to
- * disk, and skips redundant shader compilations from TGSI to bytecode.
+ * disk, and skips redundant shader compilations from NIR to bytecode.
* - It can only be used with one-variant-per-shader support, in which
* case only the main (typically middle) part of shaders is cached.
* - Only VS, TCS, TES, PS are cached, out of which only the hw VS
ubyte nr_color_samples; /* at most 8xAA */
ubyte compressed_cb_mask;
ubyte uncompressed_cb_mask;
+ ubyte displayable_dcc_cb_mask;
ubyte color_is_int8;
ubyte color_is_int10;
ubyte dirty_cbufs;
*/
union si_vgt_param_key {
struct {
-#ifdef PIPE_ARCH_LITTLE_ENDIAN
+#if UTIL_ARCH_LITTLE_ENDIAN
unsigned prim:4;
unsigned uses_instancing:1;
unsigned multi_instances_smaller_than_primgroup:1;
unsigned tess_uses_prim_id:1;
unsigned uses_gs:1;
unsigned _pad:32 - SI_NUM_VGT_PARAM_KEY_BITS;
-#else /* PIPE_ARCH_BIG_ENDIAN */
+#else /* UTIL_ARCH_BIG_ENDIAN */
unsigned _pad:32 - SI_NUM_VGT_PARAM_KEY_BITS;
unsigned uses_gs:1;
unsigned tess_uses_prim_id:1;
uint32_t index;
};
-#define SI_NUM_VGT_STAGES_KEY_BITS 4
+#define SI_NUM_VGT_STAGES_KEY_BITS 5
#define SI_NUM_VGT_STAGES_STATES (1 << SI_NUM_VGT_STAGES_KEY_BITS)
/* The VGT_SHADER_STAGES key used to index the table of precomputed values.
*/
union si_vgt_stages_key {
struct {
-#ifdef PIPE_ARCH_LITTLE_ENDIAN
+#if UTIL_ARCH_LITTLE_ENDIAN
unsigned tess:1;
unsigned gs:1;
+ unsigned ngg_passthrough:1;
unsigned ngg:1; /* gfx10+ */
unsigned streamout:1; /* only used with NGG */
unsigned _pad:32 - SI_NUM_VGT_STAGES_KEY_BITS;
-#else /* PIPE_ARCH_BIG_ENDIAN */
+#else /* UTIL_ARCH_BIG_ENDIAN */
unsigned _pad:32 - SI_NUM_VGT_STAGES_KEY_BITS;
unsigned streamout:1;
unsigned ngg:1;
+ unsigned ngg_passthrough:1;
unsigned gs:1;
unsigned tess:1;
#endif
unsigned size;
};
+struct si_small_prim_cull_info {
+ float scale[2], translate[2];
+};
+
struct si_context {
struct pipe_context b; /* base class */
struct radeon_winsys *ws;
struct radeon_winsys_ctx *ctx;
struct radeon_cmdbuf *gfx_cs; /* compute IB if graphics is disabled */
- struct radeon_cmdbuf *dma_cs;
+ struct radeon_cmdbuf *sdma_cs;
struct pipe_fence_handle *last_gfx_fence;
struct pipe_fence_handle *last_sdma_fence;
struct si_resource *eop_bug_scratch;
void *cs_copy_image_1d_array;
void *cs_clear_render_target;
void *cs_clear_render_target_1d_array;
+ void *cs_clear_12bytes_buffer;
void *cs_dcc_retile;
void *cs_fmask_expand[3][2]; /* [log2(samples)-1][is_array] */
struct si_screen *screen;
/* shader information */
struct si_vertex_elements *vertex_elements;
+ unsigned num_vertex_elements;
unsigned sprite_coord_enable;
unsigned cs_max_waves_per_sh;
bool flatshade;
bool do_update_shaders;
- /* vertex buffer descriptors */
- uint32_t *vb_descriptors_gpu_list;
- struct si_resource *vb_descriptors_buffer;
- unsigned vb_descriptors_offset;
-
/* shader descriptors */
struct si_descriptors descriptors[SI_NUM_DESCS];
unsigned descriptors_dirty;
uint32_t vs_blit_sh_data[SI_VS_BLIT_SGPRS_POS_TEXCOORD];
uint32_t cs_user_data[4];
- /* Vertex and index buffers. */
+ /* Vertex buffers. */
bool vertex_buffers_dirty;
bool vertex_buffer_pointer_dirty;
+ bool vertex_buffer_user_sgprs_dirty;
struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
uint16_t vertex_buffer_unaligned; /* bitmask of not dword-aligned buffers */
+ uint32_t *vb_descriptors_gpu_list;
+ struct si_resource *vb_descriptors_buffer;
+ unsigned vb_descriptors_offset;
+ unsigned vb_descriptor_user_sgprs[5*4];
/* MSAA config state. */
int ps_iter_samples;
int last_restart_index;
int last_prim;
int last_multi_vgt_param;
- int last_rast_prim;
- int last_flatshade_first;
+ int last_gs_out_prim;
int last_binning_enabled;
- unsigned last_sc_line_stipple;
unsigned current_vs_state;
unsigned last_vs_state;
enum pipe_prim_type current_rast_prim; /* primitive type after TES, GS */
struct radeon_saved_cs *saved, enum ring_type ring);
bool si_replace_shader(unsigned num, struct si_shader_binary *binary);
-/* si_dma.c */
-void si_init_dma_functions(struct si_context *sctx);
-
/* si_dma_cs.c */
void si_dma_emit_timestamp(struct si_context *sctx, struct si_resource *dst,
uint64_t offset);
void si_sdma_clear_buffer(struct si_context *sctx, struct pipe_resource *dst,
uint64_t offset, uint64_t size, unsigned clear_value);
+void si_sdma_copy_buffer(struct si_context *sctx, struct pipe_resource *dst,
+ struct pipe_resource *src, uint64_t dst_offset,
+ uint64_t src_offset, uint64_t size);
void si_need_dma_space(struct si_context *ctx, unsigned num_dw,
struct si_resource *dst, struct si_resource *src);
void si_flush_dma_cs(struct si_context *ctx, unsigned flags,
unsigned base_vertex,
uint64_t input_indexbuf_va,
unsigned input_indexbuf_max_elements);
-void si_initialize_prim_discard_tunables(struct si_context *sctx);
+void si_initialize_prim_discard_tunables(struct si_screen *sscreen,
+ bool is_aux_context,
+ unsigned *prim_discard_vertex_count_threshold,
+ unsigned *index_ring_size_per_ib);
+
+/* si_pipe.c */
+void si_init_compiler(struct si_screen *sscreen, struct ac_llvm_compiler *compiler);
/* si_perfcounters.c */
void si_init_perfcounters(struct si_screen *screen);
void *si_create_copy_image_compute_shader_1d_array(struct pipe_context *ctx);
void *si_clear_render_target_shader(struct pipe_context *ctx);
void *si_clear_render_target_shader_1d_array(struct pipe_context *ctx);
+void *si_clear_12bytes_buffer_shader(struct pipe_context *ctx);
void *si_create_dcc_retile_cs(struct pipe_context *ctx);
void *si_create_fmask_expand_cs(struct pipe_context *ctx, unsigned num_samples,
bool is_array);
const struct pipe_video_buffer *tmpl);
/* si_viewport.c */
+void si_get_small_prim_cull_info(struct si_context *sctx,
+ struct si_small_prim_cull_info *out);
void si_update_vs_viewport_state(struct si_context *ctx);
void si_init_viewport_functions(struct si_context *ctx);
return &sctx->vs_shader;
}
-static inline struct tgsi_shader_info *si_get_vs_info(struct si_context *sctx)
+static inline struct si_shader_info *si_get_vs_info(struct si_context *sctx)
{
struct si_shader_ctx_state *vs = si_get_vs(sctx);