X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fradeonsi%2Fsi_shader.h;h=3a1d0e44290130aabf8dd5b6f7f29d02c6d718c4;hb=735a3ba00765baa717ff541fb5aa5105dc816ad7;hp=0054a48ac15e83469279a499da4d0a05fabf2d03;hpb=902dd50cf0f31fe1e730ee077c9e23716d06e066;p=mesa.git diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h index 0054a48ac15..3a1d0e44290 100644 --- a/src/gallium/drivers/radeonsi/si_shader.h +++ b/src/gallium/drivers/radeonsi/si_shader.h @@ -131,11 +131,9 @@ #ifndef SI_SHADER_H #define SI_SHADER_H -#include /* LLVMModuleRef */ -#include -#include "tgsi/tgsi_scan.h" #include "util/u_inlines.h" #include "util/u_queue.h" +#include "util/simple_mtx.h" #include "ac_binary.h" #include "ac_llvm_build.h" @@ -159,6 +157,8 @@ struct si_context; */ #define SI_MAX_IO_GENERIC 32 +#define SI_NGG_PRIM_EDGE_FLAG_BITS ((1 << 9) | (1 << 19) | (1 << 29)) + /* SGPR user data indices */ enum { SI_SGPR_RW_BUFFERS, /* rings (& stream-out, VS only) */ @@ -211,6 +211,11 @@ enum { /* PS only */ SI_SGPR_ALPHA_REF = SI_NUM_RESOURCE_SGPRS, SI_PS_NUM_USER_SGPR, + + /* The value has to be 12, because the hw requires that descriptors + * are aligned to 4 SGPRs. + */ + SI_SGPR_VS_VB_DESCRIPTOR_FIRST = 12, }; /* LLVM function parameter indices */ @@ -251,18 +256,13 @@ enum { #define C_VS_STATE_PROVOKING_VTX_INDEX 0xFFFFFFCF #define S_VS_STATE_STREAMOUT_QUERY_ENABLED(x) (((unsigned)(x) & 0x1) << 6) #define C_VS_STATE_STREAMOUT_QUERY_ENABLED 0xFFFFFFBF -#define S_VS_STATE_LS_OUT_PATCH_SIZE(x) (((unsigned)(x) & 0x1FFF) << 8) -#define C_VS_STATE_LS_OUT_PATCH_SIZE 0xFFE000FF +#define S_VS_STATE_SMALL_PRIM_PRECISION(x) (((unsigned)(x) & 0xF) << 7) +#define C_VS_STATE_SMALL_PRIM_PRECISION 0xFFFFF87F +#define S_VS_STATE_LS_OUT_PATCH_SIZE(x) (((unsigned)(x) & 0x1FFF) << 11) +#define C_VS_STATE_LS_OUT_PATCH_SIZE 0xFF0007FF #define S_VS_STATE_LS_OUT_VERTEX_SIZE(x) (((unsigned)(x) & 0xFF) << 24) #define C_VS_STATE_LS_OUT_VERTEX_SIZE 0x00FFFFFF -/* Driver-specific system values. */ -enum { - /* Values from set_tess_state. */ - TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI = TGSI_SEMANTIC_COUNT, - TGSI_SEMANTIC_DEFAULT_TESSINNER_SI, -}; - enum { /* Use a property enum that CS wouldn't use. */ TGSI_PROPERTY_CS_LOCAL_SIZE = TGSI_PROPERTY_FS_COORD_ORIGIN, @@ -273,6 +273,13 @@ enum { SI_VS_BLIT_SGPRS_POS_TEXCOORD = 9, }; +#define SI_NGG_CULL_VIEW_SMALLPRIMS (1 << 0) /* view.xy + small prims */ +#define SI_NGG_CULL_BACK_FACE (1 << 1) /* back faces */ +#define SI_NGG_CULL_FRONT_FACE (1 << 2) /* front faces */ +#define SI_NGG_CULL_GS_FAST_LAUNCH_TRI_LIST (1 << 3) /* GS fast launch: triangles */ +#define SI_NGG_CULL_GS_FAST_LAUNCH_TRI_STRIP (1 << 4) /* GS fast launch: triangle strip */ +#define SI_NGG_CULL_GS_FAST_LAUNCH_ALL (0x3 << 3) /* GS fast launch (both prim types) */ + /** * For VS shader keys, describe any fixups required for vertex fetch. * @@ -308,8 +315,93 @@ struct si_compiler_ctx_state { bool is_debug_context; }; +struct si_shader_info { + ubyte num_inputs; + ubyte num_outputs; + ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; /**< TGSI_SEMANTIC_x */ + ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS]; + ubyte input_interpolate[PIPE_MAX_SHADER_INPUTS]; + ubyte input_interpolate_loc[PIPE_MAX_SHADER_INPUTS]; + ubyte input_usage_mask[PIPE_MAX_SHADER_INPUTS]; + ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; /**< TGSI_SEMANTIC_x */ + ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; + ubyte output_usagemask[PIPE_MAX_SHADER_OUTPUTS]; + ubyte output_streams[PIPE_MAX_SHADER_OUTPUTS]; + + ubyte processor; + + int constbuf0_num_slots; + unsigned const_buffers_declared; /**< bitmask of declared const buffers */ + unsigned samplers_declared; /**< bitmask of declared samplers */ + ubyte num_stream_output_components[4]; + + uint num_memory_instructions; /**< sampler, buffer, and image instructions */ + + /** + * If a tessellation control shader reads outputs, this describes which ones. + */ + bool reads_pervertex_outputs; + bool reads_perpatch_outputs; + bool reads_tessfactor_outputs; + + ubyte colors_read; /**< which color components are read by the FS */ + ubyte colors_written; + bool reads_samplemask; /**< does fragment shader read sample mask? */ + bool reads_tess_factors; /**< If TES reads TESSINNER or TESSOUTER */ + bool writes_z; /**< does fragment shader write Z value? */ + bool writes_stencil; /**< does fragment shader write stencil value? */ + bool writes_samplemask; /**< does fragment shader write sample mask? */ + bool writes_edgeflag; /**< vertex shader outputs edgeflag */ + bool uses_kill; /**< KILL or KILL_IF instruction used? */ + bool uses_persp_center; + bool uses_persp_centroid; + bool uses_persp_sample; + bool uses_linear_center; + bool uses_linear_centroid; + bool uses_linear_sample; + bool uses_persp_opcode_interp_sample; + bool uses_linear_opcode_interp_sample; + bool uses_instanceid; + bool uses_vertexid; + bool uses_vertexid_nobase; + bool uses_basevertex; + bool uses_drawid; + bool uses_primid; + bool uses_frontface; + bool uses_invocationid; + bool uses_thread_id[3]; + bool uses_block_id[3]; + bool uses_block_size; + bool uses_grid_size; + bool uses_subgroup_info; + bool writes_position; + bool writes_psize; + bool writes_clipvertex; + bool writes_primid; + bool writes_viewport_index; + bool writes_layer; + bool writes_memory; /**< contains stores or atomics to buffers or images */ + bool uses_derivatives; + bool uses_bindless_samplers; + bool uses_bindless_images; + bool uses_fbfetch; + unsigned clipdist_writemask; + unsigned culldist_writemask; + unsigned num_written_culldistance; + unsigned num_written_clipdistance; + + unsigned images_declared; /**< bitmask of declared images */ + unsigned msaa_images_declared; /**< bitmask of declared MSAA images */ + unsigned shader_buffers_declared; /**< bitmask of declared shader buffers */ + + unsigned properties[TGSI_PROPERTY_COUNT]; /* index with TGSI_PROPERTY_ */ + + /** Whether all codepaths write tess factors in all invocations. */ + bool tessfactors_are_def_in_all_invocs; +}; + /* A shader selector is a gallium CSO and contains shader variants and - * binaries for one TGSI program. This can be shared by multiple contexts. + * binaries for one NIR program. This can be shared by multiple contexts. */ struct si_shader_selector { struct pipe_reference reference; @@ -317,12 +409,12 @@ struct si_shader_selector { struct util_queue_fence ready; struct si_compiler_ctx_state compiler_ctx_state; - mtx_t mutex; + simple_mtx_t mutex; struct si_shader *first_variant; /* immutable after the first variant */ struct si_shader *last_variant; /* mutable */ - /* The compiled TGSI shader expecting a prolog and/or epilog (not - * uploaded to a buffer). + /* The compiled NIR shader without a prolog and/or epilog (not + * uploaded to a buffer object). */ struct si_shader *main_shader_part; struct si_shader *main_shader_part_ls; /* as_ls is set in the key */ @@ -332,19 +424,21 @@ struct si_shader_selector { struct si_shader *gs_copy_shader; - struct tgsi_token *tokens; struct nir_shader *nir; + void *nir_binary; + unsigned nir_size; + struct pipe_stream_output_info so; - struct tgsi_shader_info info; - struct tgsi_tessctrl_info tcs_info; + struct si_shader_info info; /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */ enum pipe_shader_type type; bool vs_needs_prolog; bool force_correct_derivs_after_kill; bool prim_discard_cs_allowed; - bool ngg_writes_edgeflag; - bool pos_writes_edgeflag; + bool ngg_culling_allowed; + unsigned num_vs_inputs; + unsigned num_vbos_in_user_sgprs; unsigned pa_cl_vs_out_cntl; ubyte clipdist_mask; ubyte culldist_mask; @@ -468,10 +562,13 @@ union si_shader_part_key { unsigned num_input_sgprs:6; /* For merged stages such as LS-HS, HS input VGPRs are first. */ unsigned num_merged_next_stage_vgprs:3; - unsigned last_input:4; + unsigned num_inputs:5; unsigned as_ls:1; unsigned as_es:1; unsigned as_ngg:1; + unsigned has_ngg_cull_inputs:1; /* from the NGG cull shader */ + unsigned gs_fast_launch_tri_list:1; /* for NGG culling */ + unsigned gs_fast_launch_tri_strip:1; /* for NGG culling */ /* Prologs for monolithic shaders shouldn't set EXEC. */ unsigned is_monolithic:1; } vs_prolog; @@ -562,6 +659,9 @@ struct si_shader_key { uint64_t kill_outputs; /* "get_unique_index" bits */ unsigned clip_disable:1; + /* For NGG VS and TES. */ + unsigned ngg_culling:5; /* SI_NGG_CULL_* */ + /* For shaders where monolithic variants have better code. * * This is a flag that has no effect on code generation, @@ -589,7 +689,7 @@ struct si_shader_key { #pragma pack(pop) /* GCN-specific shader info. */ -struct si_shader_info { +struct si_shader_binary_info { ubyte vs_output_param_offset[SI_MAX_VS_OUTPUTS]; ubyte num_input_sgprs; ubyte num_input_vgprs; @@ -643,7 +743,7 @@ struct si_shader { /* The following data is all that's needed for binary shaders. */ struct si_shader_binary binary; struct ac_shader_config config; - struct si_shader_info info; + struct si_shader_binary_info info; struct { uint16_t ngg_emit_size; /* in dwords */ @@ -693,6 +793,7 @@ struct si_shader { unsigned pa_cl_vte_cntl; unsigned pa_cl_ngg_cntl; unsigned vgt_gs_max_vert_out; /* for API GS */ + unsigned ge_pc_alloc; /* uconfig register */ } ngg; struct { @@ -702,6 +803,7 @@ struct si_shader { unsigned spi_vs_out_config; unsigned spi_shader_pos_format; unsigned pa_cl_vte_cntl; + unsigned ge_pc_alloc; /* uconfig register */ } vs; struct { @@ -718,6 +820,7 @@ struct si_shader { /*For save precompute registers value */ unsigned vgt_tf_param; /* VGT_TF_PARAM */ unsigned vgt_vertex_reuse_block_cntl; /* VGT_VERTEX_REUSE_BLOCK_CNTL */ + unsigned pa_cl_vs_out_cntl; unsigned ge_cntl; }; @@ -729,18 +832,14 @@ struct si_shader_part { }; /* si_shader.c */ -struct si_shader * -si_generate_gs_copy_shader(struct si_screen *sscreen, - struct ac_llvm_compiler *compiler, - struct si_shader_selector *gs_selector, - struct pipe_debug_callback *debug); -int si_compile_tgsi_shader(struct si_screen *sscreen, - struct ac_llvm_compiler *compiler, - struct si_shader *shader, - struct pipe_debug_callback *debug); -bool si_shader_create(struct si_screen *sscreen, struct ac_llvm_compiler *compiler, - struct si_shader *shader, - struct pipe_debug_callback *debug); +int si_compile_shader(struct si_screen *sscreen, + struct ac_llvm_compiler *compiler, + struct si_shader *shader, + struct pipe_debug_callback *debug); +bool si_create_shader_variant(struct si_screen *sscreen, + struct ac_llvm_compiler *compiler, + struct si_shader *shader, + struct pipe_debug_callback *debug); void si_shader_destroy(struct si_shader *shader); unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name, unsigned index); unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index, @@ -758,14 +857,18 @@ void si_multiwave_lds_size_workaround(struct si_screen *sscreen, const char *si_get_shader_name(const struct si_shader *shader); void si_shader_binary_clean(struct si_shader_binary *binary); +/* si_shader_llvm_gs.c */ +struct si_shader * +si_generate_gs_copy_shader(struct si_screen *sscreen, + struct ac_llvm_compiler *compiler, + struct si_shader_selector *gs_selector, + struct pipe_debug_callback *debug); + /* si_shader_nir.c */ void si_nir_scan_shader(const struct nir_shader *nir, - struct tgsi_shader_info *info); -void si_nir_scan_tess_ctrl(const struct nir_shader *nir, - struct tgsi_tessctrl_info *out); -void si_nir_lower_ps_inputs(struct nir_shader *nir); -void si_lower_nir(struct si_shader_selector *sel, unsigned wave_size); -void si_nir_opts(struct nir_shader *nir); + struct si_shader_info *info); +void si_nir_adjust_driver_locations(struct nir_shader *nir); +void si_finalize_nir(struct pipe_screen *screen, void *nirptr, bool optimize); /* si_state_shaders.c */ void gfx9_get_gs_info(struct si_shader_selector *es, @@ -790,6 +893,19 @@ si_get_main_shader_part(struct si_shader_selector *sel, return &sel->main_shader_part; } +static inline bool +gfx10_is_ngg_passthrough(struct si_shader *shader) +{ + struct si_shader_selector *sel = shader->selector; + + return sel->type != PIPE_SHADER_GEOMETRY && + !sel->so.num_outputs && + !sel->info.writes_edgeflag && + !shader->key.opt.ngg_culling && + (sel->type != PIPE_SHADER_VERTEX || + !shader->key.mono.u.vs_export_prim_id); +} + static inline bool si_shader_uses_bindless_samplers(struct si_shader_selector *selector) {