radeonsi: fork tgsi_shader_info and tgsi_tessctrl_info
[mesa.git] / src / gallium / drivers / radeonsi / si_shader.h
index 1699560bc8ab6e22c1e7f4b660291e8ec688b812..565bcd6b7a24e1ce082a82d0ab24284501d70b63 100644 (file)
 
 #include <llvm-c/Core.h> /* LLVMModuleRef */
 #include <llvm-c/TargetMachine.h>
-#include "tgsi/tgsi_scan.h"
 #include "util/u_inlines.h"
 #include "util/u_queue.h"
 #include "util/simple_mtx.h"
@@ -307,6 +306,157 @@ struct si_compiler_ctx_state {
        bool                            is_debug_context;
 };
 
+struct si_shader_info {
+       uint num_tokens;
+
+       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 input_cylindrical_wrap[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 num_system_values;
+       ubyte system_value_semantic_name[PIPE_MAX_SHADER_INPUTS];
+
+       ubyte processor;
+
+       uint file_mask[TGSI_FILE_COUNT];  /**< bitmask of declared registers */
+       uint file_count[TGSI_FILE_COUNT];  /**< number of declared registers */
+       int file_max[TGSI_FILE_COUNT];  /**< highest index of declared registers */
+       int const_file_max[PIPE_MAX_CONSTANT_BUFFERS];
+       unsigned const_buffers_declared; /**< bitmask of declared const buffers */
+       unsigned samplers_declared; /**< bitmask of declared samplers */
+       ubyte sampler_targets[PIPE_MAX_SHADER_SAMPLER_VIEWS];  /**< TGSI_TEXTURE_x values */
+       ubyte sampler_type[PIPE_MAX_SHADER_SAMPLER_VIEWS]; /**< TGSI_RETURN_TYPE_x */
+       ubyte num_stream_output_components[4];
+
+       ubyte input_array_first[PIPE_MAX_SHADER_INPUTS];
+       ubyte input_array_last[PIPE_MAX_SHADER_INPUTS];
+       ubyte output_array_first[PIPE_MAX_SHADER_OUTPUTS];
+       ubyte output_array_last[PIPE_MAX_SHADER_OUTPUTS];
+       unsigned array_max[TGSI_FILE_COUNT];  /**< highest index array per register file */
+
+       uint immediate_count; /**< number of immediates declared */
+       uint num_instructions;
+       uint num_memory_instructions; /**< sampler, buffer, and image instructions */
+
+       uint opcode_count[TGSI_OPCODE_LAST];  /**< opcode histogram */
+
+       /**
+        * If a tessellation control shader reads outputs, this describes which ones.
+        */
+       boolean reads_pervertex_outputs;
+       boolean reads_perpatch_outputs;
+       boolean reads_tessfactor_outputs;
+
+       ubyte colors_read; /**< which color components are read by the FS */
+       ubyte colors_written;
+       boolean reads_position; /**< does fragment shader read position? */
+       boolean reads_z; /**< does fragment shader read depth? */
+       boolean reads_samplemask; /**< does fragment shader read sample mask? */
+       boolean reads_tess_factors; /**< If TES reads TESSINNER or TESSOUTER */
+       boolean writes_z;  /**< does fragment shader write Z value? */
+       boolean writes_stencil; /**< does fragment shader write stencil value? */
+       boolean writes_samplemask; /**< does fragment shader write sample mask? */
+       boolean writes_edgeflag; /**< vertex shader outputs edgeflag */
+       boolean uses_kill;  /**< KILL or KILL_IF instruction used? */
+       boolean uses_persp_center;
+       boolean uses_persp_centroid;
+       boolean uses_persp_sample;
+       boolean uses_linear_center;
+       boolean uses_linear_centroid;
+       boolean uses_linear_sample;
+       boolean uses_persp_opcode_interp_centroid;
+       boolean uses_persp_opcode_interp_offset;
+       boolean uses_persp_opcode_interp_sample;
+       boolean uses_linear_opcode_interp_centroid;
+       boolean uses_linear_opcode_interp_offset;
+       boolean uses_linear_opcode_interp_sample;
+       boolean uses_instanceid;
+       boolean uses_vertexid;
+       boolean uses_vertexid_nobase;
+       boolean uses_basevertex;
+       boolean uses_drawid;
+       boolean uses_primid;
+       boolean uses_frontface;
+       boolean uses_invocationid;
+       boolean uses_thread_id[3];
+       boolean uses_block_id[3];
+       boolean uses_block_size;
+       boolean uses_grid_size;
+       boolean uses_subgroup_info;
+       boolean writes_position;
+       boolean writes_psize;
+       boolean writes_clipvertex;
+       boolean writes_primid;
+       boolean writes_viewport_index;
+       boolean writes_layer;
+       boolean writes_memory; /**< contains stores or atomics to buffers or images */
+       boolean uses_doubles; /**< uses any of the double instructions */
+       boolean uses_derivatives;
+       boolean uses_bindless_samplers;
+       boolean uses_bindless_images;
+       boolean 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 */
+
+       /**
+        * Bitmask indicating which declared image is a buffer.
+        */
+       unsigned images_buffers;
+       unsigned images_load; /**< bitmask of images using loads */
+       unsigned images_store; /**< bitmask of images using stores */
+       unsigned images_atomic; /**< bitmask of images using atomics */
+       unsigned shader_buffers_declared; /**< bitmask of declared shader buffers */
+       unsigned shader_buffers_load; /**< bitmask of shader buffers using loads */
+       unsigned shader_buffers_store; /**< bitmask of shader buffers using stores */
+       unsigned shader_buffers_atomic; /**< bitmask of shader buffers using atomics */
+       bool uses_bindless_buffer_load;
+       bool uses_bindless_buffer_store;
+       bool uses_bindless_buffer_atomic;
+       bool uses_bindless_image_load;
+       bool uses_bindless_image_store;
+       bool uses_bindless_image_atomic;
+
+       /**
+        * Bitmask indicating which register files are accessed with
+        * indirect addressing.  The bits are (1 << TGSI_FILE_x), etc.
+        */
+       unsigned indirect_files;
+       /**
+        * Bitmask indicating which register files are read / written with
+        * indirect addressing.  The bits are (1 << TGSI_FILE_x).
+        */
+       unsigned indirect_files_read;
+       unsigned indirect_files_written;
+       unsigned dim_indirect_files; /**< shader resource indexing */
+       unsigned const_buffers_indirect; /**< const buffers using indirect addressing */
+
+       unsigned properties[TGSI_PROPERTY_COUNT]; /* index with TGSI_PROPERTY_ */
+
+       /**
+        * Max nesting limit of loops/if's
+        */
+       unsigned max_depth;
+};
+
+struct si_tessctrl_info {
+       /** 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 NIR program. This can be shared by multiple contexts.
  */
@@ -336,8 +486,8 @@ struct si_shader_selector {
        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;
+       struct si_tessctrl_info         tcs_info;
 
        /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
        enum pipe_shader_type type;
@@ -762,9 +912,9 @@ void si_shader_binary_clean(struct si_shader_binary *binary);
 
 /* si_shader_nir.c */
 void si_nir_scan_shader(const struct nir_shader *nir,
-                       struct tgsi_shader_info *info);
+                       struct si_shader_info *info);
 void si_nir_scan_tess_ctrl(const struct nir_shader *nir,
-                          struct tgsi_tessctrl_info *out);
+                          struct si_tessctrl_info *out);
 void si_nir_adjust_driver_locations(struct nir_shader *nir);
 void si_finalize_nir(struct pipe_screen *screen, void *nirptr, bool optimize);