X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fradeonsi%2Fsi_shader_internal.h;h=2aba488b35b4fcef7ebb557c81d6911517fa84b3;hb=85a6bcca615f9aae1ffd2a1e790ee5d980e7cc43;hp=ae93f78d2fbeaea43a453984a1bdd18bed98a26f;hpb=9df23db13d706a28190325b64e6c3f61eedc6206;p=mesa.git

diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index ae93f78d2fb..2aba488b35b 100644
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -1,5 +1,6 @@
 /*
  * Copyright 2016 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -24,289 +25,277 @@
 #ifndef SI_SHADER_PRIVATE_H
 #define SI_SHADER_PRIVATE_H
 
-#include "si_shader.h"
-#include "gallivm/lp_bld_flow.h"
-#include "gallivm/lp_bld_init.h"
-#include "gallivm/lp_bld_tgsi.h"
-#include "tgsi/tgsi_parse.h"
 #include "ac_shader_abi.h"
-#include "ac_llvm_util.h"
-#include "ac_llvm_build.h"
-
-#include <llvm-c/Core.h>
-#include <llvm-c/TargetMachine.h>
+#include "si_shader.h"
 
 struct pipe_debug_callback;
-struct ac_shader_binary;
 
-#define RADEON_LLVM_MAX_INPUT_SLOTS 32
 #define RADEON_LLVM_MAX_INPUTS 32 * 4
-#define RADEON_LLVM_MAX_OUTPUTS 32 * 4
-
-#define RADEON_LLVM_INITIAL_CF_DEPTH 4
 
-#define RADEON_LLVM_MAX_SYSTEM_VALUES 11
-#define RADEON_LLVM_MAX_ADDRS 16
+/* Ideally pass the sample mask input to the PS epilog as v14, which
+ * is its usual location, so that the shader doesn't have to add v_mov.
+ */
+#define PS_EPILOG_SAMPLEMASK_MIN_LOC 14
 
-struct si_llvm_flow;
+struct si_shader_output_values {
+   LLVMValueRef values[4];
+   unsigned semantic_name;
+   unsigned semantic_index;
+   ubyte vertex_stream[4];
+};
 
 struct si_shader_context {
-	struct lp_build_tgsi_context bld_base;
-	struct gallivm_state gallivm;
-	struct ac_llvm_context ac;
-	struct si_shader *shader;
-	struct si_screen *screen;
-
-	unsigned type; /* PIPE_SHADER_* specifies the type of shader. */
-
-	/* For clamping the non-constant index in resource indexing: */
-	unsigned num_const_buffers;
-	unsigned num_shader_buffers;
-	unsigned num_images;
-	unsigned num_samplers;
-
-	/* Whether the prolog will be compiled separately. */
-	bool separate_prolog;
-
-	struct ac_shader_abi abi;
-
-	/** This function is responsible for initilizing the inputs array and will be
-	  * called once for each input declared in the TGSI shader.
-	  */
-	void (*load_input)(struct si_shader_context *,
-			   unsigned input_index,
-			   const struct tgsi_full_declaration *decl,
-			   LLVMValueRef out[4]);
-
-	void (*load_system_value)(struct si_shader_context *,
-				  unsigned index,
-				  const struct tgsi_full_declaration *decl);
-
-	void (*declare_memory_region)(struct si_shader_context *,
-				      const struct tgsi_full_declaration *decl);
-
-	/** This array contains the input values for the shader.  Typically these
-	  * values will be in the form of a target intrinsic that will inform the
-	  * backend how to load the actual inputs to the shader.
-	  */
-	struct tgsi_full_declaration input_decls[RADEON_LLVM_MAX_INPUT_SLOTS];
-	LLVMValueRef inputs[RADEON_LLVM_MAX_INPUTS];
-	LLVMValueRef outputs[RADEON_LLVM_MAX_OUTPUTS][TGSI_NUM_CHANNELS];
-	LLVMValueRef addrs[RADEON_LLVM_MAX_ADDRS][TGSI_NUM_CHANNELS];
-
-	/** This pointer is used to contain the temporary values.
-	  * The amount of temporary used in tgsi can't be bound to a max value and
-	  * thus we must allocate this array at runtime.
-	  */
-	LLVMValueRef *temps;
-	unsigned temps_count;
-	LLVMValueRef system_values[RADEON_LLVM_MAX_SYSTEM_VALUES];
-
-	LLVMValueRef *imms;
-	unsigned imms_num;
-
-	struct si_llvm_flow *flow;
-	unsigned flow_depth;
-	unsigned flow_depth_max;
-
-	struct lp_build_if_state merged_wrap_if_state;
-
-	struct tgsi_array_info *temp_arrays;
-	LLVMValueRef *temp_array_allocas;
-
-	LLVMValueRef undef_alloca;
-
-	LLVMValueRef main_fn;
-	LLVMTypeRef return_type;
-
-	/* Parameter indices for LLVMGetParam. */
-	int param_rw_buffers;
-	int param_const_and_shader_buffers;
-	int param_samplers_and_images;
-	/* Common inputs for merged shaders. */
-	int param_merged_wave_info;
-	int param_merged_scratch_offset;
-	/* API VS */
-	int param_vertex_buffers;
-	int param_rel_auto_id;
-	int param_vs_prim_id;
-	int param_vertex_index0;
-	/* VS states and layout of LS outputs / TCS inputs at the end
-	 *   [0] = clamp vertex color
-	 *   [1] = indexed
-	 *   [8:20] = stride between patches in DW = num_inputs * num_vertices * 4
-	 *            max = 32*32*4 + 32*4
-	 *   [24:31] = stride between vertices in DW = num_inputs * 4
-	 *             max = 32*4
-	 */
-	int param_vs_state_bits;
-	/* HW VS */
-	int param_streamout_config;
-	int param_streamout_write_index;
-	int param_streamout_offset[4];
-
-	/* API TCS & TES */
-	/* Layout of TCS outputs in the offchip buffer
-	 * # 6 bits
-	 *   [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
-	 * # 6 bits
-	 *   [6:11] = the number of output vertices per patch, max = 32
-	 * # 20 bits
-	 *   [12:31] = the offset of per patch attributes in the buffer in bytes.
-	 *             max = NUM_PATCHES*32*32*16
-	 */
-	int param_tcs_offchip_layout;
-
-	/* API TCS */
-	/* Offsets where TCS outputs and TCS patch outputs live in LDS:
-	 *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
-	 *   [16:31] = TCS output patch0 offset for per-patch / 16
-	 *             max = (NUM_PATCHES + 1) * 32*32
-	 */
-	int param_tcs_out_lds_offsets;
-	/* Layout of TCS outputs / TES inputs:
-	 *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
-	 *            max = 32*32*4 + 32*4
-	 *   [13:20] = stride between output vertices in DW = num_inputs * 4
-	 *             max = 32*4
-	 *   [26:31] = gl_PatchVerticesIn, max = 32
-	 */
-	int param_tcs_out_lds_layout;
-	int param_tcs_offchip_addr_base64k;
-	int param_tcs_factor_addr_base64k;
-	int param_tcs_offchip_offset;
-	int param_tcs_factor_offset;
-	int param_tcs_patch_id;
-	int param_tcs_rel_ids;
-
-	/* API TES */
-	int param_tes_u;
-	int param_tes_v;
-	int param_tes_rel_patch_id;
-	int param_tes_patch_id;
-	/* HW ES */
-	int param_es2gs_offset;
-	/* API GS */
-	int param_gs2vs_offset;
-	int param_gs_wave_id; /* GFX6 */
-	int param_gs_vtx0_offset; /* in dwords (GFX6) */
-	int param_gs_vtx1_offset; /* in dwords (GFX6) */
-	int param_gs_prim_id;
-	int param_gs_vtx2_offset; /* in dwords (GFX6) */
-	int param_gs_vtx3_offset; /* in dwords (GFX6) */
-	int param_gs_vtx4_offset; /* in dwords (GFX6) */
-	int param_gs_vtx5_offset; /* in dwords (GFX6) */
-	int param_gs_instance_id;
-	int param_gs_vtx01_offset; /* in dwords (GFX9) */
-	int param_gs_vtx23_offset; /* in dwords (GFX9) */
-	int param_gs_vtx45_offset; /* in dwords (GFX9) */
-	/* CS */
-	int param_grid_size;
-	int param_block_size;
-	int param_block_id[3];
-	int param_thread_id;
-
-	LLVMTargetMachineRef tm;
-
-	unsigned range_md_kind;
-	unsigned fpmath_md_kind;
-	LLVMValueRef fpmath_md_2p5_ulp;
-
-	/* Preloaded descriptors. */
-	LLVMValueRef esgs_ring;
-	LLVMValueRef gsvs_ring[4];
-
-	LLVMValueRef lds;
-	LLVMValueRef gs_next_vertex[4];
-	LLVMValueRef postponed_kill;
-	LLVMValueRef return_value;
-
-	LLVMTypeRef voidt;
-	LLVMTypeRef i1;
-	LLVMTypeRef i8;
-	LLVMTypeRef i32;
-	LLVMTypeRef i64;
-	LLVMTypeRef i128;
-	LLVMTypeRef f32;
-	LLVMTypeRef v2i32;
-	LLVMTypeRef v4i32;
-	LLVMTypeRef v4f32;
-	LLVMTypeRef v8i32;
-
-	LLVMValueRef i32_0;
-	LLVMValueRef i32_1;
-
-	LLVMValueRef shared_memory;
+   struct ac_llvm_context ac;
+   struct si_shader *shader;
+   struct si_screen *screen;
+
+   unsigned type; /* PIPE_SHADER_* specifies the type of shader. */
+
+   /* For clamping the non-constant index in resource indexing: */
+   unsigned num_const_buffers;
+   unsigned num_shader_buffers;
+   unsigned num_images;
+   unsigned num_samplers;
+
+   struct ac_shader_args args;
+   struct ac_shader_abi abi;
+
+   LLVMValueRef inputs[RADEON_LLVM_MAX_INPUTS];
+
+   LLVMBasicBlockRef merged_wrap_if_entry_block;
+   int merged_wrap_if_label;
+
+   LLVMValueRef main_fn;
+   LLVMTypeRef return_type;
+
+   struct ac_arg const_and_shader_buffers;
+   struct ac_arg samplers_and_images;
+
+   /* For merged shaders, the per-stage descriptors for the stage other
+    * than the one we're processing, used to pass them through from the
+    * first stage to the second.
+    */
+   struct ac_arg other_const_and_shader_buffers;
+   struct ac_arg other_samplers_and_images;
+
+   struct ac_arg rw_buffers;
+   struct ac_arg bindless_samplers_and_images;
+   /* Common inputs for merged shaders. */
+   struct ac_arg merged_wave_info;
+   struct ac_arg merged_scratch_offset;
+   struct ac_arg small_prim_cull_info;
+   /* API VS */
+   struct ac_arg vertex_buffers;
+   struct ac_arg vb_descriptors[5];
+   struct ac_arg rel_auto_id;
+   struct ac_arg vs_prim_id;
+   struct ac_arg vertex_index0;
+   /* VS states and layout of LS outputs / TCS inputs at the end
+    *   [0] = clamp vertex color
+    *   [1] = indexed
+    *   [2:3] = NGG: output primitive type
+    *   [4:5] = NGG: provoking vertex index
+    *   [6]   = NGG: streamout queries enabled
+    *   [7:10] = NGG: small prim filter precision = num_samples / quant_mode,
+    *            but in reality it's: 1/2^n, from 1/16 to 1/4096 = 1/2^4 to 1/2^12
+    *            Only the first 4 bits of the exponent are stored.
+    *            Set it like this: (fui(num_samples / quant_mode) >> 23)
+    *            Expand to FP32 like this: ((0x70 | value) << 23);
+    *            With 0x70 = 112, we get 2^(112 + value - 127) = 2^(value - 15)
+    *            = 1/2^(15 - value) in FP32
+    *   [11:23] = stride between patches in DW = num_inputs * num_vertices * 4
+    *             max = 32*32*4 + 32*4
+    *   [24:31] = stride between vertices in DW = num_inputs * 4
+    *             max = 32*4
+    */
+   struct ac_arg vs_state_bits;
+   struct ac_arg vs_blit_inputs;
+   struct ac_arg ngg_old_thread_id; /* generated by the NGG cull shader */
+   /* HW VS */
+   struct ac_arg streamout_config;
+   struct ac_arg streamout_write_index;
+   struct ac_arg streamout_offset[4];
+
+   /* API TCS & TES */
+   /* Layout of TCS outputs in the offchip buffer
+    * # 6 bits
+    *   [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
+    * # 6 bits
+    *   [6:11] = the number of output vertices per patch, max = 32
+    * # 20 bits
+    *   [12:31] = the offset of per patch attributes in the buffer in bytes.
+    *             max = NUM_PATCHES*32*32*16
+    */
+   struct ac_arg tcs_offchip_layout;
+
+   /* API TCS */
+   /* Offsets where TCS outputs and TCS patch outputs live in LDS:
+    *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
+    *   [16:31] = TCS output patch0 offset for per-patch / 16
+    *             max = (NUM_PATCHES + 1) * 32*32
+    */
+   struct ac_arg tcs_out_lds_offsets;
+   /* Layout of TCS outputs / TES inputs:
+    *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
+    *            max = 32*32*4 + 32*4
+    *   [13:18] = gl_PatchVerticesIn, max = 32
+    *   [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
+    */
+   struct ac_arg tcs_out_lds_layout;
+   struct ac_arg tcs_offchip_offset;
+   struct ac_arg tcs_factor_offset;
+
+   /* API TES */
+   struct ac_arg tes_offchip_addr;
+   struct ac_arg tes_u;
+   struct ac_arg tes_v;
+   struct ac_arg tes_rel_patch_id;
+   /* HW ES */
+   struct ac_arg es2gs_offset;
+   /* HW GS */
+   /* On gfx10:
+    *  - bits 0..11: ordered_wave_id
+    *  - bits 12..20: number of vertices in group
+    *  - bits 22..30: number of primitives in group
+    */
+   struct ac_arg gs_tg_info;
+   /* API GS */
+   struct ac_arg gs2vs_offset;
+   struct ac_arg gs_wave_id;       /* GFX6 */
+   struct ac_arg gs_vtx_offset[6]; /* in dwords (GFX6) */
+   struct ac_arg gs_vtx01_offset;  /* in dwords (GFX9) */
+   struct ac_arg gs_vtx23_offset;  /* in dwords (GFX9) */
+   struct ac_arg gs_vtx45_offset;  /* in dwords (GFX9) */
+   /* PS */
+   struct ac_arg pos_fixed_pt;
+   /* CS */
+   struct ac_arg block_size;
+   struct ac_arg cs_user_data;
+   struct ac_arg cs_shaderbuf[3];
+   struct ac_arg cs_image[3];
+
+   struct ac_llvm_compiler *compiler;
+
+   /* Preloaded descriptors. */
+   LLVMValueRef esgs_ring;
+   LLVMValueRef gsvs_ring[4];
+   LLVMValueRef tess_offchip_ring;
+
+   LLVMValueRef invoc0_tess_factors[6]; /* outer[4], inner[2] */
+   LLVMValueRef gs_next_vertex[4];
+   LLVMValueRef gs_curprim_verts[4];
+   LLVMValueRef gs_generated_prims[4];
+   LLVMValueRef gs_ngg_emit;
+   LLVMValueRef gs_ngg_scratch;
+   LLVMValueRef postponed_kill;
+   LLVMValueRef return_value;
 };
 
-static inline struct si_shader_context *
-si_shader_context(struct lp_build_tgsi_context *bld_base)
+static inline struct si_shader_context *si_shader_context_from_abi(struct ac_shader_abi *abi)
 {
-	return (struct si_shader_context*)bld_base;
+   struct si_shader_context *ctx = NULL;
+   return container_of(abi, ctx, abi);
 }
 
-void si_llvm_add_attribute(LLVMValueRef F, const char *name, int value);
-
-unsigned si_llvm_compile(LLVMModuleRef M, struct ac_shader_binary *binary,
-			 LLVMTargetMachineRef tm,
-			 struct pipe_debug_callback *debug);
-
-LLVMTypeRef tgsi2llvmtype(struct lp_build_tgsi_context *bld_base,
-			  enum tgsi_opcode_type type);
-
-LLVMValueRef bitcast(struct lp_build_tgsi_context *bld_base,
-		     enum tgsi_opcode_type type, LLVMValueRef value);
-
-LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx,
-				 LLVMValueRef index,
-				 unsigned num);
-
-void si_llvm_context_init(struct si_shader_context *ctx,
-			  struct si_screen *sscreen,
-			  LLVMTargetMachineRef tm);
-void si_llvm_context_set_tgsi(struct si_shader_context *ctx,
-			      struct si_shader *shader);
-
-void si_llvm_create_func(struct si_shader_context *ctx,
-			 const char *name,
-			 LLVMTypeRef *return_types, unsigned num_return_elems,
-			 LLVMTypeRef *ParamTypes, unsigned ParamCount);
-
-void si_llvm_dispose(struct si_shader_context *ctx);
-
+bool si_is_multi_part_shader(struct si_shader *shader);
+bool si_is_merged_shader(struct si_shader *shader);
+void si_add_arg_checked(struct ac_shader_args *args, enum ac_arg_regfile file, unsigned registers,
+                        enum ac_arg_type type, struct ac_arg *arg, unsigned idx);
+unsigned si_get_max_workgroup_size(const struct si_shader *shader);
+bool si_need_ps_prolog(const union si_shader_part_key *key);
+void si_get_ps_prolog_key(struct si_shader *shader, union si_shader_part_key *key,
+                          bool separate_prolog);
+void si_get_ps_epilog_key(struct si_shader *shader, union si_shader_part_key *key);
+void si_fix_resource_usage(struct si_screen *sscreen, struct si_shader *shader);
+void si_create_function(struct si_shader_context *ctx, bool ngg_cull_shader);
+
+bool gfx10_ngg_export_prim_early(struct si_shader *shader);
+void gfx10_ngg_build_sendmsg_gs_alloc_req(struct si_shader_context *ctx);
+void gfx10_ngg_build_export_prim(struct si_shader_context *ctx, LLVMValueRef user_edgeflags[3],
+                                 LLVMValueRef prim_passthrough);
+void gfx10_emit_ngg_culling_epilogue_4x_wave32(struct ac_shader_abi *abi, unsigned max_outputs,
+                                               LLVMValueRef *addrs);
+void gfx10_emit_ngg_epilogue(struct ac_shader_abi *abi, unsigned max_outputs, LLVMValueRef *addrs);
+void gfx10_ngg_gs_emit_vertex(struct si_shader_context *ctx, unsigned stream, LLVMValueRef *addrs);
+void gfx10_ngg_gs_emit_prologue(struct si_shader_context *ctx);
+void gfx10_ngg_gs_emit_epilogue(struct si_shader_context *ctx);
+void gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
+
+/* si_shader_llvm.c */
+bool si_compile_llvm(struct si_screen *sscreen, struct si_shader_binary *binary,
+                     struct ac_shader_config *conf, struct ac_llvm_compiler *compiler,
+                     struct ac_llvm_context *ac, struct pipe_debug_callback *debug,
+                     enum pipe_shader_type shader_type, const char *name, bool less_optimized);
+void si_llvm_context_init(struct si_shader_context *ctx, struct si_screen *sscreen,
+                          struct ac_llvm_compiler *compiler, unsigned wave_size);
+void si_llvm_create_func(struct si_shader_context *ctx, const char *name, LLVMTypeRef *return_types,
+                         unsigned num_return_elems, unsigned max_workgroup_size);
 void si_llvm_optimize_module(struct si_shader_context *ctx);
-
-LLVMValueRef si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base,
-				      enum tgsi_opcode_type type,
-				      LLVMValueRef ptr,
-				      LLVMValueRef ptr2);
-
-LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
-				const struct tgsi_full_src_register *reg,
-				enum tgsi_opcode_type type,
-				unsigned swizzle);
-
-void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
-			const struct tgsi_full_instruction *inst,
-			const struct tgsi_opcode_info *info,
-			LLVMValueRef dst[4]);
-
-/* Combine these with & instead of |. */
-#define NOOP_WAITCNT 0xf7f
-#define LGKM_CNT 0x07f
-#define VM_CNT 0xf70
-
-void si_emit_waitcnt(struct si_shader_context *ctx, unsigned simm16);
-
-LLVMValueRef si_get_bounded_indirect_index(struct si_shader_context *ctx,
-					   const struct tgsi_ind_register *ind,
-					   int rel_index, unsigned num);
-
-LLVMTypeRef si_const_array(LLVMTypeRef elem_type, int num_elements);
-
-void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base);
-void si_shader_context_init_mem(struct si_shader_context *ctx);
-
+void si_llvm_dispose(struct si_shader_context *ctx);
+LLVMValueRef si_buffer_load_const(struct si_shader_context *ctx, LLVMValueRef resource,
+                                  LLVMValueRef offset);
+void si_llvm_build_ret(struct si_shader_context *ctx, LLVMValueRef ret);
+LLVMValueRef si_insert_input_ret(struct si_shader_context *ctx, LLVMValueRef ret,
+                                 struct ac_arg param, unsigned return_index);
+LLVMValueRef si_insert_input_ret_float(struct si_shader_context *ctx, LLVMValueRef ret,
+                                       struct ac_arg param, unsigned return_index);
+LLVMValueRef si_insert_input_ptr(struct si_shader_context *ctx, LLVMValueRef ret,
+                                 struct ac_arg param, unsigned return_index);
+LLVMValueRef si_prolog_get_rw_buffers(struct si_shader_context *ctx);
+LLVMValueRef si_build_gather_64bit(struct si_shader_context *ctx, LLVMTypeRef type,
+                                   LLVMValueRef val1, LLVMValueRef val2);
+void si_llvm_emit_barrier(struct si_shader_context *ctx);
+void si_llvm_declare_esgs_ring(struct si_shader_context *ctx);
+void si_init_exec_from_input(struct si_shader_context *ctx, struct ac_arg param,
+                             unsigned bitoffset);
+LLVMValueRef si_unpack_param(struct si_shader_context *ctx, struct ac_arg param, unsigned rshift,
+                             unsigned bitwidth);
+LLVMValueRef si_get_primitive_id(struct si_shader_context *ctx, unsigned swizzle);
+LLVMValueRef si_llvm_get_block_size(struct ac_shader_abi *abi);
+void si_llvm_declare_compute_memory(struct si_shader_context *ctx);
 bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir);
+void si_build_wrapper_function(struct si_shader_context *ctx, LLVMValueRef *parts,
+                               unsigned num_parts, unsigned main_part,
+                               unsigned next_shader_first_part);
+
+/* si_shader_llvm_gs.c */
+LLVMValueRef si_is_es_thread(struct si_shader_context *ctx);
+LLVMValueRef si_is_gs_thread(struct si_shader_context *ctx);
+void si_llvm_emit_es_epilogue(struct ac_shader_abi *abi, unsigned max_outputs, LLVMValueRef *addrs);
+void si_preload_esgs_ring(struct si_shader_context *ctx);
+void si_preload_gs_rings(struct si_shader_context *ctx);
+void si_llvm_build_gs_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
+void si_llvm_init_gs_callbacks(struct si_shader_context *ctx);
+
+/* si_shader_llvm_tess.c */
+void si_llvm_preload_tes_rings(struct si_shader_context *ctx);
+void si_llvm_emit_ls_epilogue(struct ac_shader_abi *abi, unsigned max_outputs, LLVMValueRef *addrs);
+void si_llvm_build_tcs_epilog(struct si_shader_context *ctx, union si_shader_part_key *key);
+void si_llvm_init_tcs_callbacks(struct si_shader_context *ctx);
+void si_llvm_init_tes_callbacks(struct si_shader_context *ctx, bool ngg_cull_shader);
+
+/* si_shader_llvm_ps.c */
+LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
+void si_llvm_build_ps_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
+void si_llvm_build_ps_epilog(struct si_shader_context *ctx, union si_shader_part_key *key);
+void si_llvm_build_monolithic_ps(struct si_shader_context *ctx, struct si_shader *shader);
+void si_llvm_init_ps_callbacks(struct si_shader_context *ctx);
+
+/* si_shader_llvm_resources.c */
+void si_llvm_init_resource_callbacks(struct si_shader_context *ctx);
+
+/* si_shader_llvm_vs.c */
+void si_llvm_load_vs_inputs(struct si_shader_context *ctx, struct nir_shader *nir);
+void si_llvm_streamout_store_output(struct si_shader_context *ctx, LLVMValueRef const *so_buffers,
+                                    LLVMValueRef const *so_write_offsets,
+                                    struct pipe_stream_output *stream_out,
+                                    struct si_shader_output_values *shader_out);
+void si_llvm_emit_streamout(struct si_shader_context *ctx, struct si_shader_output_values *outputs,
+                            unsigned noutput, unsigned stream);
+void si_llvm_build_vs_exports(struct si_shader_context *ctx,
+                              struct si_shader_output_values *outputs, unsigned noutput);
+void si_llvm_emit_vs_epilogue(struct ac_shader_abi *abi, unsigned max_outputs, LLVMValueRef *addrs);
+void si_llvm_build_vs_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
+void si_llvm_init_vs_callbacks(struct si_shader_context *ctx, bool ngg_cull_shader);
 
 #endif