#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 <llvm-c/Core.h>
-#include <llvm-c/TargetMachine.h>
-
struct pipe_debug_callback;
-#define RADEON_LLVM_MAX_INPUT_SLOTS 32
#define RADEON_LLVM_MAX_INPUTS 32 * 4
-#define RADEON_LLVM_MAX_OUTPUTS 32 * 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_shader_output_values {
LLVMValueRef values[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;
struct ac_shader_args args;
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]);
-
- /** 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;
LLVMBasicBlockRef merged_wrap_if_entry_block;
int merged_wrap_if_label;
- struct tgsi_array_info *temp_arrays;
- LLVMValueRef *temp_array_allocas;
-
- LLVMValueRef undef_alloca;
-
LLVMValueRef main_fn;
LLVMTypeRef return_type;
/* 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
- * [8:20] = stride between patches in DW = num_inputs * num_vertices * 4
- * max = 32*32*4 + 32*4
+ * [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;
LLVMValueRef gs_ngg_scratch;
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 i1false;
- LLVMValueRef i1true;
};
-static inline struct si_shader_context *
-si_shader_context(struct lp_build_tgsi_context *bld_base)
-{
- return (struct si_shader_context*)bld_base;
-}
-
static inline struct si_shader_context *
si_shader_context_from_abi(struct ac_shader_abi *abi)
{
return container_of(abi, ctx, abi);
}
-void si_create_function(struct si_shader_context *ctx,
- const char *name,
- LLVMTypeRef *returns, unsigned num_returns,
- unsigned max_workgroup_size);
-unsigned si_llvm_compile(LLVMModuleRef M, struct si_shader_binary *binary,
- struct ac_llvm_compiler *compiler,
- struct pipe_debug_callback *debug,
- bool less_optimized, unsigned wave_size);
-
-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,
struct ac_llvm_compiler *compiler,
- unsigned wave_size,
- unsigned ballot_mask_bits);
-void si_llvm_context_set_ir(struct si_shader_context *ctx,
- struct si_shader *shader,
- struct nir_shader *nir);
-
-void si_llvm_create_func(struct si_shader_context *ctx,
- const char *name,
- LLVMTypeRef *return_types, unsigned num_return_elems);
+ 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_dispose(struct si_shader_context *ctx);
void si_llvm_optimize_module(struct si_shader_context *ctx);
-LLVMValueRef si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base,
- LLVMTypeRef 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_kill(struct ac_shader_abi *abi, LLVMValueRef visible);
-
LLVMValueRef si_nir_load_input_tes(struct ac_shader_abi *abi,
LLVMTypeRef type,
LLVMValueRef vertex_index,
bool is_patch,
bool is_compact,
bool load_input);
-
-LLVMValueRef si_llvm_load_input_gs(struct ac_shader_abi *abi,
- unsigned input_index,
- unsigned vtx_offset_param,
- LLVMTypeRef type,
- unsigned swizzle);
-
-LLVMValueRef si_nir_lookup_interp_param(struct ac_shader_abi *abi,
- enum glsl_interp_mode interp,
- unsigned location);
-
-void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
- const struct tgsi_full_instruction *inst,
- const struct tgsi_opcode_info *info,
- unsigned index,
- LLVMValueRef dst[4]);
-
-LLVMValueRef si_get_indirect_index(struct si_shader_context *ctx,
- const struct tgsi_ind_register *ind,
- unsigned addr_mul, int rel_index);
-LLVMValueRef si_get_bounded_indirect_index(struct si_shader_context *ctx,
- const struct tgsi_ind_register *ind,
- int rel_index, unsigned num);
+bool si_is_merged_shader(struct si_shader_context *ctx);
LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
-
-void si_shader_context_init_alu(struct si_shader_context *ctx);
-void si_shader_context_init_mem(struct si_shader_context *ctx);
-
-LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx,
- LLVMValueRef list, LLVMValueRef index,
- enum ac_descriptor_type type);
-LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
- LLVMValueRef list, LLVMValueRef index,
- enum ac_descriptor_type desc_type,
- bool uses_store, bool bindless);
-LLVMValueRef si_nir_emit_fbfetch(struct ac_shader_abi *abi);
-
-void si_load_system_value(struct si_shader_context *ctx,
- unsigned index,
- const struct tgsi_full_declaration *decl);
+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_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_declare_compute_memory(struct si_shader_context *ctx);
-void si_tgsi_declare_compute_memory(struct si_shader_context *ctx,
- const struct tgsi_full_declaration *decl);
-
LLVMValueRef si_get_primitive_id(struct si_shader_context *ctx,
unsigned swizzle);
void si_llvm_export_vs(struct si_shader_context *ctx,
LLVMValueRef const *so_write_offsets,
struct pipe_stream_output *stream_out,
struct si_shader_output_values *shader_out);
+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);
void si_llvm_load_input_vs(
struct si_shader_context *ctx,
unsigned input_index,
LLVMValueRef out[4]);
-void si_llvm_load_input_fs(
- struct si_shader_context *ctx,
- unsigned input_index,
- LLVMValueRef out[4]);
bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir);
LLVMValueRef si_unpack_param(struct si_shader_context *ctx,
struct ac_arg param, unsigned rshift,
unsigned bitwidth);
-LLVMValueRef si_is_es_thread(struct si_shader_context *ctx);
-LLVMValueRef si_is_gs_thread(struct si_shader_context *ctx);
-
+void si_build_wrapper_function(struct si_shader_context *ctx, LLVMValueRef *parts,
+ unsigned num_parts, unsigned main_part,
+ unsigned next_shader_first_part);
+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);
+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);
+int 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_fix_resource_usage(struct si_screen *sscreen, struct si_shader *shader);
+void si_llvm_emit_streamout(struct si_shader_context *ctx,
+ struct si_shader_output_values *outputs,
+ unsigned noutput, unsigned stream);
+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_epilogue(struct si_shader_context *ctx);
void gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
+/* 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);
+
+/* si_shader_llvm_ps.c */
+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);
+
#endif