src/gallium/drivers/radeonsi/si_shader_internal.h

   1 /*
   2  * Copyright 2016 Advanced Micro Devices, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25 #ifndef SI_SHADER_PRIVATE_H
  26 #define SI_SHADER_PRIVATE_H
  27
  28 #include "si_shader.h"
  29 #include "gallivm/lp_bld_flow.h"
  30 #include "gallivm/lp_bld_init.h"
  31 #include "gallivm/lp_bld_tgsi.h"
  32 #include "tgsi/tgsi_parse.h"
  33 #include "ac_shader_abi.h"
  34
  35 #include <llvm-c/Core.h>
  36 #include <llvm-c/TargetMachine.h>
  37
  38 struct pipe_debug_callback;
  39
  40 #define RADEON_LLVM_MAX_INPUT_SLOTS 32
  41 #define RADEON_LLVM_MAX_INPUTS 32 * 4
  42 #define RADEON_LLVM_MAX_OUTPUTS 32 * 4
  43
  44 #define RADEON_LLVM_MAX_SYSTEM_VALUES 11
  45 #define RADEON_LLVM_MAX_ADDRS 16
  46
  47 enum si_arg_regfile {
  48         ARG_SGPR,
  49         ARG_VGPR
  50 };
  51
  52 /**
  53  * Used to collect types and other info about arguments of the LLVM function
  54  * before the function is created.
  55  */
  56 struct si_function_info {
  57         LLVMTypeRef types[100];
  58         LLVMValueRef *assign[100];
  59         unsigned num_sgpr_params;
  60         unsigned num_params;
  61 };
  62
  63 struct si_shader_context {
  64         struct lp_build_tgsi_context bld_base;
  65         struct gallivm_state gallivm;
  66         struct ac_llvm_context ac;
  67         struct si_shader *shader;
  68         struct si_screen *screen;
  69
  70         unsigned type; /* PIPE_SHADER_* specifies the type of shader. */
  71
  72         /* For clamping the non-constant index in resource indexing: */
  73         unsigned num_const_buffers;
  74         unsigned num_shader_buffers;
  75         unsigned num_images;
  76         unsigned num_samplers;
  77
  78         struct ac_shader_abi abi;
  79
  80         /** This function is responsible for initilizing the inputs array and will be
  81           * called once for each input declared in the TGSI shader.
  82           */
  83         void (*load_input)(struct si_shader_context *,
  84                            unsigned input_index,
  85                            const struct tgsi_full_declaration *decl,
  86                            LLVMValueRef out[4]);
  87
  88         /** This array contains the input values for the shader.  Typically these
  89           * values will be in the form of a target intrinsic that will inform the
  90           * backend how to load the actual inputs to the shader.
  91           */
  92         struct tgsi_full_declaration input_decls[RADEON_LLVM_MAX_INPUT_SLOTS];
  93         LLVMValueRef inputs[RADEON_LLVM_MAX_INPUTS];
  94         LLVMValueRef outputs[RADEON_LLVM_MAX_OUTPUTS][TGSI_NUM_CHANNELS];
  95         LLVMValueRef addrs[RADEON_LLVM_MAX_ADDRS][TGSI_NUM_CHANNELS];
  96
  97         /** This pointer is used to contain the temporary values.
  98           * The amount of temporary used in tgsi can't be bound to a max value and
  99           * thus we must allocate this array at runtime.
 100           */
 101         LLVMValueRef *temps;
 102         unsigned temps_count;
 103         LLVMValueRef system_values[RADEON_LLVM_MAX_SYSTEM_VALUES];
 104
 105         LLVMValueRef *imms;
 106         unsigned imms_num;
 107
 108         struct lp_build_if_state merged_wrap_if_state;
 109
 110         struct tgsi_array_info *temp_arrays;
 111         LLVMValueRef *temp_array_allocas;
 112
 113         LLVMValueRef undef_alloca;
 114
 115         LLVMValueRef main_fn;
 116         LLVMTypeRef return_type;
 117
 118         /* Parameter indices for LLVMGetParam. */
 119         int param_rw_buffers;
 120         int param_const_and_shader_buffers;
 121         int param_samplers_and_images;
 122         int param_bindless_samplers_and_images;
 123         /* Common inputs for merged shaders. */
 124         int param_merged_wave_info;
 125         int param_merged_scratch_offset;
 126         /* API VS */
 127         int param_vertex_buffers;
 128         int param_rel_auto_id;
 129         int param_vs_prim_id;
 130         int param_vertex_index0;
 131         /* VS states and layout of LS outputs / TCS inputs at the end
 132          *   [0] = clamp vertex color
 133          *   [1] = indexed
 134          *   [8:20] = stride between patches in DW = num_inputs * num_vertices * 4
 135          *            max = 32*32*4 + 32*4
 136          *   [24:31] = stride between vertices in DW = num_inputs * 4
 137          *             max = 32*4
 138          */
 139         int param_vs_state_bits;
 140         int param_vs_blit_inputs;
 141         /* HW VS */
 142         int param_streamout_config;
 143         int param_streamout_write_index;
 144         int param_streamout_offset[4];
 145
 146         /* API TCS & TES */
 147         /* Layout of TCS outputs in the offchip buffer
 148          * # 6 bits
 149          *   [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
 150          * # 6 bits
 151          *   [6:11] = the number of output vertices per patch, max = 32
 152          * # 20 bits
 153          *   [12:31] = the offset of per patch attributes in the buffer in bytes.
 154          *             max = NUM_PATCHES*32*32*16
 155          */
 156         int param_tcs_offchip_layout;
 157
 158         /* API TCS */
 159         /* Offsets where TCS outputs and TCS patch outputs live in LDS:
 160          *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
 161          *   [16:31] = TCS output patch0 offset for per-patch / 16
 162          *             max = (NUM_PATCHES + 1) * 32*32
 163          */
 164         int param_tcs_out_lds_offsets;
 165         /* Layout of TCS outputs / TES inputs:
 166          *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
 167          *            max = 32*32*4 + 32*4
 168          *   [13:18] = gl_PatchVerticesIn, max = 32
 169          *   [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
 170          */
 171         int param_tcs_out_lds_layout;
 172         int param_tcs_offchip_offset;
 173         int param_tcs_factor_offset;
 174
 175         /* API TES */
 176         int param_tes_offchip_addr;
 177         int param_tes_u;
 178         int param_tes_v;
 179         int param_tes_rel_patch_id;
 180         /* HW ES */
 181         int param_es2gs_offset;
 182         /* API GS */
 183         int param_gs2vs_offset;
 184         int param_gs_wave_id; /* GFX6 */
 185         LLVMValueRef gs_vtx_offset[6]; /* in dwords (GFX6) */
 186         int param_gs_vtx01_offset; /* in dwords (GFX9) */
 187         int param_gs_vtx23_offset; /* in dwords (GFX9) */
 188         int param_gs_vtx45_offset; /* in dwords (GFX9) */
 189         /* CS */
 190         int param_block_size;
 191         int param_cs_user_data;
 192
 193         struct ac_llvm_compiler *compiler;
 194
 195         /* Preloaded descriptors. */
 196         LLVMValueRef esgs_ring;
 197         LLVMValueRef gsvs_ring[4];
 198         LLVMValueRef tess_offchip_ring;
 199
 200         LLVMValueRef invoc0_tess_factors[6]; /* outer[4], inner[2] */
 201         LLVMValueRef gs_next_vertex[4];
 202         LLVMValueRef postponed_kill;
 203         LLVMValueRef return_value;
 204
 205         LLVMTypeRef voidt;
 206         LLVMTypeRef i1;
 207         LLVMTypeRef i8;
 208         LLVMTypeRef i32;
 209         LLVMTypeRef i64;
 210         LLVMTypeRef i128;
 211         LLVMTypeRef f32;
 212         LLVMTypeRef v2i32;
 213         LLVMTypeRef v4i32;
 214         LLVMTypeRef v4f32;
 215         LLVMTypeRef v8i32;
 216
 217         LLVMValueRef i32_0;
 218         LLVMValueRef i32_1;
 219         LLVMValueRef i1false;
 220         LLVMValueRef i1true;
 221 };
 222
 223 static inline struct si_shader_context *
 224 si_shader_context(struct lp_build_tgsi_context *bld_base)
 225 {
 226         return (struct si_shader_context*)bld_base;
 227 }
 228
 229 static inline struct si_shader_context *
 230 si_shader_context_from_abi(struct ac_shader_abi *abi)
 231 {
 232         struct si_shader_context *ctx = NULL;
 233         return container_of(abi, ctx, abi);
 234 }
 235
 236 void si_init_function_info(struct si_function_info *fninfo);
 237 unsigned add_arg_assign(struct si_function_info *fninfo,
 238                         enum si_arg_regfile regfile, LLVMTypeRef type,
 239                         LLVMValueRef *assign);
 240 void si_create_function(struct si_shader_context *ctx,
 241                         const char *name,
 242                         LLVMTypeRef *returns, unsigned num_returns,
 243                         struct si_function_info *fninfo,
 244                         unsigned max_workgroup_size);
 245 unsigned si_llvm_compile(LLVMModuleRef M, struct si_shader_binary *binary,
 246                          struct ac_llvm_compiler *compiler,
 247                          struct pipe_debug_callback *debug,
 248                          bool less_optimized);
 249
 250 LLVMTypeRef tgsi2llvmtype(struct lp_build_tgsi_context *bld_base,
 251                           enum tgsi_opcode_type type);
 252
 253 LLVMValueRef bitcast(struct lp_build_tgsi_context *bld_base,
 254                      enum tgsi_opcode_type type, LLVMValueRef value);
 255
 256 LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx,
 257                                  LLVMValueRef index,
 258                                  unsigned num);
 259
 260 void si_llvm_context_init(struct si_shader_context *ctx,
 261                           struct si_screen *sscreen,
 262                           struct ac_llvm_compiler *compiler);
 263 void si_llvm_context_set_tgsi(struct si_shader_context *ctx,
 264                               struct si_shader *shader);
 265
 266 void si_llvm_create_func(struct si_shader_context *ctx,
 267                          const char *name,
 268                          LLVMTypeRef *return_types, unsigned num_return_elems,
 269                          LLVMTypeRef *ParamTypes, unsigned ParamCount);
 270
 271 void si_llvm_dispose(struct si_shader_context *ctx);
 272
 273 void si_llvm_optimize_module(struct si_shader_context *ctx);
 274
 275 LLVMValueRef si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base,
 276                                       LLVMTypeRef type,
 277                                       LLVMValueRef ptr,
 278                                       LLVMValueRef ptr2);
 279
 280 LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
 281                                 const struct tgsi_full_src_register *reg,
 282                                 enum tgsi_opcode_type type,
 283                                 unsigned swizzle);
 284
 285 void si_llvm_emit_kill(struct ac_shader_abi *abi, LLVMValueRef visible);
 286
 287 LLVMValueRef si_nir_load_input_tes(struct ac_shader_abi *abi,
 288                                    LLVMTypeRef type,
 289                                    LLVMValueRef vertex_index,
 290                                    LLVMValueRef param_index,
 291                                    unsigned const_index,
 292                                    unsigned location,
 293                                    unsigned driver_location,
 294                                    unsigned component,
 295                                    unsigned num_components,
 296                                    bool is_patch,
 297                                    bool is_compact,
 298                                    bool load_input);
 299
 300 LLVMValueRef si_llvm_load_input_gs(struct ac_shader_abi *abi,
 301                                    unsigned input_index,
 302                                    unsigned vtx_offset_param,
 303                                    LLVMTypeRef type,
 304                                    unsigned swizzle);
 305
 306 LLVMValueRef si_nir_lookup_interp_param(struct ac_shader_abi *abi,
 307                                         enum glsl_interp_mode interp,
 308                                         unsigned location);
 309
 310 void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
 311                         const struct tgsi_full_instruction *inst,
 312                         const struct tgsi_opcode_info *info,
 313                         unsigned index,
 314                         LLVMValueRef dst[4]);
 315
 316 LLVMValueRef si_get_indirect_index(struct si_shader_context *ctx,
 317                                    const struct tgsi_ind_register *ind,
 318                                    unsigned addr_mul, int rel_index);
 319 LLVMValueRef si_get_bounded_indirect_index(struct si_shader_context *ctx,
 320                                            const struct tgsi_ind_register *ind,
 321                                            int rel_index, unsigned num);
 322 LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
 323
 324 void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base);
 325 void si_shader_context_init_mem(struct si_shader_context *ctx);
 326
 327 LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx,
 328                                   LLVMValueRef list, LLVMValueRef index,
 329                                   enum ac_descriptor_type type);
 330 LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 331                                 LLVMValueRef list, LLVMValueRef index,
 332                                 enum ac_descriptor_type desc_type, bool dcc_off,
 333                                 bool bindless);
 334
 335 void si_load_system_value(struct si_shader_context *ctx,
 336                           unsigned index,
 337                           const struct tgsi_full_declaration *decl);
 338 void si_declare_compute_memory(struct si_shader_context *ctx);
 339 void si_tgsi_declare_compute_memory(struct si_shader_context *ctx,
 340                                     const struct tgsi_full_declaration *decl);
 341
 342 void si_llvm_load_input_vs(
 343         struct si_shader_context *ctx,
 344         unsigned input_index,
 345         LLVMValueRef out[4]);
 346 void si_llvm_load_input_fs(
 347         struct si_shader_context *ctx,
 348         unsigned input_index,
 349         LLVMValueRef out[4]);
 350
 351 bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir);
 352
 353 LLVMValueRef si_unpack_param(struct si_shader_context *ctx,
 354                              unsigned param, unsigned rshift,
 355                              unsigned bitwidth);
 356
 357 #endif