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 "ac_shader_abi.h"
  30
  31 #include <llvm-c/Core.h>
  32 #include <llvm-c/TargetMachine.h>
  33
  34 struct pipe_debug_callback;
  35
  36 #define RADEON_LLVM_MAX_INPUTS 32 * 4
  37
  38 /* Ideally pass the sample mask input to the PS epilog as v14, which
  39  * is its usual location, so that the shader doesn't have to add v_mov.
  40  */
  41 #define PS_EPILOG_SAMPLEMASK_MIN_LOC 14
  42
  43 struct si_shader_output_values {
  44         LLVMValueRef values[4];
  45         unsigned semantic_name;
  46         unsigned semantic_index;
  47         ubyte vertex_stream[4];
  48 };
  49
  50 struct si_shader_context {
  51         struct ac_llvm_context ac;
  52         struct si_shader *shader;
  53         struct si_screen *screen;
  54
  55         unsigned type; /* PIPE_SHADER_* specifies the type of shader. */
  56
  57         /* For clamping the non-constant index in resource indexing: */
  58         unsigned num_const_buffers;
  59         unsigned num_shader_buffers;
  60         unsigned num_images;
  61         unsigned num_samplers;
  62
  63         struct ac_shader_args args;
  64         struct ac_shader_abi abi;
  65
  66         LLVMValueRef inputs[RADEON_LLVM_MAX_INPUTS];
  67
  68         LLVMBasicBlockRef merged_wrap_if_entry_block;
  69         int merged_wrap_if_label;
  70
  71         LLVMValueRef main_fn;
  72         LLVMTypeRef return_type;
  73
  74         struct ac_arg const_and_shader_buffers;
  75         struct ac_arg samplers_and_images;
  76
  77         /* For merged shaders, the per-stage descriptors for the stage other
  78          * than the one we're processing, used to pass them through from the
  79          * first stage to the second.
  80          */
  81         struct ac_arg other_const_and_shader_buffers;
  82         struct ac_arg other_samplers_and_images;
  83
  84         struct ac_arg rw_buffers;
  85         struct ac_arg bindless_samplers_and_images;
  86         /* Common inputs for merged shaders. */
  87         struct ac_arg merged_wave_info;
  88         struct ac_arg merged_scratch_offset;
  89         /* API VS */
  90         struct ac_arg vertex_buffers;
  91         struct ac_arg vb_descriptors[5];
  92         struct ac_arg rel_auto_id;
  93         struct ac_arg vs_prim_id;
  94         struct ac_arg vertex_index0;
  95         /* VS states and layout of LS outputs / TCS inputs at the end
  96          *   [0] = clamp vertex color
  97          *   [1] = indexed
  98          *   [2:3] = NGG: output primitive type
  99          *   [4:5] = NGG: provoking vertex index
 100          *   [6]   = NGG: streamout queries enabled
 101          *   [11:23] = stride between patches in DW = num_inputs * num_vertices * 4
 102          *             max = 32*32*4 + 32*4
 103          *   [24:31] = stride between vertices in DW = num_inputs * 4
 104          *             max = 32*4
 105          */
 106         struct ac_arg vs_state_bits;
 107         struct ac_arg vs_blit_inputs;
 108         /* HW VS */
 109         struct ac_arg streamout_config;
 110         struct ac_arg streamout_write_index;
 111         struct ac_arg streamout_offset[4];
 112
 113         /* API TCS & TES */
 114         /* Layout of TCS outputs in the offchip buffer
 115          * # 6 bits
 116          *   [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
 117          * # 6 bits
 118          *   [6:11] = the number of output vertices per patch, max = 32
 119          * # 20 bits
 120          *   [12:31] = the offset of per patch attributes in the buffer in bytes.
 121          *             max = NUM_PATCHES*32*32*16
 122          */
 123         struct ac_arg tcs_offchip_layout;
 124
 125         /* API TCS */
 126         /* Offsets where TCS outputs and TCS patch outputs live in LDS:
 127          *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
 128          *   [16:31] = TCS output patch0 offset for per-patch / 16
 129          *             max = (NUM_PATCHES + 1) * 32*32
 130          */
 131         struct ac_arg tcs_out_lds_offsets;
 132         /* Layout of TCS outputs / TES inputs:
 133          *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
 134          *            max = 32*32*4 + 32*4
 135          *   [13:18] = gl_PatchVerticesIn, max = 32
 136          *   [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
 137          */
 138         struct ac_arg tcs_out_lds_layout;
 139         struct ac_arg tcs_offchip_offset;
 140         struct ac_arg tcs_factor_offset;
 141
 142         /* API TES */
 143         struct ac_arg tes_offchip_addr;
 144         struct ac_arg tes_u;
 145         struct ac_arg tes_v;
 146         struct ac_arg tes_rel_patch_id;
 147         /* HW ES */
 148         struct ac_arg es2gs_offset;
 149         /* HW GS */
 150         /* On gfx10:
 151          *  - bits 0..11: ordered_wave_id
 152          *  - bits 12..20: number of vertices in group
 153          *  - bits 22..30: number of primitives in group
 154          */
 155         struct ac_arg gs_tg_info;
 156         /* API GS */
 157         struct ac_arg gs2vs_offset;
 158         struct ac_arg gs_wave_id; /* GFX6 */
 159         struct ac_arg gs_vtx_offset[6]; /* in dwords (GFX6) */
 160         struct ac_arg gs_vtx01_offset; /* in dwords (GFX9) */
 161         struct ac_arg gs_vtx23_offset; /* in dwords (GFX9) */
 162         struct ac_arg gs_vtx45_offset; /* in dwords (GFX9) */
 163         /* PS */
 164         struct ac_arg pos_fixed_pt;
 165         /* CS */
 166         struct ac_arg block_size;
 167         struct ac_arg cs_user_data;
 168
 169         struct ac_llvm_compiler *compiler;
 170
 171         /* Preloaded descriptors. */
 172         LLVMValueRef esgs_ring;
 173         LLVMValueRef gsvs_ring[4];
 174         LLVMValueRef tess_offchip_ring;
 175
 176         LLVMValueRef invoc0_tess_factors[6]; /* outer[4], inner[2] */
 177         LLVMValueRef gs_next_vertex[4];
 178         LLVMValueRef gs_curprim_verts[4];
 179         LLVMValueRef gs_generated_prims[4];
 180         LLVMValueRef gs_ngg_emit;
 181         LLVMValueRef gs_ngg_scratch;
 182         LLVMValueRef postponed_kill;
 183         LLVMValueRef return_value;
 184
 185         LLVMTypeRef voidt;
 186         LLVMTypeRef i1;
 187         LLVMTypeRef i8;
 188         LLVMTypeRef i32;
 189         LLVMTypeRef i64;
 190         LLVMTypeRef i128;
 191         LLVMTypeRef f32;
 192         LLVMTypeRef v2i32;
 193         LLVMTypeRef v4i32;
 194         LLVMTypeRef v4f32;
 195         LLVMTypeRef v8i32;
 196
 197         LLVMValueRef i32_0;
 198         LLVMValueRef i32_1;
 199         LLVMValueRef i1false;
 200         LLVMValueRef i1true;
 201 };
 202
 203 static inline struct si_shader_context *
 204 si_shader_context_from_abi(struct ac_shader_abi *abi)
 205 {
 206         struct si_shader_context *ctx = NULL;
 207         return container_of(abi, ctx, abi);
 208 }
 209
 210 unsigned si_llvm_compile(LLVMModuleRef M, struct si_shader_binary *binary,
 211                          struct ac_llvm_compiler *compiler,
 212                          struct pipe_debug_callback *debug,
 213                          bool less_optimized, unsigned wave_size);
 214
 215 LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx,
 216                                  LLVMValueRef index,
 217                                  unsigned num);
 218
 219 void si_llvm_context_init(struct si_shader_context *ctx,
 220                           struct si_screen *sscreen,
 221                           struct ac_llvm_compiler *compiler,
 222                           unsigned wave_size);
 223 void si_llvm_context_set_ir(struct si_shader_context *ctx,
 224                             struct si_shader *shader);
 225
 226 void si_llvm_create_func(struct si_shader_context *ctx, const char *name,
 227                          LLVMTypeRef *return_types, unsigned num_return_elems,
 228                          unsigned max_workgroup_size);
 229
 230 void si_llvm_dispose(struct si_shader_context *ctx);
 231
 232 void si_llvm_optimize_module(struct si_shader_context *ctx);
 233
 234 LLVMValueRef si_nir_load_input_tes(struct ac_shader_abi *abi,
 235                                    LLVMTypeRef type,
 236                                    LLVMValueRef vertex_index,
 237                                    LLVMValueRef param_index,
 238                                    unsigned const_index,
 239                                    unsigned location,
 240                                    unsigned driver_location,
 241                                    unsigned component,
 242                                    unsigned num_components,
 243                                    bool is_patch,
 244                                    bool is_compact,
 245                                    bool load_input);
 246 bool si_is_merged_shader(struct si_shader_context *ctx);
 247 LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
 248 LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx,
 249                                   LLVMValueRef list, LLVMValueRef index,
 250                                   enum ac_descriptor_type type);
 251 LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 252                                 LLVMValueRef list, LLVMValueRef index,
 253                                 enum ac_descriptor_type desc_type,
 254                                 bool uses_store, bool bindless);
 255 LLVMValueRef si_buffer_load_const(struct si_shader_context *ctx,
 256                                   LLVMValueRef resource, LLVMValueRef offset);
 257 void si_llvm_build_ret(struct si_shader_context *ctx, LLVMValueRef ret);
 258 LLVMValueRef si_prolog_get_rw_buffers(struct si_shader_context *ctx);
 259 LLVMValueRef si_build_gather_64bit(struct si_shader_context *ctx,
 260                                    LLVMTypeRef type, LLVMValueRef val1,
 261                                    LLVMValueRef val2);
 262 void si_llvm_emit_barrier(struct si_shader_context *ctx);
 263 void si_declare_compute_memory(struct si_shader_context *ctx);
 264 LLVMValueRef si_get_primitive_id(struct si_shader_context *ctx,
 265                                  unsigned swizzle);
 266 void si_llvm_export_vs(struct si_shader_context *ctx,
 267                        struct si_shader_output_values *outputs,
 268                        unsigned noutput);
 269 void si_emit_streamout_output(struct si_shader_context *ctx,
 270                               LLVMValueRef const *so_buffers,
 271                               LLVMValueRef const *so_write_offsets,
 272                               struct pipe_stream_output *stream_out,
 273                               struct si_shader_output_values *shader_out);
 274 void si_add_arg_checked(struct ac_shader_args *args,
 275                         enum ac_arg_regfile file,
 276                         unsigned registers, enum ac_arg_type type,
 277                         struct ac_arg *arg,
 278                         unsigned idx);
 279
 280 void si_llvm_load_input_vs(
 281         struct si_shader_context *ctx,
 282         unsigned input_index,
 283         LLVMValueRef out[4]);
 284
 285 bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir);
 286
 287 LLVMValueRef si_unpack_param(struct si_shader_context *ctx,
 288                              struct ac_arg param, unsigned rshift,
 289                              unsigned bitwidth);
 290 LLVMValueRef si_is_es_thread(struct si_shader_context *ctx);
 291 LLVMValueRef si_is_gs_thread(struct si_shader_context *ctx);
 292 void si_build_wrapper_function(struct si_shader_context *ctx, LLVMValueRef *parts,
 293                                unsigned num_parts, unsigned main_part,
 294                                unsigned next_shader_first_part);
 295 bool si_need_ps_prolog(const union si_shader_part_key *key);
 296 void si_get_ps_prolog_key(struct si_shader *shader,
 297                           union si_shader_part_key *key,
 298                           bool separate_prolog);
 299 void si_get_ps_epilog_key(struct si_shader *shader,
 300                           union si_shader_part_key *key);
 301 LLVMValueRef si_insert_input_ret(struct si_shader_context *ctx, LLVMValueRef ret,
 302                                  struct ac_arg param, unsigned return_index);
 303 LLVMValueRef si_insert_input_ret_float(struct si_shader_context *ctx, LLVMValueRef ret,
 304                                        struct ac_arg param, unsigned return_index);
 305 LLVMValueRef si_insert_input_ptr(struct si_shader_context *ctx, LLVMValueRef ret,
 306                                  struct ac_arg param, unsigned return_index);
 307
 308 void gfx10_emit_ngg_epilogue(struct ac_shader_abi *abi,
 309                              unsigned max_outputs,
 310                              LLVMValueRef *addrs);
 311 void gfx10_ngg_gs_emit_vertex(struct si_shader_context *ctx,
 312                               unsigned stream,
 313                               LLVMValueRef *addrs);
 314 void gfx10_ngg_gs_emit_prologue(struct si_shader_context *ctx);
 315 void gfx10_ngg_gs_emit_epilogue(struct si_shader_context *ctx);
 316 void gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
 317
 318 /* si_shader_llvm_tess.c */
 319 void si_llvm_preload_tes_rings(struct si_shader_context *ctx);
 320 void si_llvm_emit_ls_epilogue(struct ac_shader_abi *abi, unsigned max_outputs,
 321                               LLVMValueRef *addrs);
 322 void si_llvm_build_tcs_epilog(struct si_shader_context *ctx,
 323                               union si_shader_part_key *key);
 324 void si_llvm_init_tcs_callbacks(struct si_shader_context *ctx);
 325 void si_llvm_init_tes_callbacks(struct si_shader_context *ctx);
 326
 327 /* si_shader_llvm_ps.c */
 328 void si_llvm_build_ps_prolog(struct si_shader_context *ctx,
 329                              union si_shader_part_key *key);
 330 void si_llvm_build_ps_epilog(struct si_shader_context *ctx,
 331                              union si_shader_part_key *key);
 332 void si_llvm_build_monolithic_ps(struct si_shader_context *ctx,
 333                                  struct si_shader *shader);
 334 void si_llvm_init_ps_callbacks(struct si_shader_context *ctx);
 335
 336 #endif