src/amd/vulkan/radv_shader_info.c

   1 /*
   2  * Copyright © 2017 Red Hat
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23 #include "radv_private.h"
  24 #include "radv_shader.h"
  25 #include "nir/nir.h"
  26 #include "nir/nir_deref.h"
  27 #include "nir/nir_xfb_info.h"
  28
  29 static void mark_sampler_desc(const nir_variable *var,
  30                               struct radv_shader_info *info)
  31 {
  32         info->desc_set_used_mask |= (1 << var->data.descriptor_set);
  33 }
  34
  35 static void mark_ls_output(struct radv_shader_info *info,
  36                            uint32_t param, int num_slots)
  37 {
  38         uint64_t mask = (1ull << num_slots) - 1ull;
  39         info->vs.ls_outputs_written |= (mask << param);
  40 }
  41
  42 static void mark_tess_output(struct radv_shader_info *info,
  43                              bool is_patch, uint32_t param, int num_slots)
  44 {
  45         uint64_t mask = (1ull << num_slots) - 1ull;
  46         if (is_patch)
  47                 info->tcs.patch_outputs_written |= (mask << param);
  48         else
  49                 info->tcs.outputs_written |= (mask << param);
  50 }
  51
  52 static void
  53 get_deref_offset(nir_deref_instr *instr,
  54                  unsigned *const_out)
  55 {
  56         nir_variable *var = nir_deref_instr_get_variable(instr);
  57         nir_deref_path path;
  58         unsigned idx_lvl = 1;
  59
  60         if (var->data.compact) {
  61                 assert(instr->deref_type == nir_deref_type_array);
  62                 *const_out = nir_src_as_uint(instr->arr.index);
  63                 return;
  64         }
  65
  66         nir_deref_path_init(&path, instr, NULL);
  67
  68         uint32_t const_offset = 0;
  69
  70         for (; path.path[idx_lvl]; ++idx_lvl) {
  71                 const struct glsl_type *parent_type = path.path[idx_lvl - 1]->type;
  72                 if (path.path[idx_lvl]->deref_type == nir_deref_type_struct) {
  73                         unsigned index = path.path[idx_lvl]->strct.index;
  74
  75                         for (unsigned i = 0; i < index; i++) {
  76                                 const struct glsl_type *ft = glsl_get_struct_field(parent_type, i);
  77                                 const_offset += glsl_count_attribute_slots(ft, false);
  78                         }
  79                 } else if(path.path[idx_lvl]->deref_type == nir_deref_type_array) {
  80                         unsigned size = glsl_count_attribute_slots(path.path[idx_lvl]->type, false);
  81                         if (nir_src_is_const(path.path[idx_lvl]->arr.index))
  82                                 const_offset += nir_src_as_uint(path.path[idx_lvl]->arr.index) * size;
  83                 } else
  84                         unreachable("Uhandled deref type in get_deref_instr_offset");
  85         }
  86
  87         *const_out = const_offset;
  88
  89         nir_deref_path_finish(&path);
  90 }
  91
  92 static void
  93 gather_intrinsic_load_deref_info(const nir_shader *nir,
  94                                const nir_intrinsic_instr *instr,
  95                                struct radv_shader_info *info)
  96 {
  97         switch (nir->info.stage) {
  98         case MESA_SHADER_VERTEX: {
  99                 nir_variable *var = nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[0].ssa->parent_instr));
 100
 101                 if (var && var->data.mode == nir_var_shader_in) {
 102                         unsigned idx = var->data.location;
 103                         uint8_t mask = nir_ssa_def_components_read(&instr->dest.ssa);
 104
 105                         info->vs.input_usage_mask[idx] |=
 106                                 mask << var->data.location_frac;
 107                 }
 108                 break;
 109         }
 110         default:
 111                 break;
 112         }
 113 }
 114
 115 static uint32_t
 116 widen_writemask(uint32_t wrmask)
 117 {
 118         uint32_t new_wrmask = 0;
 119         for(unsigned i = 0; i < 4; i++)
 120                 new_wrmask |= (wrmask & (1 << i) ? 0x3 : 0x0) << (i * 2);
 121         return new_wrmask;
 122 }
 123
 124 static void
 125 set_output_usage_mask(const nir_shader *nir, const nir_intrinsic_instr *instr,
 126                       uint8_t *output_usage_mask)
 127 {
 128         nir_deref_instr *deref_instr =
 129                 nir_instr_as_deref(instr->src[0].ssa->parent_instr);
 130         nir_variable *var = nir_deref_instr_get_variable(deref_instr);
 131         unsigned attrib_count = glsl_count_attribute_slots(deref_instr->type, false);
 132         unsigned idx = var->data.location;
 133         unsigned comp = var->data.location_frac;
 134         unsigned const_offset = 0;
 135
 136         get_deref_offset(deref_instr, &const_offset);
 137
 138         if (var->data.compact) {
 139                 assert(!glsl_type_is_64bit(deref_instr->type));
 140                 const_offset += comp;
 141                 output_usage_mask[idx + const_offset / 4] |= 1 << (const_offset % 4);
 142                 return;
 143         }
 144
 145         uint32_t wrmask = nir_intrinsic_write_mask(instr);
 146         if (glsl_type_is_64bit(deref_instr->type))
 147                 wrmask = widen_writemask(wrmask);
 148
 149         for (unsigned i = 0; i < attrib_count; i++)
 150                 output_usage_mask[idx + i + const_offset] |=
 151                         ((wrmask >> (i * 4)) & 0xf) << comp;
 152 }
 153
 154 static void
 155 set_writes_memory(const nir_shader *nir, struct radv_shader_info *info)
 156 {
 157         if (nir->info.stage == MESA_SHADER_FRAGMENT)
 158                 info->ps.writes_memory = true;
 159         else if (nir->info.stage == MESA_SHADER_GEOMETRY)
 160                 info->gs.writes_memory = true;
 161 }
 162
 163 static void
 164 gather_intrinsic_store_deref_info(const nir_shader *nir,
 165                                 const nir_intrinsic_instr *instr,
 166                                 struct radv_shader_info *info)
 167 {
 168         nir_variable *var = nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[0].ssa->parent_instr));
 169
 170         if (var && var->data.mode == nir_var_shader_out) {
 171                 unsigned idx = var->data.location;
 172
 173                 switch (nir->info.stage) {
 174                 case MESA_SHADER_VERTEX:
 175                         set_output_usage_mask(nir, instr,
 176                                               info->vs.output_usage_mask);
 177                         break;
 178                 case MESA_SHADER_GEOMETRY:
 179                         set_output_usage_mask(nir, instr,
 180                                               info->gs.output_usage_mask);
 181                         break;
 182                 case MESA_SHADER_TESS_EVAL:
 183                         set_output_usage_mask(nir, instr,
 184                                               info->tes.output_usage_mask);
 185                         break;
 186                 case MESA_SHADER_TESS_CTRL: {
 187                         unsigned param = shader_io_get_unique_index(idx);
 188                         const struct glsl_type *type = var->type;
 189
 190                         if (!var->data.patch)
 191                                 type = glsl_get_array_element(var->type);
 192
 193                         unsigned slots =
 194                                 var->data.compact ? DIV_ROUND_UP(var->data.location_frac + glsl_get_length(type), 4)
 195                                                   : glsl_count_attribute_slots(type, false);
 196
 197                         mark_tess_output(info, var->data.patch, param, slots);
 198                         break;
 199                 }
 200                 default:
 201                         break;
 202                 }
 203         }
 204 }
 205
 206 static void
 207 gather_push_constant_info(const nir_shader *nir,
 208                           const nir_intrinsic_instr *instr,
 209                           struct radv_shader_info *info)
 210 {
 211         int base = nir_intrinsic_base(instr);
 212
 213         if (!nir_src_is_const(instr->src[0])) {
 214                 info->has_indirect_push_constants = true;
 215         } else {
 216                 uint32_t min = base + nir_src_as_uint(instr->src[0]);
 217                 uint32_t max = min + instr->num_components * 4;
 218
 219                 info->max_push_constant_used =
 220                         MAX2(max, info->max_push_constant_used);
 221                 info->min_push_constant_used =
 222                         MIN2(min, info->min_push_constant_used);
 223         }
 224
 225         if (instr->dest.ssa.bit_size != 32)
 226                 info->has_only_32bit_push_constants = false;
 227
 228         info->loads_push_constants = true;
 229 }
 230
 231 static void
 232 gather_intrinsic_info(const nir_shader *nir, const nir_intrinsic_instr *instr,
 233                       struct radv_shader_info *info)
 234 {
 235         switch (instr->intrinsic) {
 236         case nir_intrinsic_load_barycentric_at_sample:
 237                 info->ps.needs_sample_positions = true;
 238                 break;
 239         case nir_intrinsic_load_draw_id:
 240                 info->vs.needs_draw_id = true;
 241                 break;
 242         case nir_intrinsic_load_instance_id:
 243                 info->vs.needs_instance_id = true;
 244                 break;
 245         case nir_intrinsic_load_num_work_groups:
 246                 info->cs.uses_grid_size = true;
 247                 break;
 248         case nir_intrinsic_load_local_invocation_id:
 249         case nir_intrinsic_load_work_group_id: {
 250                 unsigned mask = nir_ssa_def_components_read(&instr->dest.ssa);
 251                 while (mask) {
 252                         unsigned i = u_bit_scan(&mask);
 253
 254                         if (instr->intrinsic == nir_intrinsic_load_work_group_id)
 255                                 info->cs.uses_block_id[i] = true;
 256                         else
 257                                 info->cs.uses_thread_id[i] = true;
 258                 }
 259                 break;
 260         }
 261         case nir_intrinsic_load_local_invocation_index:
 262         case nir_intrinsic_load_subgroup_id:
 263         case nir_intrinsic_load_num_subgroups:
 264                 info->cs.uses_local_invocation_idx = true;
 265                 break;
 266         case nir_intrinsic_load_sample_id:
 267                 info->ps.force_persample = true;
 268                 break;
 269         case nir_intrinsic_load_sample_pos:
 270                 info->ps.force_persample = true;
 271                 break;
 272         case nir_intrinsic_load_view_index:
 273                 info->needs_multiview_view_index = true;
 274                 if (nir->info.stage == MESA_SHADER_FRAGMENT)
 275                         info->ps.layer_input = true;
 276                 break;
 277         case nir_intrinsic_load_layer_id:
 278                 if (nir->info.stage == MESA_SHADER_FRAGMENT)
 279                         info->ps.layer_input = true;
 280                 break;
 281         case nir_intrinsic_load_invocation_id:
 282                 info->uses_invocation_id = true;
 283                 break;
 284         case nir_intrinsic_load_primitive_id:
 285                 info->uses_prim_id = true;
 286                 break;
 287         case nir_intrinsic_load_push_constant:
 288                 gather_push_constant_info(nir, instr, info);
 289                 break;
 290         case nir_intrinsic_vulkan_resource_index:
 291                 info->desc_set_used_mask |= (1 << nir_intrinsic_desc_set(instr));
 292                 break;
 293         case nir_intrinsic_image_deref_load:
 294         case nir_intrinsic_image_deref_store:
 295         case nir_intrinsic_image_deref_atomic_add:
 296         case nir_intrinsic_image_deref_atomic_imin:
 297         case nir_intrinsic_image_deref_atomic_umin:
 298         case nir_intrinsic_image_deref_atomic_imax:
 299         case nir_intrinsic_image_deref_atomic_umax:
 300         case nir_intrinsic_image_deref_atomic_and:
 301         case nir_intrinsic_image_deref_atomic_or:
 302         case nir_intrinsic_image_deref_atomic_xor:
 303         case nir_intrinsic_image_deref_atomic_exchange:
 304         case nir_intrinsic_image_deref_atomic_comp_swap:
 305         case nir_intrinsic_image_deref_size: {
 306                 nir_variable *var = nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[0].ssa->parent_instr));
 307                 mark_sampler_desc(var, info);
 308
 309                 if (instr->intrinsic == nir_intrinsic_image_deref_store ||
 310                     instr->intrinsic == nir_intrinsic_image_deref_atomic_add ||
 311                     instr->intrinsic == nir_intrinsic_image_deref_atomic_imin ||
 312                     instr->intrinsic == nir_intrinsic_image_deref_atomic_umin ||
 313                     instr->intrinsic == nir_intrinsic_image_deref_atomic_imax ||
 314                     instr->intrinsic == nir_intrinsic_image_deref_atomic_umax ||
 315                     instr->intrinsic == nir_intrinsic_image_deref_atomic_and ||
 316                     instr->intrinsic == nir_intrinsic_image_deref_atomic_or ||
 317                     instr->intrinsic == nir_intrinsic_image_deref_atomic_xor ||
 318                     instr->intrinsic == nir_intrinsic_image_deref_atomic_exchange ||
 319                     instr->intrinsic == nir_intrinsic_image_deref_atomic_comp_swap) {
 320                         set_writes_memory(nir, info);
 321                 }
 322                 break;
 323         }
 324         case nir_intrinsic_store_ssbo:
 325         case nir_intrinsic_ssbo_atomic_add:
 326         case nir_intrinsic_ssbo_atomic_imin:
 327         case nir_intrinsic_ssbo_atomic_umin:
 328         case nir_intrinsic_ssbo_atomic_imax:
 329         case nir_intrinsic_ssbo_atomic_umax:
 330         case nir_intrinsic_ssbo_atomic_and:
 331         case nir_intrinsic_ssbo_atomic_or:
 332         case nir_intrinsic_ssbo_atomic_xor:
 333         case nir_intrinsic_ssbo_atomic_exchange:
 334         case nir_intrinsic_ssbo_atomic_comp_swap:
 335                 set_writes_memory(nir, info);
 336                 break;
 337         case nir_intrinsic_load_deref:
 338                 gather_intrinsic_load_deref_info(nir, instr, info);
 339                 break;
 340         case nir_intrinsic_store_deref:
 341                 gather_intrinsic_store_deref_info(nir, instr, info);
 342                 /* fallthrough */
 343         case nir_intrinsic_deref_atomic_add:
 344         case nir_intrinsic_deref_atomic_imin:
 345         case nir_intrinsic_deref_atomic_umin:
 346         case nir_intrinsic_deref_atomic_imax:
 347         case nir_intrinsic_deref_atomic_umax:
 348         case nir_intrinsic_deref_atomic_and:
 349         case nir_intrinsic_deref_atomic_or:
 350         case nir_intrinsic_deref_atomic_xor:
 351         case nir_intrinsic_deref_atomic_exchange:
 352         case nir_intrinsic_deref_atomic_comp_swap: {
 353                 if (nir_src_as_deref(instr->src[0])->mode & (nir_var_mem_global | nir_var_mem_ssbo))
 354                         set_writes_memory(nir, info);
 355                 break;
 356         }
 357         default:
 358                 break;
 359         }
 360 }
 361
 362 static void
 363 gather_tex_info(const nir_shader *nir, const nir_tex_instr *instr,
 364                 struct radv_shader_info *info)
 365 {
 366         for (unsigned i = 0; i < instr->num_srcs; i++) {
 367                 switch (instr->src[i].src_type) {
 368                 case nir_tex_src_texture_deref:
 369                         mark_sampler_desc(nir_deref_instr_get_variable(nir_src_as_deref(instr->src[i].src)), info);
 370                         break;
 371                 case nir_tex_src_sampler_deref:
 372                         mark_sampler_desc(nir_deref_instr_get_variable(nir_src_as_deref(instr->src[i].src)), info);
 373                         break;
 374                 default:
 375                         break;
 376                 }
 377         }
 378 }
 379
 380 static void
 381 gather_info_block(const nir_shader *nir, const nir_block *block,
 382                   struct radv_shader_info *info)
 383 {
 384         nir_foreach_instr(instr, block) {
 385                 switch (instr->type) {
 386                 case nir_instr_type_intrinsic:
 387                         gather_intrinsic_info(nir, nir_instr_as_intrinsic(instr), info);
 388                         break;
 389                 case nir_instr_type_tex:
 390                         gather_tex_info(nir, nir_instr_as_tex(instr), info);
 391                         break;
 392                 default:
 393                         break;
 394                 }
 395         }
 396 }
 397
 398 static void
 399 gather_info_input_decl_vs(const nir_shader *nir, const nir_variable *var,
 400                           struct radv_shader_info *info,
 401                           const struct radv_shader_variant_key *key)
 402 {
 403         unsigned attrib_count = glsl_count_attribute_slots(var->type, true);
 404         int idx = var->data.location;
 405
 406         if (idx >= VERT_ATTRIB_GENERIC0 && idx <= VERT_ATTRIB_GENERIC15)
 407                 info->vs.has_vertex_buffers = true;
 408
 409         for (unsigned i = 0; i < attrib_count; ++i) {
 410                 unsigned attrib_index = var->data.location + i - VERT_ATTRIB_GENERIC0;
 411
 412                 if (key->vs.instance_rate_inputs & (1u << attrib_index))
 413                         info->vs.needs_instance_id = true;
 414         }
 415 }
 416
 417 static void
 418 mark_16bit_ps_input(struct radv_shader_info *info, const struct glsl_type *type,
 419                     int location)
 420 {
 421         if (glsl_type_is_scalar(type) || glsl_type_is_vector(type) || glsl_type_is_matrix(type)) {
 422                 unsigned attrib_count = glsl_count_attribute_slots(type, false);
 423                 if (glsl_type_is_16bit(type)) {
 424                         info->ps.float16_shaded_mask |= ((1ull << attrib_count) - 1) << location;
 425                 }
 426         } else if (glsl_type_is_array(type)) {
 427                 unsigned stride = glsl_count_attribute_slots(glsl_get_array_element(type), false);
 428                 for (unsigned i = 0; i < glsl_get_length(type); ++i) {
 429                         mark_16bit_ps_input(info, glsl_get_array_element(type), location + i * stride);
 430                 }
 431         } else {
 432                 assert(glsl_type_is_struct_or_ifc(type));
 433                 for (unsigned i = 0; i < glsl_get_length(type); i++) {
 434                         mark_16bit_ps_input(info, glsl_get_struct_field(type, i), location);
 435                         location += glsl_count_attribute_slots(glsl_get_struct_field(type, i), false);
 436                 }
 437         }
 438 }
 439 static void
 440 gather_info_input_decl_ps(const nir_shader *nir, const nir_variable *var,
 441                           struct radv_shader_info *info)
 442 {
 443         unsigned attrib_count = glsl_count_attribute_slots(var->type, false);
 444         const struct glsl_type *type = glsl_without_array(var->type);
 445         int idx = var->data.location;
 446
 447         switch (idx) {
 448         case VARYING_SLOT_PNTC:
 449                 info->ps.has_pcoord = true;
 450                 break;
 451         case VARYING_SLOT_PRIMITIVE_ID:
 452                 info->ps.prim_id_input = true;
 453                 break;
 454         case VARYING_SLOT_LAYER:
 455                 info->ps.layer_input = true;
 456                 break;
 457         case VARYING_SLOT_CLIP_DIST0:
 458         case VARYING_SLOT_CLIP_DIST1:
 459                 info->ps.num_input_clips_culls += attrib_count;
 460                 break;
 461         case VARYING_SLOT_VIEWPORT:
 462                 info->ps.viewport_index_input = true;
 463                 break;
 464         default:
 465                 break;
 466         }
 467
 468         if (glsl_get_base_type(type) == GLSL_TYPE_FLOAT) {
 469                 if (var->data.sample)
 470                         info->ps.force_persample = true;
 471         }
 472
 473         if (var->data.compact) {
 474                 unsigned component_count = var->data.location_frac +
 475                                            glsl_get_length(var->type);
 476                 attrib_count = (component_count + 3) / 4;
 477         } else {
 478                 mark_16bit_ps_input(info, var->type, var->data.driver_location);
 479         }
 480
 481         uint64_t mask = ((1ull << attrib_count) - 1);
 482
 483         if (var->data.interpolation == INTERP_MODE_FLAT)
 484                 info->ps.flat_shaded_mask |= mask << var->data.driver_location;
 485         if (var->data.interpolation == INTERP_MODE_EXPLICIT)
 486                 info->ps.explicit_shaded_mask |= mask << var->data.driver_location;
 487
 488         if (var->data.location >= VARYING_SLOT_VAR0)
 489                 info->ps.input_mask |= mask << (var->data.location - VARYING_SLOT_VAR0);
 490 }
 491
 492 static void
 493 gather_info_input_decl(const nir_shader *nir, const nir_variable *var,
 494                        struct radv_shader_info *info,
 495                        const struct radv_shader_variant_key *key)
 496 {
 497         switch (nir->info.stage) {
 498         case MESA_SHADER_VERTEX:
 499                 gather_info_input_decl_vs(nir, var, info, key);
 500                 break;
 501         case MESA_SHADER_FRAGMENT:
 502                 gather_info_input_decl_ps(nir, var, info);
 503                 break;
 504         default:
 505                 break;
 506         }
 507 }
 508
 509 static void
 510 gather_info_output_decl_ls(const nir_shader *nir, const nir_variable *var,
 511                            struct radv_shader_info *info)
 512 {
 513         int idx = var->data.location;
 514         unsigned param = shader_io_get_unique_index(idx);
 515         int num_slots = glsl_count_attribute_slots(var->type, false);
 516         if (var->data.compact)
 517                 num_slots = DIV_ROUND_UP(var->data.location_frac + glsl_get_length(var->type), 4);
 518         mark_ls_output(info, param, num_slots);
 519 }
 520
 521 static void
 522 gather_info_output_decl_ps(const nir_shader *nir, const nir_variable *var,
 523                            struct radv_shader_info *info)
 524 {
 525         int idx = var->data.location;
 526
 527         switch (idx) {
 528         case FRAG_RESULT_DEPTH:
 529                 info->ps.writes_z = true;
 530                 break;
 531         case FRAG_RESULT_STENCIL:
 532                 info->ps.writes_stencil = true;
 533                 break;
 534         case FRAG_RESULT_SAMPLE_MASK:
 535                 info->ps.writes_sample_mask = true;
 536                 break;
 537         default:
 538                 break;
 539         }
 540 }
 541
 542 static void
 543 gather_info_output_decl_gs(const nir_shader *nir, const nir_variable *var,
 544                            struct radv_shader_info *info)
 545 {
 546         unsigned num_components = glsl_get_component_slots(var->type);
 547         unsigned stream = var->data.stream;
 548         unsigned idx = var->data.location;
 549
 550         assert(stream < 4);
 551
 552         info->gs.max_stream = MAX2(info->gs.max_stream, stream);
 553         info->gs.num_stream_output_components[stream] += num_components;
 554         info->gs.output_streams[idx] = stream;
 555 }
 556
 557 static void
 558 gather_info_output_decl(const nir_shader *nir, const nir_variable *var,
 559                         struct radv_shader_info *info,
 560                         const struct radv_shader_variant_key *key)
 561 {
 562         struct radv_vs_output_info *vs_info = NULL;
 563
 564         switch (nir->info.stage) {
 565         case MESA_SHADER_FRAGMENT:
 566                 gather_info_output_decl_ps(nir, var, info);
 567                 break;
 568         case MESA_SHADER_VERTEX:
 569                 if (!key->vs_common_out.as_ls &&
 570                     !key->vs_common_out.as_es)
 571                         vs_info = &info->vs.outinfo;
 572
 573                 if (key->vs_common_out.as_ls)
 574                         gather_info_output_decl_ls(nir, var, info);
 575                 else if (key->vs_common_out.as_ngg)
 576                         gather_info_output_decl_gs(nir, var, info);
 577                 break;
 578         case MESA_SHADER_GEOMETRY:
 579                 vs_info = &info->vs.outinfo;
 580                 gather_info_output_decl_gs(nir, var, info);
 581                 break;
 582         case MESA_SHADER_TESS_EVAL:
 583                 if (!key->vs_common_out.as_es)
 584                         vs_info = &info->tes.outinfo;
 585                 break;
 586         default:
 587                 break;
 588         }
 589
 590         if (vs_info) {
 591                 switch (var->data.location) {
 592                 case VARYING_SLOT_CLIP_DIST0:
 593                         vs_info->clip_dist_mask =
 594                                 (1 << nir->info.clip_distance_array_size) - 1;
 595                         vs_info->cull_dist_mask =
 596                                 (1 << nir->info.cull_distance_array_size) - 1;
 597                         vs_info->cull_dist_mask <<= nir->info.clip_distance_array_size;
 598                         break;
 599                 case VARYING_SLOT_PSIZ:
 600                         vs_info->writes_pointsize = true;
 601                         break;
 602                 case VARYING_SLOT_VIEWPORT:
 603                         vs_info->writes_viewport_index = true;
 604                         break;
 605                 case VARYING_SLOT_LAYER:
 606                         vs_info->writes_layer = true;
 607                         break;
 608                 default:
 609                         break;
 610                 }
 611         }
 612 }
 613
 614 static void
 615 gather_xfb_info(const nir_shader *nir, struct radv_shader_info *info)
 616 {
 617         nir_xfb_info *xfb = nir_gather_xfb_info(nir, NULL);
 618         struct radv_streamout_info *so = &info->so;
 619
 620         if (!xfb)
 621                 return;
 622
 623         assert(xfb->output_count < MAX_SO_OUTPUTS);
 624         so->num_outputs = xfb->output_count;
 625
 626         for (unsigned i = 0; i < xfb->output_count; i++) {
 627                 struct radv_stream_output *output = &so->outputs[i];
 628
 629                 output->buffer = xfb->outputs[i].buffer;
 630                 output->stream = xfb->buffer_to_stream[xfb->outputs[i].buffer];
 631                 output->offset = xfb->outputs[i].offset;
 632                 output->location = xfb->outputs[i].location;
 633                 output->component_mask = xfb->outputs[i].component_mask;
 634
 635                 so->enabled_stream_buffers_mask |=
 636                         (1 << output->buffer) << (output->stream * 4);
 637
 638         }
 639
 640         for (unsigned i = 0; i < NIR_MAX_XFB_BUFFERS; i++) {
 641                 so->strides[i] = xfb->buffers[i].stride / 4;
 642         }
 643
 644         ralloc_free(xfb);
 645 }
 646
 647 void
 648 radv_nir_shader_info_init(struct radv_shader_info *info)
 649 {
 650         /* Assume that shaders only have 32-bit push constants by default. */
 651         info->min_push_constant_used = UINT8_MAX;
 652         info->has_only_32bit_push_constants = true;
 653 }
 654
 655 void
 656 radv_nir_shader_info_pass(const struct nir_shader *nir,
 657                           const struct radv_pipeline_layout *layout,
 658                           const struct radv_shader_variant_key *key,
 659                           struct radv_shader_info *info)
 660 {
 661         struct nir_function *func =
 662                 (struct nir_function *)exec_list_get_head_const(&nir->functions);
 663
 664         if (layout && layout->dynamic_offset_count &&
 665             (layout->dynamic_shader_stages & mesa_to_vk_shader_stage(nir->info.stage))) {
 666                 info->loads_push_constants = true;
 667                 info->loads_dynamic_offsets = true;
 668         }
 669
 670         nir_foreach_variable(variable, &nir->inputs)
 671                 gather_info_input_decl(nir, variable, info, key);
 672
 673         nir_foreach_block(block, func->impl) {
 674                 gather_info_block(nir, block, info);
 675         }
 676
 677         nir_foreach_variable(variable, &nir->outputs)
 678                 gather_info_output_decl(nir, variable, info, key);
 679
 680         if (nir->info.stage == MESA_SHADER_VERTEX ||
 681             nir->info.stage == MESA_SHADER_TESS_EVAL ||
 682             nir->info.stage == MESA_SHADER_GEOMETRY)
 683                 gather_xfb_info(nir, info);
 684
 685         /* Make sure to export the LayerID if the fragment shader needs it. */
 686         if (key->vs_common_out.export_layer_id) {
 687                 switch (nir->info.stage) {
 688                 case MESA_SHADER_VERTEX:
 689                         info->vs.output_usage_mask[VARYING_SLOT_LAYER] |= 0x1;
 690                         break;
 691                 case MESA_SHADER_TESS_EVAL:
 692                         info->tes.output_usage_mask[VARYING_SLOT_LAYER] |= 0x1;
 693                         break;
 694                 case MESA_SHADER_GEOMETRY:
 695                         info->gs.output_usage_mask[VARYING_SLOT_LAYER] |= 0x1;
 696                         break;
 697                 default:
 698                         break;
 699                 }
 700         }
 701
 702         /* Make sure to export the LayerID if the subpass has multiviews. */
 703         if (key->has_multiview_view_index) {
 704                 switch (nir->info.stage) {
 705                 case MESA_SHADER_VERTEX:
 706                         info->vs.outinfo.writes_layer = true;
 707                         break;
 708                 case MESA_SHADER_TESS_EVAL:
 709                         info->tes.outinfo.writes_layer = true;
 710                         break;
 711                 case MESA_SHADER_GEOMETRY:
 712                         info->vs.outinfo.writes_layer = true;
 713                         break;
 714                 default:
 715                         break;
 716                 }
 717         }
 718
 719         /* Make sure to export the PrimitiveID if the fragment shader needs it. */
 720         if (key->vs_common_out.export_prim_id) {
 721                 switch (nir->info.stage) {
 722                 case MESA_SHADER_VERTEX:
 723                         info->vs.outinfo.export_prim_id = true;
 724                         break;
 725                 case MESA_SHADER_TESS_EVAL:
 726                         info->tes.outinfo.export_prim_id = true;
 727                         break;
 728                 case MESA_SHADER_GEOMETRY:
 729                         info->vs.outinfo.export_prim_id = true;
 730                         break;
 731                 default:
 732                         break;
 733                 }
 734         }
 735
 736         /* Make sure to export the ViewportIndex if the fragment shader needs it. */
 737         if (key->vs_common_out.export_viewport_index) {
 738                 switch (nir->info.stage) {
 739                 case MESA_SHADER_VERTEX:
 740                         info->vs.output_usage_mask[VARYING_SLOT_VIEWPORT] |= 0x1;
 741                         break;
 742                 case MESA_SHADER_TESS_EVAL:
 743                         info->tes.output_usage_mask[VARYING_SLOT_VIEWPORT] |= 0x1;
 744                         break;
 745                 case MESA_SHADER_GEOMETRY:
 746                         info->gs.output_usage_mask[VARYING_SLOT_VIEWPORT] |= 0x1;
 747                         break;
 748                 default:
 749                         break;
 750                 }
 751         }
 752
 753         if (nir->info.stage == MESA_SHADER_FRAGMENT)
 754                 info->ps.num_interp = nir->num_inputs;
 755
 756         switch (nir->info.stage) {
 757         case MESA_SHADER_COMPUTE:
 758                 for (int i = 0; i < 3; ++i)
 759                         info->cs.block_size[i] = nir->info.cs.local_size[i];
 760                 break;
 761         case MESA_SHADER_FRAGMENT:
 762                 info->ps.can_discard = nir->info.fs.uses_discard;
 763                 info->ps.early_fragment_test = nir->info.fs.early_fragment_tests;
 764                 info->ps.post_depth_coverage = nir->info.fs.post_depth_coverage;
 765                 break;
 766         case MESA_SHADER_GEOMETRY:
 767                 info->gs.vertices_in = nir->info.gs.vertices_in;
 768                 info->gs.vertices_out = nir->info.gs.vertices_out;
 769                 info->gs.output_prim = nir->info.gs.output_primitive;
 770                 info->gs.invocations = nir->info.gs.invocations;
 771                 break;
 772         case MESA_SHADER_TESS_EVAL:
 773                 info->tes.primitive_mode = nir->info.tess.primitive_mode;
 774                 info->tes.spacing = nir->info.tess.spacing;
 775                 info->tes.ccw = nir->info.tess.ccw;
 776                 info->tes.point_mode = nir->info.tess.point_mode;
 777                 info->tes.as_es = key->vs_common_out.as_es;
 778                 info->tes.export_prim_id = key->vs_common_out.export_prim_id;
 779                 info->is_ngg = key->vs_common_out.as_ngg;
 780                 info->is_ngg_passthrough = key->vs_common_out.as_ngg_passthrough;
 781                 break;
 782         case MESA_SHADER_TESS_CTRL:
 783                 info->tcs.tcs_vertices_out = nir->info.tess.tcs_vertices_out;
 784                 break;
 785         case MESA_SHADER_VERTEX:
 786                 info->vs.as_es = key->vs_common_out.as_es;
 787                 info->vs.as_ls = key->vs_common_out.as_ls;
 788                 info->vs.export_prim_id = key->vs_common_out.export_prim_id;
 789                 info->is_ngg = key->vs_common_out.as_ngg;
 790                 info->is_ngg_passthrough = key->vs_common_out.as_ngg_passthrough;
 791                 break;
 792         default:
 793                 break;
 794         }
 795
 796         if (nir->info.stage == MESA_SHADER_GEOMETRY) {
 797                 unsigned add_clip = nir->info.clip_distance_array_size +
 798                                     nir->info.cull_distance_array_size > 4;
 799                 info->gs.gsvs_vertex_size =
 800                         (util_bitcount64(nir->info.outputs_written) + add_clip) * 16;
 801                 info->gs.max_gsvs_emit_size =
 802                         info->gs.gsvs_vertex_size * nir->info.gs.vertices_out;
 803         }
 804
 805         /* Compute the ESGS item size for VS or TES as ES. */
 806         if ((nir->info.stage == MESA_SHADER_VERTEX ||
 807              nir->info.stage == MESA_SHADER_TESS_EVAL) &&
 808             key->vs_common_out.as_es) {
 809                 struct radv_es_output_info *es_info =
 810                         nir->info.stage == MESA_SHADER_VERTEX ? &info->vs.es_info : &info->tes.es_info;
 811                 uint32_t max_output_written = 0;
 812
 813                 uint64_t output_mask = nir->info.outputs_written;
 814                 while (output_mask) {
 815                         const int i = u_bit_scan64(&output_mask);
 816                         unsigned param_index = shader_io_get_unique_index(i);
 817
 818                         max_output_written = MAX2(param_index, max_output_written);
 819                 }
 820
 821                 es_info->esgs_itemsize = (max_output_written + 1) * 16;
 822         }
 823
 824         info->float_controls_mode = nir->info.float_controls_execution_mode;
 825 }