src/panfrost/bifrost/bifrost_compile.c

   1 /*
   2  * Copyright (C) 2020 Collabora Ltd.
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors (Collabora):
  24  *      Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  25  */
  26
  27 #include "main/mtypes.h"
  28 #include "compiler/glsl/glsl_to_nir.h"
  29 #include "compiler/nir_types.h"
  30 #include "main/imports.h"
  31 #include "compiler/nir/nir_builder.h"
  32
  33 #include "disassemble.h"
  34 #include "bifrost_compile.h"
  35 #include "compiler.h"
  36 #include "bi_quirks.h"
  37 #include "bi_print.h"
  38
  39 static bi_block *emit_cf_list(bi_context *ctx, struct exec_list *list);
  40 static bi_instruction *bi_emit_branch(bi_context *ctx);
  41 static void bi_block_add_successor(bi_block *block, bi_block *successor);
  42 static void bi_schedule_barrier(bi_context *ctx);
  43
  44 static void
  45 emit_jump(bi_context *ctx, nir_jump_instr *instr)
  46 {
  47         bi_instruction *branch = bi_emit_branch(ctx);
  48
  49         switch (instr->type) {
  50         case nir_jump_break:
  51                 branch->branch.target = ctx->break_block;
  52                 break;
  53         case nir_jump_continue:
  54                 branch->branch.target = ctx->continue_block;
  55                 break;
  56         default:
  57                 unreachable("Unhandled jump type");
  58         }
  59
  60         bi_block_add_successor(ctx->current_block, branch->branch.target);
  61 }
  62
  63 /* Gets a bytemask for a complete vecN write */
  64 static unsigned
  65 bi_mask_for_channels_32(unsigned i)
  66 {
  67         return (1 << (4 * i)) - 1;
  68 }
  69
  70 static bi_instruction
  71 bi_load(enum bi_class T, nir_intrinsic_instr *instr)
  72 {
  73         bi_instruction load = {
  74                 .type = T,
  75                 .writemask = bi_mask_for_channels_32(instr->num_components),
  76                 .src = { BIR_INDEX_CONSTANT },
  77                 .constant = { .u64 = nir_intrinsic_base(instr) },
  78         };
  79
  80         const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];
  81
  82         if (info->has_dest)
  83                 load.dest = bir_dest_index(&instr->dest);
  84
  85         if (info->has_dest && info->index_map[NIR_INTRINSIC_TYPE] > 0)
  86                 load.dest_type = nir_intrinsic_type(instr);
  87
  88         nir_src *offset = nir_get_io_offset_src(instr);
  89
  90         if (nir_src_is_const(*offset))
  91                 load.constant.u64 += nir_src_as_uint(*offset);
  92         else
  93                 load.src[0] = bir_src_index(offset);
  94
  95         return load;
  96 }
  97
  98 static void
  99 bi_emit_ld_vary(bi_context *ctx, nir_intrinsic_instr *instr)
 100 {
 101         bi_instruction ins = bi_load(BI_LOAD_VAR, instr);
 102         ins.load_vary.interp_mode = BIFROST_INTERP_DEFAULT; /* TODO */
 103         ins.load_vary.reuse = false; /* TODO */
 104         ins.load_vary.flat = instr->intrinsic != nir_intrinsic_load_interpolated_input;
 105         ins.dest_type = nir_type_float | nir_dest_bit_size(instr->dest),
 106         bi_emit(ctx, ins);
 107 }
 108
 109 static void
 110 bi_emit_frag_out(bi_context *ctx, nir_intrinsic_instr *instr)
 111 {
 112         if (!ctx->emitted_atest) {
 113                 bi_instruction ins = {
 114                         .type = BI_ATEST
 115                 };
 116
 117                 bi_emit(ctx, ins);
 118                 bi_schedule_barrier(ctx);
 119                 ctx->emitted_atest = true;
 120         }
 121
 122         bi_instruction blend = {
 123                 .type = BI_BLEND,
 124                 .blend_location = nir_intrinsic_base(instr),
 125                 .src = {
 126                         bir_src_index(&instr->src[0])
 127                 },
 128                 .swizzle = {
 129                         { 0, 1, 2, 3 }
 130                 }
 131         };
 132
 133         bi_emit(ctx, blend);
 134         bi_schedule_barrier(ctx);
 135 }
 136
 137 static void
 138 bi_emit_st_vary(bi_context *ctx, nir_intrinsic_instr *instr)
 139 {
 140         bi_instruction address = bi_load(BI_LOAD_VAR_ADDRESS, instr);
 141         address.dest = bi_make_temp(ctx);
 142         address.dest_type = nir_type_uint64;
 143         address.writemask = (1 << 8) - 1;
 144
 145         bi_instruction st = {
 146                 .type = BI_STORE_VAR,
 147                 .src = {
 148                         address.dest,
 149                         bir_src_index(&instr->src[0])
 150                 },
 151                 .swizzle = {
 152                         { 0, 1, 2, 3 }
 153                 }
 154         };
 155
 156         bi_emit(ctx, address);
 157         bi_emit(ctx, st);
 158 }
 159
 160 static void
 161 bi_emit_ld_uniform(bi_context *ctx, nir_intrinsic_instr *instr)
 162 {
 163         bi_instruction ld = bi_load(BI_LOAD_UNIFORM, instr);
 164         ld.src[1] = BIR_INDEX_ZERO; /* TODO: UBO index */
 165         bi_emit(ctx, ld);
 166 }
 167
 168 static void
 169 emit_intrinsic(bi_context *ctx, nir_intrinsic_instr *instr)
 170 {
 171
 172         switch (instr->intrinsic) {
 173         case nir_intrinsic_load_barycentric_pixel:
 174                 /* stub */
 175                 break;
 176         case nir_intrinsic_load_interpolated_input:
 177         case nir_intrinsic_load_input:
 178                 if (ctx->stage == MESA_SHADER_FRAGMENT)
 179                         bi_emit_ld_vary(ctx, instr);
 180                 else if (ctx->stage == MESA_SHADER_VERTEX)
 181                         bi_emit(ctx, bi_load(BI_LOAD_ATTR, instr));
 182                 else {
 183                         unreachable("Unsupported shader stage");
 184                 }
 185                 break;
 186
 187         case nir_intrinsic_store_output:
 188                 if (ctx->stage == MESA_SHADER_FRAGMENT)
 189                         bi_emit_frag_out(ctx, instr);
 190                 else if (ctx->stage == MESA_SHADER_VERTEX)
 191                         bi_emit_st_vary(ctx, instr);
 192                 else
 193                         unreachable("Unsupported shader stage");
 194                 break;
 195
 196         case nir_intrinsic_load_uniform:
 197                 bi_emit_ld_uniform(ctx, instr);
 198                 break;
 199
 200         default:
 201                 /* todo */
 202                 break;
 203         }
 204 }
 205
 206 static void
 207 emit_load_const(bi_context *ctx, nir_load_const_instr *instr)
 208 {
 209         /* Make sure we've been lowered */
 210         assert(instr->def.num_components == 1);
 211
 212         bi_instruction move = {
 213                 .type = BI_MOV,
 214                 .dest = bir_ssa_index(&instr->def),
 215                 .dest_type = instr->def.bit_size | nir_type_uint,
 216                 .writemask = (1 << (instr->def.bit_size / 8)) - 1,
 217                 .src = {
 218                         BIR_INDEX_CONSTANT
 219                 },
 220                 .constant = {
 221                         .u64 = nir_const_value_as_uint(instr->value[0], instr->def.bit_size)
 222                 }
 223         };
 224
 225         bi_emit(ctx, move);
 226 }
 227
 228 static enum bi_class
 229 bi_class_for_nir_alu(nir_op op)
 230 {
 231         switch (op) {
 232         case nir_op_fadd: return BI_ADD;
 233         case nir_op_fmul: return BI_FMA;
 234         case nir_op_mov:  return BI_MOV;
 235         default: unreachable("Unknown ALU op");
 236         }
 237 }
 238
 239 static void
 240 emit_alu(bi_context *ctx, nir_alu_instr *instr)
 241 {
 242         /* Assume it's something we can handle normally */
 243         bi_instruction alu = {
 244                 .type = bi_class_for_nir_alu(instr->op),
 245                 .dest = bir_dest_index(&instr->dest.dest),
 246                 .dest_type = nir_op_infos[instr->op].output_type
 247                         | nir_dest_bit_size(instr->dest.dest),
 248         };
 249
 250         if (instr->dest.dest.is_ssa) {
 251                 /* Construct a writemask */
 252                 unsigned bits_per_comp = instr->dest.dest.ssa.bit_size;
 253                 unsigned comps = instr->dest.dest.ssa.num_components;
 254                 assert(comps == 1);
 255                 unsigned bits = bits_per_comp * comps;
 256                 unsigned bytes = MAX2(bits / 8, 1);
 257                 alu.writemask = (1 << bytes) - 1;
 258         } else {
 259                 unsigned comp_mask = instr->dest.write_mask;
 260
 261                 alu.writemask = pan_to_bytemask(nir_dest_bit_size(instr->dest.dest),
 262                                 comp_mask);
 263         }
 264
 265         /* Copy sources */
 266
 267         unsigned num_inputs = nir_op_infos[instr->op].num_inputs;
 268         assert(num_inputs <= ARRAY_SIZE(alu.src));
 269
 270         for (unsigned i = 0; i < num_inputs; ++i) {
 271                 alu.src[i] = bir_src_index(&instr->src[i].src);
 272
 273                 alu.src_types[i] = nir_op_infos[instr->op].input_types[i]
 274                         | nir_src_bit_size(instr->src[i].src);
 275
 276                 /* We assert scalarization above */
 277                 alu.swizzle[i][0] = instr->src[i].swizzle[0];
 278         }
 279
 280         /* Op-specific fixup */
 281         switch (instr->op) {
 282         case nir_op_fmul:
 283                 alu.src[2] = BIR_INDEX_ZERO; /* FMA */
 284                 break;
 285         default:
 286                 break;
 287         }
 288
 289         bi_emit(ctx, alu);
 290 }
 291
 292 static void
 293 emit_instr(bi_context *ctx, struct nir_instr *instr)
 294 {
 295         switch (instr->type) {
 296         case nir_instr_type_load_const:
 297                 emit_load_const(ctx, nir_instr_as_load_const(instr));
 298                 break;
 299
 300         case nir_instr_type_intrinsic:
 301                 emit_intrinsic(ctx, nir_instr_as_intrinsic(instr));
 302                 break;
 303
 304         case nir_instr_type_alu:
 305                 emit_alu(ctx, nir_instr_as_alu(instr));
 306                 break;
 307
 308 #if 0
 309         case nir_instr_type_tex:
 310                 emit_tex(ctx, nir_instr_as_tex(instr));
 311                 break;
 312 #endif
 313
 314         case nir_instr_type_jump:
 315                 emit_jump(ctx, nir_instr_as_jump(instr));
 316                 break;
 317
 318         case nir_instr_type_ssa_undef:
 319                 /* Spurious */
 320                 break;
 321
 322         default:
 323                 //unreachable("Unhandled instruction type");
 324                 break;
 325         }
 326 }
 327
 328
 329
 330 static bi_block *
 331 create_empty_block(bi_context *ctx)
 332 {
 333         bi_block *blk = rzalloc(ctx, bi_block);
 334
 335         blk->predecessors = _mesa_set_create(blk,
 336                         _mesa_hash_pointer,
 337                         _mesa_key_pointer_equal);
 338
 339         blk->name = ctx->block_name_count++;
 340
 341         return blk;
 342 }
 343
 344 static void
 345 bi_block_add_successor(bi_block *block, bi_block *successor)
 346 {
 347         assert(block);
 348         assert(successor);
 349
 350         for (unsigned i = 0; i < ARRAY_SIZE(block->successors); ++i) {
 351                 if (block->successors[i]) {
 352                        if (block->successors[i] == successor)
 353                                return;
 354                        else
 355                                continue;
 356                 }
 357
 358                 block->successors[i] = successor;
 359                 _mesa_set_add(successor->predecessors, block);
 360                 return;
 361         }
 362
 363         unreachable("Too many successors");
 364 }
 365
 366 static void
 367 bi_schedule_barrier(bi_context *ctx)
 368 {
 369         bi_block *temp = ctx->after_block;
 370         ctx->after_block = create_empty_block(ctx);
 371         list_addtail(&ctx->after_block->link, &ctx->blocks);
 372         list_inithead(&ctx->after_block->instructions);
 373         bi_block_add_successor(ctx->current_block, ctx->after_block);
 374         ctx->current_block = ctx->after_block;
 375         ctx->after_block = temp;
 376 }
 377
 378 static bi_block *
 379 emit_block(bi_context *ctx, nir_block *block)
 380 {
 381         if (ctx->after_block) {
 382                 ctx->current_block = ctx->after_block;
 383                 ctx->after_block = NULL;
 384         } else {
 385                 ctx->current_block = create_empty_block(ctx);
 386         }
 387
 388         list_addtail(&ctx->current_block->link, &ctx->blocks);
 389         list_inithead(&ctx->current_block->instructions);
 390
 391         nir_foreach_instr(instr, block) {
 392                 emit_instr(ctx, instr);
 393                 ++ctx->instruction_count;
 394         }
 395
 396         return ctx->current_block;
 397 }
 398
 399 /* Emits an unconditional branch to the end of the current block, returning a
 400  * pointer so the user can fill in details */
 401
 402 static bi_instruction *
 403 bi_emit_branch(bi_context *ctx)
 404 {
 405         bi_instruction branch = {
 406                 .type = BI_BRANCH,
 407                 .branch = {
 408                         .cond = BI_COND_ALWAYS
 409                 }
 410         };
 411
 412         return bi_emit(ctx, branch);
 413 }
 414
 415 /* Sets a condition for a branch by examing the NIR condition. If we're
 416  * familiar with the condition, we unwrap it to fold it into the branch
 417  * instruction. Otherwise, we consume the condition directly. We
 418  * generally use 1-bit booleans which allows us to use small types for
 419  * the conditions.
 420  */
 421
 422 static void
 423 bi_set_branch_cond(bi_instruction *branch, nir_src *cond, bool invert)
 424 {
 425         /* TODO: Try to unwrap instead of always bailing */
 426         branch->src[0] = bir_src_index(cond);
 427         branch->src[1] = BIR_INDEX_ZERO;
 428         branch->src_types[0] = branch->src_types[1] = nir_type_uint16;
 429         branch->branch.cond = invert ? BI_COND_EQ : BI_COND_NE;
 430 }
 431
 432 static void
 433 emit_if(bi_context *ctx, nir_if *nif)
 434 {
 435         bi_block *before_block = ctx->current_block;
 436
 437         /* Speculatively emit the branch, but we can't fill it in until later */
 438         bi_instruction *then_branch = bi_emit_branch(ctx);
 439         bi_set_branch_cond(then_branch, &nif->condition, true);
 440
 441         /* Emit the two subblocks. */
 442         bi_block *then_block = emit_cf_list(ctx, &nif->then_list);
 443         bi_block *end_then_block = ctx->current_block;
 444
 445         /* Emit a jump from the end of the then block to the end of the else */
 446         bi_instruction *then_exit = bi_emit_branch(ctx);
 447
 448         /* Emit second block, and check if it's empty */
 449
 450         int count_in = ctx->instruction_count;
 451         bi_block *else_block = emit_cf_list(ctx, &nif->else_list);
 452         bi_block *end_else_block = ctx->current_block;
 453         ctx->after_block = create_empty_block(ctx);
 454
 455         /* Now that we have the subblocks emitted, fix up the branches */
 456
 457         assert(then_block);
 458         assert(else_block);
 459
 460         if (ctx->instruction_count == count_in) {
 461                 /* The else block is empty, so don't emit an exit jump */
 462                 bi_remove_instruction(then_exit);
 463                 then_branch->branch.target = ctx->after_block;
 464         } else {
 465                 then_branch->branch.target = else_block;
 466                 then_exit->branch.target = ctx->after_block;
 467                 bi_block_add_successor(end_then_block, then_exit->branch.target);
 468         }
 469
 470         /* Wire up the successors */
 471
 472         bi_block_add_successor(before_block, then_branch->branch.target); /* then_branch */
 473
 474         bi_block_add_successor(before_block, then_block); /* fallthrough */
 475         bi_block_add_successor(end_else_block, ctx->after_block); /* fallthrough */
 476 }
 477
 478 static void
 479 emit_loop(bi_context *ctx, nir_loop *nloop)
 480 {
 481         /* Remember where we are */
 482         bi_block *start_block = ctx->current_block;
 483
 484         bi_block *saved_break = ctx->break_block;
 485         bi_block *saved_continue = ctx->continue_block;
 486
 487         ctx->continue_block = create_empty_block(ctx);
 488         ctx->break_block = create_empty_block(ctx);
 489         ctx->after_block = ctx->continue_block;
 490
 491         /* Emit the body itself */
 492         emit_cf_list(ctx, &nloop->body);
 493
 494         /* Branch back to loop back */
 495         bi_instruction *br_back = bi_emit_branch(ctx);
 496         br_back->branch.target = ctx->continue_block;
 497         bi_block_add_successor(start_block, ctx->continue_block);
 498         bi_block_add_successor(ctx->current_block, ctx->continue_block);
 499
 500         ctx->after_block = ctx->break_block;
 501
 502         /* Pop off */
 503         ctx->break_block = saved_break;
 504         ctx->continue_block = saved_continue;
 505         ++ctx->loop_count;
 506 }
 507
 508 static bi_block *
 509 emit_cf_list(bi_context *ctx, struct exec_list *list)
 510 {
 511         bi_block *start_block = NULL;
 512
 513         foreach_list_typed(nir_cf_node, node, node, list) {
 514                 switch (node->type) {
 515                 case nir_cf_node_block: {
 516                         bi_block *block = emit_block(ctx, nir_cf_node_as_block(node));
 517
 518                         if (!start_block)
 519                                 start_block = block;
 520
 521                         break;
 522                 }
 523
 524                 case nir_cf_node_if:
 525                         emit_if(ctx, nir_cf_node_as_if(node));
 526                         break;
 527
 528                 case nir_cf_node_loop:
 529                         emit_loop(ctx, nir_cf_node_as_loop(node));
 530                         break;
 531
 532                 default:
 533                         unreachable("Unknown control flow");
 534                 }
 535         }
 536
 537         return start_block;
 538 }
 539
 540 static int
 541 glsl_type_size(const struct glsl_type *type, bool bindless)
 542 {
 543         return glsl_count_attribute_slots(type, false);
 544 }
 545
 546 static void
 547 bi_optimize_nir(nir_shader *nir)
 548 {
 549         bool progress;
 550         unsigned lower_flrp = 16 | 32 | 64;
 551
 552         NIR_PASS(progress, nir, nir_lower_regs_to_ssa);
 553         NIR_PASS(progress, nir, nir_lower_idiv, nir_lower_idiv_fast);
 554
 555         nir_lower_tex_options lower_tex_options = {
 556                 .lower_txs_lod = true,
 557                 .lower_txp = ~0,
 558                 .lower_tex_without_implicit_lod = true,
 559                 .lower_txd = true,
 560         };
 561
 562         NIR_PASS(progress, nir, nir_lower_tex, &lower_tex_options);
 563         NIR_PASS(progress, nir, nir_lower_alu_to_scalar, NULL, NULL);
 564         NIR_PASS(progress, nir, nir_lower_load_const_to_scalar);
 565
 566         do {
 567                 progress = false;
 568
 569                 NIR_PASS(progress, nir, nir_lower_var_copies);
 570                 NIR_PASS(progress, nir, nir_lower_vars_to_ssa);
 571
 572                 NIR_PASS(progress, nir, nir_copy_prop);
 573                 NIR_PASS(progress, nir, nir_opt_remove_phis);
 574                 NIR_PASS(progress, nir, nir_opt_dce);
 575                 NIR_PASS(progress, nir, nir_opt_dead_cf);
 576                 NIR_PASS(progress, nir, nir_opt_cse);
 577                 NIR_PASS(progress, nir, nir_opt_peephole_select, 64, false, true);
 578                 NIR_PASS(progress, nir, nir_opt_algebraic);
 579                 NIR_PASS(progress, nir, nir_opt_constant_folding);
 580
 581                 if (lower_flrp != 0) {
 582                         bool lower_flrp_progress = false;
 583                         NIR_PASS(lower_flrp_progress,
 584                                  nir,
 585                                  nir_lower_flrp,
 586                                  lower_flrp,
 587                                  false /* always_precise */,
 588                                  nir->options->lower_ffma);
 589                         if (lower_flrp_progress) {
 590                                 NIR_PASS(progress, nir,
 591                                          nir_opt_constant_folding);
 592                                 progress = true;
 593                         }
 594
 595                         /* Nothing should rematerialize any flrps, so we only
 596                          * need to do this lowering once.
 597                          */
 598                         lower_flrp = 0;
 599                 }
 600
 601                 NIR_PASS(progress, nir, nir_opt_undef);
 602                 NIR_PASS(progress, nir, nir_opt_loop_unroll,
 603                          nir_var_shader_in |
 604                          nir_var_shader_out |
 605                          nir_var_function_temp);
 606         } while (progress);
 607
 608         NIR_PASS(progress, nir, nir_opt_algebraic_late);
 609         NIR_PASS(progress, nir, nir_lower_alu_to_scalar, NULL, NULL);
 610         NIR_PASS(progress, nir, nir_lower_load_const_to_scalar);
 611
 612         /* Take us out of SSA */
 613         NIR_PASS(progress, nir, nir_lower_locals_to_regs);
 614         NIR_PASS(progress, nir, nir_convert_from_ssa, true);
 615
 616         /* We're a primary scalar architecture but there's enough vector that
 617          * we use a vector IR so let's not also deal with scalar hacks on top
 618          * of the vector hacks */
 619
 620         NIR_PASS(progress, nir, nir_move_vec_src_uses_to_dest);
 621         NIR_PASS(progress, nir, nir_lower_vec_to_movs);
 622         NIR_PASS(progress, nir, nir_opt_dce);
 623 }
 624
 625 void
 626 bifrost_compile_shader_nir(nir_shader *nir, bifrost_program *program, unsigned product_id)
 627 {
 628         bi_context *ctx = rzalloc(NULL, bi_context);
 629         ctx->nir = nir;
 630         ctx->stage = nir->info.stage;
 631         ctx->quirks = bifrost_get_quirks(product_id);
 632         list_inithead(&ctx->blocks);
 633
 634         /* Lower gl_Position pre-optimisation, but after lowering vars to ssa
 635          * (so we don't accidentally duplicate the epilogue since mesa/st has
 636          * messed with our I/O quite a bit already) */
 637
 638         NIR_PASS_V(nir, nir_lower_vars_to_ssa);
 639
 640         if (ctx->stage == MESA_SHADER_VERTEX) {
 641                 NIR_PASS_V(nir, nir_lower_viewport_transform);
 642                 NIR_PASS_V(nir, nir_lower_point_size, 1.0, 1024.0);
 643         }
 644
 645         NIR_PASS_V(nir, nir_split_var_copies);
 646         NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 647         NIR_PASS_V(nir, nir_lower_var_copies);
 648         NIR_PASS_V(nir, nir_lower_vars_to_ssa);
 649         NIR_PASS_V(nir, nir_lower_io, nir_var_all, glsl_type_size, 0);
 650         NIR_PASS_V(nir, nir_lower_ssbo);
 651
 652         bi_optimize_nir(nir);
 653         nir_print_shader(nir, stdout);
 654
 655         nir_foreach_function(func, nir) {
 656                 if (!func->impl)
 657                         continue;
 658
 659                 ctx->impl = func->impl;
 660                 emit_cf_list(ctx, &func->impl->body);
 661                 break; /* TODO: Multi-function shaders */
 662         }
 663
 664         bi_print_shader(ctx, stdout);
 665         bi_schedule(ctx);
 666
 667         ralloc_free(ctx);
 668 }