src/gallium/drivers/vc4/vc4_qir.c

   1 /*
   2  * Copyright © 2014 Broadcom
   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
  24 #include "util/u_memory.h"
  25 #include "util/ralloc.h"
  26
  27 #include "vc4_qir.h"
  28 #include "vc4_qpu.h"
  29
  30 struct qir_op_info {
  31         const char *name;
  32         uint8_t ndst, nsrc;
  33         bool has_side_effects;
  34 };
  35
  36 static const struct qir_op_info qir_op_info[] = {
  37         [QOP_MOV] = { "mov", 1, 1 },
  38         [QOP_FMOV] = { "fmov", 1, 1 },
  39         [QOP_MMOV] = { "mmov", 1, 1 },
  40         [QOP_FADD] = { "fadd", 1, 2 },
  41         [QOP_FSUB] = { "fsub", 1, 2 },
  42         [QOP_FMUL] = { "fmul", 1, 2 },
  43         [QOP_MUL24] = { "mul24", 1, 2 },
  44         [QOP_V8MULD] = {"v8muld", 1, 2 },
  45         [QOP_V8MIN] = {"v8min", 1, 2 },
  46         [QOP_V8MAX] = {"v8max", 1, 2 },
  47         [QOP_V8ADDS] = {"v8adds", 1, 2 },
  48         [QOP_V8SUBS] = {"v8subs", 1, 2 },
  49         [QOP_FMIN] = { "fmin", 1, 2 },
  50         [QOP_FMAX] = { "fmax", 1, 2 },
  51         [QOP_FMINABS] = { "fminabs", 1, 2 },
  52         [QOP_FMAXABS] = { "fmaxabs", 1, 2 },
  53         [QOP_FTOI] = { "ftoi", 1, 1 },
  54         [QOP_ITOF] = { "itof", 1, 1 },
  55         [QOP_ADD] = { "add", 1, 2 },
  56         [QOP_SUB] = { "sub", 1, 2 },
  57         [QOP_SHR] = { "shr", 1, 2 },
  58         [QOP_ASR] = { "asr", 1, 2 },
  59         [QOP_SHL] = { "shl", 1, 2 },
  60         [QOP_MIN] = { "min", 1, 2 },
  61         [QOP_MAX] = { "max", 1, 2 },
  62         [QOP_AND] = { "and", 1, 2 },
  63         [QOP_OR] = { "or", 1, 2 },
  64         [QOP_XOR] = { "xor", 1, 2 },
  65         [QOP_NOT] = { "not", 1, 1 },
  66
  67         [QOP_RCP] = { "rcp", 1, 1 },
  68         [QOP_RSQ] = { "rsq", 1, 1 },
  69         [QOP_EXP2] = { "exp2", 1, 1 },
  70         [QOP_LOG2] = { "log2", 1, 1 },
  71         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0 },
  72         [QOP_MS_MASK] = { "ms_mask", 0, 1, true },
  73         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  74
  75         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  76         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  77
  78         [QOP_TEX_S] = { "tex_s", 0, 2, true },
  79         [QOP_TEX_T] = { "tex_t", 0, 2, true },
  80         [QOP_TEX_R] = { "tex_r", 0, 2, true },
  81         [QOP_TEX_B] = { "tex_b", 0, 2, true },
  82         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2, true },
  83         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
  84
  85         [QOP_LOAD_IMM] = { "load_imm", 0, 1 },
  86         [QOP_LOAD_IMM_U2] = { "load_imm_u2", 0, 1 },
  87         [QOP_LOAD_IMM_I2] = { "load_imm_i2", 0, 1 },
  88
  89         [QOP_ROT_MUL] = { "rot_mul", 0, 2 },
  90
  91         [QOP_BRANCH] = { "branch", 0, 0, true },
  92         [QOP_UNIFORMS_RESET] = { "uniforms_reset", 0, 2, true },
  93 };
  94
  95 static const char *
  96 qir_get_op_name(enum qop qop)
  97 {
  98         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
  99                 return qir_op_info[qop].name;
 100         else
 101                 return "???";
 102 }
 103
 104 int
 105 qir_get_op_nsrc(enum qop qop)
 106 {
 107         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 108                 return qir_op_info[qop].nsrc;
 109         else
 110                 abort();
 111 }
 112
 113 /**
 114  * Returns whether the instruction has any side effects that must be
 115  * preserved.
 116  */
 117 bool
 118 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 119 {
 120         switch (inst->dst.file) {
 121         case QFILE_TLB_Z_WRITE:
 122         case QFILE_TLB_COLOR_WRITE:
 123         case QFILE_TLB_COLOR_WRITE_MS:
 124         case QFILE_TLB_STENCIL_SETUP:
 125                 return true;
 126         default:
 127                 break;
 128         }
 129
 130         return qir_op_info[inst->op].has_side_effects;
 131 }
 132
 133 bool
 134 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 135 {
 136         /* We can dead-code eliminate varyings, because we only tell the VS
 137          * about the live ones at the end.  But we have to preserve the
 138          * point/line coordinates reads, because they're generated by
 139          * fixed-function hardware.
 140          */
 141         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 142                 if (inst->src[i].file == QFILE_VARY &&
 143                     c->input_slots[inst->src[i].index].slot == 0xff) {
 144                         return true;
 145                 }
 146
 147                 if (inst->src[i].file == QFILE_VPM)
 148                         return true;
 149         }
 150
 151         if (inst->dst.file == QFILE_VPM)
 152                 return true;
 153
 154         return false;
 155 }
 156
 157 bool
 158 qir_is_mul(struct qinst *inst)
 159 {
 160         switch (inst->op) {
 161         case QOP_MMOV:
 162         case QOP_FMUL:
 163         case QOP_MUL24:
 164         case QOP_V8MULD:
 165         case QOP_V8MIN:
 166         case QOP_V8MAX:
 167         case QOP_V8ADDS:
 168         case QOP_V8SUBS:
 169         case QOP_ROT_MUL:
 170                 return true;
 171         default:
 172                 return false;
 173         }
 174 }
 175
 176 bool
 177 qir_is_float_input(struct qinst *inst)
 178 {
 179         switch (inst->op) {
 180         case QOP_FMOV:
 181         case QOP_FMUL:
 182         case QOP_FADD:
 183         case QOP_FSUB:
 184         case QOP_FMIN:
 185         case QOP_FMAX:
 186         case QOP_FMINABS:
 187         case QOP_FMAXABS:
 188         case QOP_FTOI:
 189                 return true;
 190         default:
 191                 return false;
 192         }
 193 }
 194
 195 bool
 196 qir_is_raw_mov(struct qinst *inst)
 197 {
 198         return ((inst->op == QOP_MOV ||
 199                  inst->op == QOP_FMOV ||
 200                  inst->op == QOP_MMOV) &&
 201                 inst->cond == QPU_COND_ALWAYS &&
 202                 !inst->dst.pack &&
 203                 !inst->src[0].pack);
 204 }
 205
 206 bool
 207 qir_is_tex(struct qinst *inst)
 208 {
 209         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 210 }
 211
 212 bool
 213 qir_depends_on_flags(struct qinst *inst)
 214 {
 215         if (inst->op == QOP_BRANCH) {
 216                 return inst->cond != QPU_COND_BRANCH_ALWAYS;
 217         } else {
 218                 return (inst->cond != QPU_COND_ALWAYS &&
 219                         inst->cond != QPU_COND_NEVER);
 220         }
 221 }
 222
 223 bool
 224 qir_writes_r4(struct qinst *inst)
 225 {
 226         switch (inst->op) {
 227         case QOP_TEX_RESULT:
 228         case QOP_TLB_COLOR_READ:
 229         case QOP_RCP:
 230         case QOP_RSQ:
 231         case QOP_EXP2:
 232         case QOP_LOG2:
 233                 return true;
 234         default:
 235                 return false;
 236         }
 237 }
 238
 239 uint8_t
 240 qir_channels_written(struct qinst *inst)
 241 {
 242         if (qir_is_mul(inst)) {
 243                 switch (inst->dst.pack) {
 244                 case QPU_PACK_MUL_NOP:
 245                 case QPU_PACK_MUL_8888:
 246                         return 0xf;
 247                 case QPU_PACK_MUL_8A:
 248                         return 0x1;
 249                 case QPU_PACK_MUL_8B:
 250                         return 0x2;
 251                 case QPU_PACK_MUL_8C:
 252                         return 0x4;
 253                 case QPU_PACK_MUL_8D:
 254                         return 0x8;
 255                 }
 256         } else {
 257                 switch (inst->dst.pack) {
 258                 case QPU_PACK_A_NOP:
 259                 case QPU_PACK_A_8888:
 260                 case QPU_PACK_A_8888_SAT:
 261                 case QPU_PACK_A_32_SAT:
 262                         return 0xf;
 263                 case QPU_PACK_A_8A:
 264                 case QPU_PACK_A_8A_SAT:
 265                         return 0x1;
 266                 case QPU_PACK_A_8B:
 267                 case QPU_PACK_A_8B_SAT:
 268                         return 0x2;
 269                 case QPU_PACK_A_8C:
 270                 case QPU_PACK_A_8C_SAT:
 271                         return 0x4;
 272                 case QPU_PACK_A_8D:
 273                 case QPU_PACK_A_8D_SAT:
 274                         return 0x8;
 275                 case QPU_PACK_A_16A:
 276                 case QPU_PACK_A_16A_SAT:
 277                         return 0x3;
 278                 case QPU_PACK_A_16B:
 279                 case QPU_PACK_A_16B_SAT:
 280                         return 0xc;
 281                 }
 282         }
 283         unreachable("Bad pack field");
 284 }
 285
 286 static void
 287 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 288 {
 289         static const char *files[] = {
 290                 [QFILE_TEMP] = "t",
 291                 [QFILE_VARY] = "v",
 292                 [QFILE_UNIF] = "u",
 293                 [QFILE_TLB_COLOR_WRITE] = "tlb_c",
 294                 [QFILE_TLB_COLOR_WRITE_MS] = "tlb_c_ms",
 295                 [QFILE_TLB_Z_WRITE] = "tlb_z",
 296                 [QFILE_TLB_STENCIL_SETUP] = "tlb_stencil",
 297                 [QFILE_FRAG_X] = "frag_x",
 298                 [QFILE_FRAG_Y] = "frag_y",
 299                 [QFILE_FRAG_REV_FLAG] = "frag_rev_flag",
 300                 [QFILE_QPU_ELEMENT] = "elem",
 301         };
 302
 303         switch (reg.file) {
 304
 305         case QFILE_NULL:
 306                 fprintf(stderr, "null");
 307                 break;
 308
 309         case QFILE_LOAD_IMM:
 310                 fprintf(stderr, "0x%08x (%f)", reg.index, uif(reg.index));
 311                 break;
 312
 313         case QFILE_SMALL_IMM:
 314                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 315                         fprintf(stderr, "%d", reg.index);
 316                 else
 317                         fprintf(stderr, "%f", uif(reg.index));
 318                 break;
 319
 320         case QFILE_VPM:
 321                 if (write) {
 322                         fprintf(stderr, "vpm");
 323                 } else {
 324                         fprintf(stderr, "vpm%d.%d",
 325                                 reg.index / 4, reg.index % 4);
 326                 }
 327                 break;
 328
 329         case QFILE_TLB_COLOR_WRITE:
 330         case QFILE_TLB_COLOR_WRITE_MS:
 331         case QFILE_TLB_Z_WRITE:
 332         case QFILE_TLB_STENCIL_SETUP:
 333                 fprintf(stderr, "%s", files[reg.file]);
 334                 break;
 335
 336         default:
 337                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 338                 break;
 339         }
 340
 341         if (reg.file == QFILE_UNIF &&
 342             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 343                 fprintf(stderr, " (0x%08x / %f)",
 344                         c->uniform_data[reg.index],
 345                         uif(c->uniform_data[reg.index]));
 346         }
 347 }
 348
 349 void
 350 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 351 {
 352         fprintf(stderr, "%s", qir_get_op_name(inst->op));
 353         if (inst->op == QOP_BRANCH)
 354                 vc4_qpu_disasm_cond_branch(stderr, inst->cond);
 355         else
 356                 vc4_qpu_disasm_cond(stderr, inst->cond);
 357         if (inst->sf)
 358                 fprintf(stderr, ".sf");
 359         fprintf(stderr, " ");
 360
 361         if (inst->op != QOP_BRANCH) {
 362                 qir_print_reg(c, inst->dst, true);
 363                 if (inst->dst.pack) {
 364                         if (inst->dst.pack) {
 365                                 if (qir_is_mul(inst))
 366                                         vc4_qpu_disasm_pack_mul(stderr, inst->dst.pack);
 367                                 else
 368                                         vc4_qpu_disasm_pack_a(stderr, inst->dst.pack);
 369                         }
 370                 }
 371         }
 372
 373         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 374                 fprintf(stderr, ", ");
 375                 qir_print_reg(c, inst->src[i], false);
 376                 vc4_qpu_disasm_unpack(stderr, inst->src[i].pack);
 377         }
 378 }
 379
 380 void
 381 qir_dump(struct vc4_compile *c)
 382 {
 383         int ip = 0;
 384
 385         qir_for_each_block(block, c) {
 386                 fprintf(stderr, "BLOCK %d:\n", block->index);
 387                 qir_for_each_inst(inst, block) {
 388                         if (c->temp_start) {
 389                                 bool first = true;
 390
 391                                 for (int i = 0; i < c->num_temps; i++) {
 392                                         if (c->temp_start[i] != ip)
 393                                                 continue;
 394
 395                                         if (first) {
 396                                                 first = false;
 397                                         } else {
 398                                                 fprintf(stderr, ", ");
 399                                         }
 400                                         fprintf(stderr, "S%4d", i);
 401                                 }
 402
 403                                 if (first)
 404                                         fprintf(stderr, "      ");
 405                                 else
 406                                         fprintf(stderr, " ");
 407                         }
 408
 409                         if (c->temp_end) {
 410                                 bool first = true;
 411
 412                                 for (int i = 0; i < c->num_temps; i++) {
 413                                         if (c->temp_end[i] != ip)
 414                                                 continue;
 415
 416                                         if (first) {
 417                                                 first = false;
 418                                         } else {
 419                                                 fprintf(stderr, ", ");
 420                                         }
 421                                         fprintf(stderr, "E%4d", i);
 422                                 }
 423
 424                                 if (first)
 425                                         fprintf(stderr, "      ");
 426                                 else
 427                                         fprintf(stderr, " ");
 428                         }
 429
 430                         qir_dump_inst(c, inst);
 431                         fprintf(stderr, "\n");
 432                         ip++;
 433                 }
 434                 if (block->successors[1]) {
 435                         fprintf(stderr, "-> BLOCK %d, %d\n",
 436                                 block->successors[0]->index,
 437                                 block->successors[1]->index);
 438                 } else if (block->successors[0]) {
 439                         fprintf(stderr, "-> BLOCK %d\n",
 440                                 block->successors[0]->index);
 441                 }
 442         }
 443 }
 444
 445 struct qreg
 446 qir_get_temp(struct vc4_compile *c)
 447 {
 448         struct qreg reg;
 449
 450         reg.file = QFILE_TEMP;
 451         reg.index = c->num_temps++;
 452         reg.pack = 0;
 453
 454         if (c->num_temps > c->defs_array_size) {
 455                 uint32_t old_size = c->defs_array_size;
 456                 c->defs_array_size = MAX2(old_size * 2, 16);
 457                 c->defs = reralloc(c, c->defs, struct qinst *,
 458                                    c->defs_array_size);
 459                 memset(&c->defs[old_size], 0,
 460                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 461         }
 462
 463         return reg;
 464 }
 465
 466 struct qinst *
 467 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 468 {
 469         struct qinst *inst = CALLOC_STRUCT(qinst);
 470
 471         inst->op = op;
 472         inst->dst = dst;
 473         inst->src = calloc(2, sizeof(inst->src[0]));
 474         inst->src[0] = src0;
 475         inst->src[1] = src1;
 476         inst->cond = QPU_COND_ALWAYS;
 477
 478         return inst;
 479 }
 480
 481 struct qinst *
 482 qir_inst4(enum qop op, struct qreg dst,
 483           struct qreg a,
 484           struct qreg b,
 485           struct qreg c,
 486           struct qreg d)
 487 {
 488         struct qinst *inst = CALLOC_STRUCT(qinst);
 489
 490         inst->op = op;
 491         inst->dst = dst;
 492         inst->src = calloc(4, sizeof(*inst->src));
 493         inst->src[0] = a;
 494         inst->src[1] = b;
 495         inst->src[2] = c;
 496         inst->src[3] = d;
 497
 498         return inst;
 499 }
 500
 501 static void
 502 qir_emit(struct vc4_compile *c, struct qinst *inst)
 503 {
 504         list_addtail(&inst->link, &c->cur_block->instructions);
 505 }
 506
 507 /* Updates inst to write to a new temporary, emits it, and notes the def. */
 508 struct qreg
 509 qir_emit_def(struct vc4_compile *c, struct qinst *inst)
 510 {
 511         assert(inst->dst.file == QFILE_NULL);
 512
 513         inst->dst = qir_get_temp(c);
 514
 515         if (inst->dst.file == QFILE_TEMP)
 516                 c->defs[inst->dst.index] = inst;
 517
 518         qir_emit(c, inst);
 519
 520         return inst->dst;
 521 }
 522
 523 struct qinst *
 524 qir_emit_nondef(struct vc4_compile *c, struct qinst *inst)
 525 {
 526         if (inst->dst.file == QFILE_TEMP)
 527                 c->defs[inst->dst.index] = NULL;
 528
 529         qir_emit(c, inst);
 530
 531         return inst;
 532 }
 533
 534 bool
 535 qir_reg_equals(struct qreg a, struct qreg b)
 536 {
 537         return a.file == b.file && a.index == b.index && a.pack == b.pack;
 538 }
 539
 540 struct qblock *
 541 qir_new_block(struct vc4_compile *c)
 542 {
 543         struct qblock *block = rzalloc(c, struct qblock);
 544
 545         list_inithead(&block->instructions);
 546         list_inithead(&block->qpu_inst_list);
 547
 548         block->predecessors = _mesa_set_create(block,
 549                                                _mesa_hash_pointer,
 550                                                _mesa_key_pointer_equal);
 551
 552         block->index = c->next_block_index++;
 553
 554         return block;
 555 }
 556
 557 void
 558 qir_set_emit_block(struct vc4_compile *c, struct qblock *block)
 559 {
 560         c->cur_block = block;
 561         list_addtail(&block->link, &c->blocks);
 562 }
 563
 564 struct qblock *
 565 qir_entry_block(struct vc4_compile *c)
 566 {
 567         return list_first_entry(&c->blocks, struct qblock, link);
 568 }
 569
 570 struct qblock *
 571 qir_exit_block(struct vc4_compile *c)
 572 {
 573         return list_last_entry(&c->blocks, struct qblock, link);
 574 }
 575
 576 void
 577 qir_link_blocks(struct qblock *predecessor, struct qblock *successor)
 578 {
 579         _mesa_set_add(successor->predecessors, predecessor);
 580         if (predecessor->successors[0]) {
 581                 assert(!predecessor->successors[1]);
 582                 predecessor->successors[1] = successor;
 583         } else {
 584                 predecessor->successors[0] = successor;
 585         }
 586 }
 587
 588 struct vc4_compile *
 589 qir_compile_init(void)
 590 {
 591         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 592
 593         list_inithead(&c->blocks);
 594         qir_set_emit_block(c, qir_new_block(c));
 595
 596         c->output_position_index = -1;
 597         c->output_color_index = -1;
 598         c->output_point_size_index = -1;
 599         c->output_sample_mask_index = -1;
 600
 601         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 602                                             _mesa_key_pointer_equal);
 603
 604         return c;
 605 }
 606
 607 void
 608 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 609 {
 610         if (qinst->dst.file == QFILE_TEMP)
 611                 c->defs[qinst->dst.index] = NULL;
 612
 613         list_del(&qinst->link);
 614         free(qinst->src);
 615         free(qinst);
 616 }
 617
 618 struct qreg
 619 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 620 {
 621         int pack = reg.pack;
 622
 623         while (reg.file == QFILE_TEMP &&
 624                c->defs[reg.index] &&
 625                (c->defs[reg.index]->op == QOP_MOV ||
 626                 c->defs[reg.index]->op == QOP_FMOV ||
 627                 c->defs[reg.index]->op == QOP_MMOV)&&
 628                !c->defs[reg.index]->dst.pack &&
 629                !c->defs[reg.index]->src[0].pack) {
 630                 reg = c->defs[reg.index]->src[0];
 631         }
 632
 633         reg.pack = pack;
 634         return reg;
 635 }
 636
 637 void
 638 qir_compile_destroy(struct vc4_compile *c)
 639 {
 640         qir_for_each_block(block, c) {
 641                 while (!list_empty(&block->instructions)) {
 642                         struct qinst *qinst =
 643                                 list_first_entry(&block->instructions,
 644                                                  struct qinst, link);
 645                         qir_remove_instruction(c, qinst);
 646                 }
 647         }
 648
 649         ralloc_free(c);
 650 }
 651
 652 const char *
 653 qir_get_stage_name(enum qstage stage)
 654 {
 655         static const char *names[] = {
 656                 [QSTAGE_FRAG] = "FS",
 657                 [QSTAGE_VERT] = "VS",
 658                 [QSTAGE_COORD] = "CS",
 659         };
 660
 661         return names[stage];
 662 }
 663
 664 struct qreg
 665 qir_uniform(struct vc4_compile *c,
 666             enum quniform_contents contents,
 667             uint32_t data)
 668 {
 669         for (int i = 0; i < c->num_uniforms; i++) {
 670                 if (c->uniform_contents[i] == contents &&
 671                     c->uniform_data[i] == data) {
 672                         return qir_reg(QFILE_UNIF, i);
 673                 }
 674         }
 675
 676         uint32_t uniform = c->num_uniforms++;
 677
 678         if (uniform >= c->uniform_array_size) {
 679                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 680                                              c->uniform_array_size * 2);
 681
 682                 c->uniform_data = reralloc(c, c->uniform_data,
 683                                            uint32_t,
 684                                            c->uniform_array_size);
 685                 c->uniform_contents = reralloc(c, c->uniform_contents,
 686                                                enum quniform_contents,
 687                                                c->uniform_array_size);
 688         }
 689
 690         c->uniform_contents[uniform] = contents;
 691         c->uniform_data[uniform] = data;
 692
 693         return qir_reg(QFILE_UNIF, uniform);
 694 }
 695
 696 void
 697 qir_SF(struct vc4_compile *c, struct qreg src)
 698 {
 699         struct qinst *last_inst = NULL;
 700
 701         if (!list_empty(&c->cur_block->instructions))
 702                 last_inst = (struct qinst *)c->cur_block->instructions.prev;
 703
 704         /* We don't have any way to guess which kind of MOV is implied. */
 705         assert(!src.pack);
 706
 707         if (src.file != QFILE_TEMP ||
 708             !c->defs[src.index] ||
 709             last_inst != c->defs[src.index]) {
 710                 last_inst = qir_MOV_dest(c, qir_reg(QFILE_NULL, 0), src);
 711                 last_inst = (struct qinst *)c->cur_block->instructions.prev;
 712         }
 713         last_inst->sf = true;
 714 }
 715
 716 #define OPTPASS(func)                                                   \
 717         do {                                                            \
 718                 bool stage_progress = func(c);                          \
 719                 if (stage_progress) {                                   \
 720                         progress = true;                                \
 721                         if (print_opt_debug) {                          \
 722                                 fprintf(stderr,                         \
 723                                         "QIR opt pass %2d: %s progress\n", \
 724                                         pass, #func);                   \
 725                         }                                               \
 726                         qir_validate(c);                                \
 727                 }                                                       \
 728         } while (0)
 729
 730 void
 731 qir_optimize(struct vc4_compile *c)
 732 {
 733         bool print_opt_debug = false;
 734         int pass = 1;
 735
 736         while (true) {
 737                 bool progress = false;
 738
 739                 OPTPASS(qir_opt_algebraic);
 740                 OPTPASS(qir_opt_constant_folding);
 741                 OPTPASS(qir_opt_copy_propagation);
 742                 OPTPASS(qir_opt_peephole_sf);
 743                 OPTPASS(qir_opt_dead_code);
 744                 OPTPASS(qir_opt_small_immediates);
 745                 OPTPASS(qir_opt_vpm);
 746
 747                 if (!progress)
 748                         break;
 749
 750                 pass++;
 751         }
 752 }