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, 2 },
  70         [QOP_LOG2] = { "log2", 1, 2 },
  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 },
  79         [QOP_TEX_T] = { "tex_t", 0, 2 },
  80         [QOP_TEX_R] = { "tex_r", 0, 2 },
  81         [QOP_TEX_B] = { "tex_b", 0, 2 },
  82         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
  83         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
  84 };
  85
  86 static const char *
  87 qir_get_op_name(enum qop qop)
  88 {
  89         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
  90                 return qir_op_info[qop].name;
  91         else
  92                 return "???";
  93 }
  94
  95 int
  96 qir_get_op_nsrc(enum qop qop)
  97 {
  98         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
  99                 return qir_op_info[qop].nsrc;
 100         else
 101                 abort();
 102 }
 103
 104 /**
 105  * Returns whether the instruction has any side effects that must be
 106  * preserved.
 107  */
 108 bool
 109 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 110 {
 111         switch (inst->dst.file) {
 112         case QFILE_TLB_Z_WRITE:
 113         case QFILE_TLB_COLOR_WRITE:
 114         case QFILE_TLB_COLOR_WRITE_MS:
 115         case QFILE_TLB_STENCIL_SETUP:
 116                 return true;
 117         default:
 118                 break;
 119         }
 120
 121         return qir_op_info[inst->op].has_side_effects;
 122 }
 123
 124 bool
 125 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 126 {
 127         /* We can dead-code eliminate varyings, because we only tell the VS
 128          * about the live ones at the end.  But we have to preserve the
 129          * point/line coordinates reads, because they're generated by
 130          * fixed-function hardware.
 131          */
 132         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 133                 if (inst->src[i].file == QFILE_VARY &&
 134                     c->input_slots[inst->src[i].index].slot == 0xff) {
 135                         return true;
 136                 }
 137
 138                 if (inst->src[i].file == QFILE_VPM)
 139                         return true;
 140         }
 141
 142         if (inst->dst.file == QFILE_VPM)
 143                 return true;
 144
 145         return false;
 146 }
 147
 148 bool
 149 qir_is_mul(struct qinst *inst)
 150 {
 151         switch (inst->op) {
 152         case QOP_MMOV:
 153         case QOP_FMUL:
 154         case QOP_MUL24:
 155         case QOP_V8MULD:
 156         case QOP_V8MIN:
 157         case QOP_V8MAX:
 158         case QOP_V8ADDS:
 159         case QOP_V8SUBS:
 160                 return true;
 161         default:
 162                 return false;
 163         }
 164 }
 165
 166 bool
 167 qir_is_float_input(struct qinst *inst)
 168 {
 169         switch (inst->op) {
 170         case QOP_FMOV:
 171         case QOP_FMUL:
 172         case QOP_FADD:
 173         case QOP_FSUB:
 174         case QOP_FMIN:
 175         case QOP_FMAX:
 176         case QOP_FMINABS:
 177         case QOP_FMAXABS:
 178         case QOP_FTOI:
 179                 return true;
 180         default:
 181                 return false;
 182         }
 183 }
 184
 185 bool
 186 qir_is_raw_mov(struct qinst *inst)
 187 {
 188         return ((inst->op == QOP_MOV ||
 189                  inst->op == QOP_FMOV ||
 190                  inst->op == QOP_MMOV) &&
 191                 inst->cond == QPU_COND_ALWAYS &&
 192                 !inst->dst.pack &&
 193                 !inst->src[0].pack);
 194 }
 195
 196 bool
 197 qir_is_tex(struct qinst *inst)
 198 {
 199         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 200 }
 201
 202 bool
 203 qir_depends_on_flags(struct qinst *inst)
 204 {
 205         return (inst->cond != QPU_COND_ALWAYS &&
 206                 inst->cond != QPU_COND_NEVER);
 207 }
 208
 209 bool
 210 qir_writes_r4(struct qinst *inst)
 211 {
 212         switch (inst->op) {
 213         case QOP_TEX_RESULT:
 214         case QOP_TLB_COLOR_READ:
 215         case QOP_RCP:
 216         case QOP_RSQ:
 217         case QOP_EXP2:
 218         case QOP_LOG2:
 219                 return true;
 220         default:
 221                 return false;
 222         }
 223 }
 224
 225 static void
 226 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 227 {
 228         static const char *files[] = {
 229                 [QFILE_TEMP] = "t",
 230                 [QFILE_VARY] = "v",
 231                 [QFILE_UNIF] = "u",
 232                 [QFILE_TLB_COLOR_WRITE] = "tlb_c",
 233                 [QFILE_TLB_COLOR_WRITE_MS] = "tlb_c_ms",
 234                 [QFILE_TLB_Z_WRITE] = "tlb_z",
 235                 [QFILE_TLB_STENCIL_SETUP] = "tlb_stencil",
 236                 [QFILE_FRAG_X] = "frag_x",
 237                 [QFILE_FRAG_Y] = "frag_y",
 238                 [QFILE_FRAG_REV_FLAG] = "frag_rev_flag",
 239         };
 240
 241         switch (reg.file) {
 242
 243         case QFILE_NULL:
 244                 fprintf(stderr, "null");
 245                 break;
 246
 247         case QFILE_SMALL_IMM:
 248                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 249                         fprintf(stderr, "%d", reg.index);
 250                 else
 251                         fprintf(stderr, "%f", uif(reg.index));
 252                 break;
 253
 254         case QFILE_VPM:
 255                 if (write) {
 256                         fprintf(stderr, "vpm");
 257                 } else {
 258                         fprintf(stderr, "vpm%d.%d",
 259                                 reg.index / 4, reg.index % 4);
 260                 }
 261                 break;
 262
 263         case QFILE_TLB_COLOR_WRITE:
 264         case QFILE_TLB_COLOR_WRITE_MS:
 265         case QFILE_TLB_Z_WRITE:
 266         case QFILE_TLB_STENCIL_SETUP:
 267                 fprintf(stderr, "%s", files[reg.file]);
 268                 break;
 269
 270         default:
 271                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 272                 break;
 273         }
 274
 275         if (reg.file == QFILE_UNIF &&
 276             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 277                 fprintf(stderr, " (0x%08x / %f)",
 278                         c->uniform_data[reg.index],
 279                         uif(c->uniform_data[reg.index]));
 280         }
 281 }
 282
 283 void
 284 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 285 {
 286         fprintf(stderr, "%s", qir_get_op_name(inst->op));
 287         vc4_qpu_disasm_cond(stderr, inst->cond);
 288         if (inst->sf)
 289                 fprintf(stderr, ".sf");
 290         fprintf(stderr, " ");
 291
 292         qir_print_reg(c, inst->dst, true);
 293         if (inst->dst.pack) {
 294                 if (inst->dst.pack) {
 295                         if (qir_is_mul(inst))
 296                                 vc4_qpu_disasm_pack_mul(stderr, inst->dst.pack);
 297                         else
 298                                 vc4_qpu_disasm_pack_a(stderr, inst->dst.pack);
 299                 }
 300         }
 301         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 302                 fprintf(stderr, ", ");
 303                 qir_print_reg(c, inst->src[i], false);
 304                 vc4_qpu_disasm_unpack(stderr, inst->src[i].pack);
 305         }
 306 }
 307
 308 void
 309 qir_dump(struct vc4_compile *c)
 310 {
 311         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 312                 qir_dump_inst(c, inst);
 313                 fprintf(stderr, "\n");
 314         }
 315 }
 316
 317 struct qreg
 318 qir_get_temp(struct vc4_compile *c)
 319 {
 320         struct qreg reg;
 321
 322         reg.file = QFILE_TEMP;
 323         reg.index = c->num_temps++;
 324         reg.pack = 0;
 325
 326         if (c->num_temps > c->defs_array_size) {
 327                 uint32_t old_size = c->defs_array_size;
 328                 c->defs_array_size = MAX2(old_size * 2, 16);
 329                 c->defs = reralloc(c, c->defs, struct qinst *,
 330                                    c->defs_array_size);
 331                 memset(&c->defs[old_size], 0,
 332                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 333         }
 334
 335         return reg;
 336 }
 337
 338 struct qinst *
 339 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 340 {
 341         struct qinst *inst = CALLOC_STRUCT(qinst);
 342
 343         inst->op = op;
 344         inst->dst = dst;
 345         inst->src = calloc(2, sizeof(inst->src[0]));
 346         inst->src[0] = src0;
 347         inst->src[1] = src1;
 348         inst->cond = QPU_COND_ALWAYS;
 349
 350         return inst;
 351 }
 352
 353 struct qinst *
 354 qir_inst4(enum qop op, struct qreg dst,
 355           struct qreg a,
 356           struct qreg b,
 357           struct qreg c,
 358           struct qreg d)
 359 {
 360         struct qinst *inst = CALLOC_STRUCT(qinst);
 361
 362         inst->op = op;
 363         inst->dst = dst;
 364         inst->src = calloc(4, sizeof(*inst->src));
 365         inst->src[0] = a;
 366         inst->src[1] = b;
 367         inst->src[2] = c;
 368         inst->src[3] = d;
 369
 370         return inst;
 371 }
 372
 373 void
 374 qir_emit(struct vc4_compile *c, struct qinst *inst)
 375 {
 376         if (inst->dst.file == QFILE_TEMP)
 377                 c->defs[inst->dst.index] = inst;
 378
 379         qir_emit_nodef(c, inst);
 380 }
 381
 382 bool
 383 qir_reg_equals(struct qreg a, struct qreg b)
 384 {
 385         return a.file == b.file && a.index == b.index;
 386 }
 387
 388 struct vc4_compile *
 389 qir_compile_init(void)
 390 {
 391         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 392
 393         list_inithead(&c->instructions);
 394
 395         c->output_position_index = -1;
 396         c->output_color_index = -1;
 397         c->output_point_size_index = -1;
 398         c->output_sample_mask_index = -1;
 399
 400         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 401                                             _mesa_key_pointer_equal);
 402
 403         return c;
 404 }
 405
 406 void
 407 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 408 {
 409         if (qinst->dst.file == QFILE_TEMP)
 410                 c->defs[qinst->dst.index] = NULL;
 411
 412         list_del(&qinst->link);
 413         free(qinst->src);
 414         free(qinst);
 415 }
 416
 417 struct qreg
 418 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 419 {
 420         int pack = reg.pack;
 421
 422         while (reg.file == QFILE_TEMP &&
 423                c->defs[reg.index] &&
 424                (c->defs[reg.index]->op == QOP_MOV ||
 425                 c->defs[reg.index]->op == QOP_FMOV ||
 426                 c->defs[reg.index]->op == QOP_MMOV)&&
 427                !c->defs[reg.index]->dst.pack &&
 428                !c->defs[reg.index]->src[0].pack) {
 429                 reg = c->defs[reg.index]->src[0];
 430         }
 431
 432         reg.pack = pack;
 433         return reg;
 434 }
 435
 436 void
 437 qir_compile_destroy(struct vc4_compile *c)
 438 {
 439         while (!list_empty(&c->instructions)) {
 440                 struct qinst *qinst =
 441                         (struct qinst *)c->instructions.next;
 442                 qir_remove_instruction(c, qinst);
 443         }
 444
 445         ralloc_free(c);
 446 }
 447
 448 const char *
 449 qir_get_stage_name(enum qstage stage)
 450 {
 451         static const char *names[] = {
 452                 [QSTAGE_FRAG] = "FS",
 453                 [QSTAGE_VERT] = "VS",
 454                 [QSTAGE_COORD] = "CS",
 455         };
 456
 457         return names[stage];
 458 }
 459
 460 struct qreg
 461 qir_uniform(struct vc4_compile *c,
 462             enum quniform_contents contents,
 463             uint32_t data)
 464 {
 465         for (int i = 0; i < c->num_uniforms; i++) {
 466                 if (c->uniform_contents[i] == contents &&
 467                     c->uniform_data[i] == data) {
 468                         return qir_reg(QFILE_UNIF, i);
 469                 }
 470         }
 471
 472         uint32_t uniform = c->num_uniforms++;
 473
 474         if (uniform >= c->uniform_array_size) {
 475                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 476                                              c->uniform_array_size * 2);
 477
 478                 c->uniform_data = reralloc(c, c->uniform_data,
 479                                            uint32_t,
 480                                            c->uniform_array_size);
 481                 c->uniform_contents = reralloc(c, c->uniform_contents,
 482                                                enum quniform_contents,
 483                                                c->uniform_array_size);
 484         }
 485
 486         c->uniform_contents[uniform] = contents;
 487         c->uniform_data[uniform] = data;
 488
 489         return qir_reg(QFILE_UNIF, uniform);
 490 }
 491
 492 void
 493 qir_SF(struct vc4_compile *c, struct qreg src)
 494 {
 495         struct qinst *last_inst = NULL;
 496         if (!list_empty(&c->instructions))
 497                 last_inst = (struct qinst *)c->instructions.prev;
 498
 499         /* We don't have any way to guess which kind of MOV is implied. */
 500         assert(!src.pack);
 501
 502         if (src.file != QFILE_TEMP ||
 503             !c->defs[src.index] ||
 504             last_inst != c->defs[src.index]) {
 505                 last_inst = qir_MOV_dest(c, qir_reg(QFILE_NULL, 0), src);
 506                 last_inst = (struct qinst *)c->instructions.prev;
 507         }
 508         last_inst->sf = true;
 509 }
 510
 511 #define OPTPASS(func)                                                   \
 512         do {                                                            \
 513                 bool stage_progress = func(c);                          \
 514                 if (stage_progress) {                                   \
 515                         progress = true;                                \
 516                         if (print_opt_debug) {                          \
 517                                 fprintf(stderr,                         \
 518                                         "QIR opt pass %2d: %s progress\n", \
 519                                         pass, #func);                   \
 520                         }                                               \
 521                 }                                                       \
 522         } while (0)
 523
 524 void
 525 qir_optimize(struct vc4_compile *c)
 526 {
 527         bool print_opt_debug = false;
 528         int pass = 1;
 529
 530         while (true) {
 531                 bool progress = false;
 532
 533                 OPTPASS(qir_opt_algebraic);
 534                 OPTPASS(qir_opt_constant_folding);
 535                 OPTPASS(qir_opt_copy_propagation);
 536                 OPTPASS(qir_opt_dead_code);
 537                 OPTPASS(qir_opt_small_immediates);
 538                 OPTPASS(qir_opt_vpm);
 539
 540                 if (!progress)
 541                         break;
 542
 543                 pass++;
 544         }
 545 }