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_X] = { "frag_x", 1, 0 },
  76         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  77         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  78         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  79         [QOP_FRAG_REV_FLAG] = { "frag_rev_flag", 1, 0 },
  80
  81         [QOP_TEX_S] = { "tex_s", 0, 2 },
  82         [QOP_TEX_T] = { "tex_t", 0, 2 },
  83         [QOP_TEX_R] = { "tex_r", 0, 2 },
  84         [QOP_TEX_B] = { "tex_b", 0, 2 },
  85         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
  86         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
  87 };
  88
  89 static const char *
  90 qir_get_op_name(enum qop qop)
  91 {
  92         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
  93                 return qir_op_info[qop].name;
  94         else
  95                 return "???";
  96 }
  97
  98 int
  99 qir_get_op_nsrc(enum qop qop)
 100 {
 101         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 102                 return qir_op_info[qop].nsrc;
 103         else
 104                 abort();
 105 }
 106
 107 /**
 108  * Returns whether the instruction has any side effects that must be
 109  * preserved.
 110  */
 111 bool
 112 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 113 {
 114         switch (inst->dst.file) {
 115         case QFILE_TLB_Z_WRITE:
 116         case QFILE_TLB_COLOR_WRITE:
 117         case QFILE_TLB_COLOR_WRITE_MS:
 118         case QFILE_TLB_STENCIL_SETUP:
 119                 return true;
 120         default:
 121                 break;
 122         }
 123
 124         return qir_op_info[inst->op].has_side_effects;
 125 }
 126
 127 bool
 128 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 129 {
 130         /* We can dead-code eliminate varyings, because we only tell the VS
 131          * about the live ones at the end.  But we have to preserve the
 132          * point/line coordinates reads, because they're generated by
 133          * fixed-function hardware.
 134          */
 135         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 136                 if (inst->src[i].file == QFILE_VARY &&
 137                     c->input_slots[inst->src[i].index].slot == 0xff) {
 138                         return true;
 139                 }
 140
 141                 if (inst->src[i].file == QFILE_VPM)
 142                         return true;
 143         }
 144
 145         if (inst->dst.file == QFILE_VPM)
 146                 return true;
 147
 148         return false;
 149 }
 150
 151 bool
 152 qir_is_mul(struct qinst *inst)
 153 {
 154         switch (inst->op) {
 155         case QOP_MMOV:
 156         case QOP_FMUL:
 157         case QOP_MUL24:
 158         case QOP_V8MULD:
 159         case QOP_V8MIN:
 160         case QOP_V8MAX:
 161         case QOP_V8ADDS:
 162         case QOP_V8SUBS:
 163                 return true;
 164         default:
 165                 return false;
 166         }
 167 }
 168
 169 bool
 170 qir_is_float_input(struct qinst *inst)
 171 {
 172         switch (inst->op) {
 173         case QOP_FMOV:
 174         case QOP_FMUL:
 175         case QOP_FADD:
 176         case QOP_FSUB:
 177         case QOP_FMIN:
 178         case QOP_FMAX:
 179         case QOP_FMINABS:
 180         case QOP_FMAXABS:
 181         case QOP_FTOI:
 182                 return true;
 183         default:
 184                 return false;
 185         }
 186 }
 187
 188 bool
 189 qir_is_raw_mov(struct qinst *inst)
 190 {
 191         return ((inst->op == QOP_MOV ||
 192                  inst->op == QOP_FMOV ||
 193                  inst->op == QOP_MMOV) &&
 194                 inst->cond == QPU_COND_ALWAYS &&
 195                 !inst->dst.pack &&
 196                 !inst->src[0].pack);
 197 }
 198
 199 bool
 200 qir_is_tex(struct qinst *inst)
 201 {
 202         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 203 }
 204
 205 bool
 206 qir_depends_on_flags(struct qinst *inst)
 207 {
 208         return (inst->cond != QPU_COND_ALWAYS &&
 209                 inst->cond != QPU_COND_NEVER);
 210 }
 211
 212 bool
 213 qir_writes_r4(struct qinst *inst)
 214 {
 215         switch (inst->op) {
 216         case QOP_TEX_RESULT:
 217         case QOP_TLB_COLOR_READ:
 218         case QOP_RCP:
 219         case QOP_RSQ:
 220         case QOP_EXP2:
 221         case QOP_LOG2:
 222                 return true;
 223         default:
 224                 return false;
 225         }
 226 }
 227
 228 static void
 229 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 230 {
 231         static const char *files[] = {
 232                 [QFILE_TEMP] = "t",
 233                 [QFILE_VARY] = "v",
 234                 [QFILE_UNIF] = "u",
 235                 [QFILE_TLB_COLOR_WRITE] = "tlb_c",
 236                 [QFILE_TLB_COLOR_WRITE_MS] = "tlb_c_ms",
 237                 [QFILE_TLB_Z_WRITE] = "tlb_z",
 238                 [QFILE_TLB_STENCIL_SETUP] = "tlb_stencil",
 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         static const char *conditions[] = {
 287                 [QPU_COND_ALWAYS] = "",
 288                 [QPU_COND_NEVER] = ".never",
 289                 [QPU_COND_ZS] = ".zs",
 290                 [QPU_COND_ZC] = ".zc",
 291                 [QPU_COND_NS] = ".ns",
 292                 [QPU_COND_NC] = ".nc",
 293                 [QPU_COND_CS] = ".cs",
 294                 [QPU_COND_CC] = ".cc",
 295         };
 296         fprintf(stderr, "%s%s%s ",
 297                 qir_get_op_name(inst->op),
 298                 conditions[inst->cond],
 299                 inst->sf ? ".sf" : "");
 300
 301         qir_print_reg(c, inst->dst, true);
 302         if (inst->dst.pack) {
 303                 if (inst->dst.pack) {
 304                         if (qir_is_mul(inst))
 305                                 vc4_qpu_disasm_pack_mul(stderr, inst->dst.pack);
 306                         else
 307                                 vc4_qpu_disasm_pack_a(stderr, inst->dst.pack);
 308                 }
 309         }
 310         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 311                 fprintf(stderr, ", ");
 312                 qir_print_reg(c, inst->src[i], false);
 313                 vc4_qpu_disasm_unpack(stderr, inst->src[i].pack);
 314         }
 315 }
 316
 317 void
 318 qir_dump(struct vc4_compile *c)
 319 {
 320         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 321                 qir_dump_inst(c, inst);
 322                 fprintf(stderr, "\n");
 323         }
 324 }
 325
 326 struct qreg
 327 qir_get_temp(struct vc4_compile *c)
 328 {
 329         struct qreg reg;
 330
 331         reg.file = QFILE_TEMP;
 332         reg.index = c->num_temps++;
 333         reg.pack = 0;
 334
 335         if (c->num_temps > c->defs_array_size) {
 336                 uint32_t old_size = c->defs_array_size;
 337                 c->defs_array_size = MAX2(old_size * 2, 16);
 338                 c->defs = reralloc(c, c->defs, struct qinst *,
 339                                    c->defs_array_size);
 340                 memset(&c->defs[old_size], 0,
 341                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 342         }
 343
 344         return reg;
 345 }
 346
 347 struct qinst *
 348 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 349 {
 350         struct qinst *inst = CALLOC_STRUCT(qinst);
 351
 352         inst->op = op;
 353         inst->dst = dst;
 354         inst->src = calloc(2, sizeof(inst->src[0]));
 355         inst->src[0] = src0;
 356         inst->src[1] = src1;
 357         inst->cond = QPU_COND_ALWAYS;
 358
 359         return inst;
 360 }
 361
 362 struct qinst *
 363 qir_inst4(enum qop op, struct qreg dst,
 364           struct qreg a,
 365           struct qreg b,
 366           struct qreg c,
 367           struct qreg d)
 368 {
 369         struct qinst *inst = CALLOC_STRUCT(qinst);
 370
 371         inst->op = op;
 372         inst->dst = dst;
 373         inst->src = calloc(4, sizeof(*inst->src));
 374         inst->src[0] = a;
 375         inst->src[1] = b;
 376         inst->src[2] = c;
 377         inst->src[3] = d;
 378
 379         return inst;
 380 }
 381
 382 void
 383 qir_emit(struct vc4_compile *c, struct qinst *inst)
 384 {
 385         if (inst->dst.file == QFILE_TEMP)
 386                 c->defs[inst->dst.index] = inst;
 387
 388         qir_emit_nodef(c, inst);
 389 }
 390
 391 bool
 392 qir_reg_equals(struct qreg a, struct qreg b)
 393 {
 394         return a.file == b.file && a.index == b.index;
 395 }
 396
 397 struct vc4_compile *
 398 qir_compile_init(void)
 399 {
 400         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 401
 402         list_inithead(&c->instructions);
 403
 404         c->output_position_index = -1;
 405         c->output_color_index = -1;
 406         c->output_point_size_index = -1;
 407         c->output_sample_mask_index = -1;
 408
 409         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 410                                             _mesa_key_pointer_equal);
 411
 412         return c;
 413 }
 414
 415 void
 416 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 417 {
 418         if (qinst->dst.file == QFILE_TEMP)
 419                 c->defs[qinst->dst.index] = NULL;
 420
 421         list_del(&qinst->link);
 422         free(qinst->src);
 423         free(qinst);
 424 }
 425
 426 struct qreg
 427 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 428 {
 429         int pack = reg.pack;
 430
 431         while (reg.file == QFILE_TEMP &&
 432                c->defs[reg.index] &&
 433                (c->defs[reg.index]->op == QOP_MOV ||
 434                 c->defs[reg.index]->op == QOP_FMOV ||
 435                 c->defs[reg.index]->op == QOP_MMOV)&&
 436                !c->defs[reg.index]->dst.pack &&
 437                !c->defs[reg.index]->src[0].pack) {
 438                 reg = c->defs[reg.index]->src[0];
 439         }
 440
 441         reg.pack = pack;
 442         return reg;
 443 }
 444
 445 void
 446 qir_compile_destroy(struct vc4_compile *c)
 447 {
 448         while (!list_empty(&c->instructions)) {
 449                 struct qinst *qinst =
 450                         (struct qinst *)c->instructions.next;
 451                 qir_remove_instruction(c, qinst);
 452         }
 453
 454         ralloc_free(c);
 455 }
 456
 457 const char *
 458 qir_get_stage_name(enum qstage stage)
 459 {
 460         static const char *names[] = {
 461                 [QSTAGE_FRAG] = "FS",
 462                 [QSTAGE_VERT] = "VS",
 463                 [QSTAGE_COORD] = "CS",
 464         };
 465
 466         return names[stage];
 467 }
 468
 469 struct qreg
 470 qir_uniform(struct vc4_compile *c,
 471             enum quniform_contents contents,
 472             uint32_t data)
 473 {
 474         for (int i = 0; i < c->num_uniforms; i++) {
 475                 if (c->uniform_contents[i] == contents &&
 476                     c->uniform_data[i] == data) {
 477                         return qir_reg(QFILE_UNIF, i);
 478                 }
 479         }
 480
 481         uint32_t uniform = c->num_uniforms++;
 482
 483         if (uniform >= c->uniform_array_size) {
 484                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 485                                              c->uniform_array_size * 2);
 486
 487                 c->uniform_data = reralloc(c, c->uniform_data,
 488                                            uint32_t,
 489                                            c->uniform_array_size);
 490                 c->uniform_contents = reralloc(c, c->uniform_contents,
 491                                                enum quniform_contents,
 492                                                c->uniform_array_size);
 493         }
 494
 495         c->uniform_contents[uniform] = contents;
 496         c->uniform_data[uniform] = data;
 497
 498         return qir_reg(QFILE_UNIF, uniform);
 499 }
 500
 501 void
 502 qir_SF(struct vc4_compile *c, struct qreg src)
 503 {
 504         struct qinst *last_inst = NULL;
 505         if (!list_empty(&c->instructions))
 506                 last_inst = (struct qinst *)c->instructions.prev;
 507
 508         /* We don't have any way to guess which kind of MOV is implied. */
 509         assert(!src.pack);
 510
 511         if (src.file != QFILE_TEMP ||
 512             !c->defs[src.index] ||
 513             last_inst != c->defs[src.index]) {
 514                 last_inst = qir_MOV_dest(c, qir_reg(QFILE_NULL, 0), src);
 515                 last_inst = (struct qinst *)c->instructions.prev;
 516         }
 517         last_inst->sf = true;
 518 }
 519
 520 #define OPTPASS(func)                                                   \
 521         do {                                                            \
 522                 bool stage_progress = func(c);                          \
 523                 if (stage_progress) {                                   \
 524                         progress = true;                                \
 525                         if (print_opt_debug) {                          \
 526                                 fprintf(stderr,                         \
 527                                         "QIR opt pass %2d: %s progress\n", \
 528                                         pass, #func);                   \
 529                         }                                               \
 530                 }                                                       \
 531         } while (0)
 532
 533 void
 534 qir_optimize(struct vc4_compile *c)
 535 {
 536         bool print_opt_debug = false;
 537         int pass = 1;
 538
 539         while (true) {
 540                 bool progress = false;
 541
 542                 OPTPASS(qir_opt_algebraic);
 543                 OPTPASS(qir_opt_cse);
 544                 OPTPASS(qir_opt_constant_folding);
 545                 OPTPASS(qir_opt_copy_propagation);
 546                 OPTPASS(qir_opt_dead_code);
 547                 OPTPASS(qir_opt_small_immediates);
 548                 OPTPASS(qir_opt_vpm);
 549
 550                 if (!progress)
 551                         break;
 552
 553                 pass++;
 554         }
 555 }