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         bool multi_instruction;
  35 };
  36
  37 static const struct qir_op_info qir_op_info[] = {
  38         [QOP_MOV] = { "mov", 1, 1 },
  39         [QOP_FMOV] = { "fmov", 1, 1 },
  40         [QOP_MMOV] = { "mmov", 1, 1 },
  41         [QOP_FADD] = { "fadd", 1, 2 },
  42         [QOP_FSUB] = { "fsub", 1, 2 },
  43         [QOP_FMUL] = { "fmul", 1, 2 },
  44         [QOP_MUL24] = { "mul24", 1, 2 },
  45         [QOP_V8MULD] = {"v8muld", 1, 2 },
  46         [QOP_V8MIN] = {"v8min", 1, 2 },
  47         [QOP_V8MAX] = {"v8max", 1, 2 },
  48         [QOP_V8ADDS] = {"v8adds", 1, 2 },
  49         [QOP_V8SUBS] = {"v8subs", 1, 2 },
  50         [QOP_FMIN] = { "fmin", 1, 2 },
  51         [QOP_FMAX] = { "fmax", 1, 2 },
  52         [QOP_FMINABS] = { "fminabs", 1, 2 },
  53         [QOP_FMAXABS] = { "fmaxabs", 1, 2 },
  54         [QOP_FTOI] = { "ftoi", 1, 1 },
  55         [QOP_ITOF] = { "itof", 1, 1 },
  56         [QOP_ADD] = { "add", 1, 2 },
  57         [QOP_SUB] = { "sub", 1, 2 },
  58         [QOP_SHR] = { "shr", 1, 2 },
  59         [QOP_ASR] = { "asr", 1, 2 },
  60         [QOP_SHL] = { "shl", 1, 2 },
  61         [QOP_MIN] = { "min", 1, 2 },
  62         [QOP_MAX] = { "max", 1, 2 },
  63         [QOP_AND] = { "and", 1, 2 },
  64         [QOP_OR] = { "or", 1, 2 },
  65         [QOP_XOR] = { "xor", 1, 2 },
  66         [QOP_NOT] = { "not", 1, 1 },
  67
  68         [QOP_RCP] = { "rcp", 1, 1, false, true },
  69         [QOP_RSQ] = { "rsq", 1, 1, false, true },
  70         [QOP_EXP2] = { "exp2", 1, 2, false, true },
  71         [QOP_LOG2] = { "log2", 1, 2, false, true },
  72         [QOP_TLB_DISCARD_SETUP] = { "discard", 0, 1, true },
  73         [QOP_TLB_STENCIL_SETUP] = { "tlb_stencil_setup", 0, 1, true },
  74         [QOP_TLB_Z_WRITE] = { "tlb_z", 0, 1, true },
  75         [QOP_TLB_COLOR_WRITE] = { "tlb_color", 0, 1, true },
  76         [QOP_TLB_COLOR_WRITE_MS] = { "tlb_color_ms", 0, 1, true },
  77         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0 },
  78         [QOP_MS_MASK] = { "ms_mask", 0, 1, true },
  79         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  80
  81         [QOP_FRAG_X] = { "frag_x", 1, 0 },
  82         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  83         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  84         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  85         [QOP_FRAG_REV_FLAG] = { "frag_rev_flag", 1, 0 },
  86
  87         [QOP_TEX_S] = { "tex_s", 0, 2 },
  88         [QOP_TEX_T] = { "tex_t", 0, 2 },
  89         [QOP_TEX_R] = { "tex_r", 0, 2 },
  90         [QOP_TEX_B] = { "tex_b", 0, 2 },
  91         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
  92         [QOP_TEX_RESULT] = { "tex_result", 1, 0, 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         return qir_op_info[inst->op].has_side_effects;
 121 }
 122
 123 bool
 124 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 125 {
 126         /* We can dead-code eliminate varyings, because we only tell the VS
 127          * about the live ones at the end.  But we have to preserve the
 128          * point/line coordinates reads, because they're generated by
 129          * fixed-function hardware.
 130          */
 131         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 132                 if (inst->src[i].file == QFILE_VARY &&
 133                     c->input_slots[inst->src[i].index].slot == 0xff) {
 134                         return true;
 135                 }
 136
 137                 if (inst->src[i].file == QFILE_VPM)
 138                         return true;
 139         }
 140
 141         if (inst->dst.file == QFILE_VPM)
 142                 return true;
 143
 144         return false;
 145 }
 146
 147 bool
 148 qir_is_multi_instruction(struct qinst *inst)
 149 {
 150         return qir_op_info[inst->op].multi_instruction;
 151 }
 152
 153 bool
 154 qir_is_mul(struct qinst *inst)
 155 {
 156         switch (inst->op) {
 157         case QOP_MMOV:
 158         case QOP_FMUL:
 159         case QOP_MUL24:
 160         case QOP_V8MULD:
 161         case QOP_V8MIN:
 162         case QOP_V8MAX:
 163         case QOP_V8ADDS:
 164         case QOP_V8SUBS:
 165                 return true;
 166         default:
 167                 return false;
 168         }
 169 }
 170
 171 bool
 172 qir_is_float_input(struct qinst *inst)
 173 {
 174         switch (inst->op) {
 175         case QOP_FMOV:
 176         case QOP_FMUL:
 177         case QOP_FADD:
 178         case QOP_FSUB:
 179         case QOP_FMIN:
 180         case QOP_FMAX:
 181         case QOP_FMINABS:
 182         case QOP_FMAXABS:
 183         case QOP_FTOI:
 184                 return true;
 185         default:
 186                 return false;
 187         }
 188 }
 189
 190 bool
 191 qir_is_raw_mov(struct qinst *inst)
 192 {
 193         return ((inst->op == QOP_MOV ||
 194                  inst->op == QOP_FMOV ||
 195                  inst->op == QOP_MMOV) &&
 196                 !inst->dst.pack &&
 197                 !inst->src[0].pack);
 198 }
 199
 200 bool
 201 qir_is_tex(struct qinst *inst)
 202 {
 203         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 204 }
 205
 206 bool
 207 qir_depends_on_flags(struct qinst *inst)
 208 {
 209         return (inst->cond != QPU_COND_ALWAYS &&
 210                 inst->cond != QPU_COND_NEVER);
 211 }
 212
 213 bool
 214 qir_writes_r4(struct qinst *inst)
 215 {
 216         switch (inst->op) {
 217         case QOP_TEX_RESULT:
 218         case QOP_TLB_COLOR_READ:
 219         case QOP_RCP:
 220         case QOP_RSQ:
 221         case QOP_EXP2:
 222         case QOP_LOG2:
 223                 return true;
 224         default:
 225                 return false;
 226         }
 227 }
 228
 229 static void
 230 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 231 {
 232         static const char *files[] = {
 233                 [QFILE_TEMP] = "t",
 234                 [QFILE_VARY] = "v",
 235                 [QFILE_UNIF] = "u",
 236         };
 237
 238         if (reg.file == QFILE_NULL) {
 239                 fprintf(stderr, "null");
 240         } else if (reg.file == QFILE_SMALL_IMM) {
 241                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 242                         fprintf(stderr, "%d", reg.index);
 243                 else
 244                         fprintf(stderr, "%f", uif(reg.index));
 245         } else if (reg.file == QFILE_VPM) {
 246                 if (write) {
 247                         fprintf(stderr, "vpm");
 248                 } else {
 249                         fprintf(stderr, "vpm%d.%d",
 250                                 reg.index / 4, reg.index % 4);
 251                 }
 252         } else {
 253                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 254         }
 255
 256         if (reg.file == QFILE_UNIF &&
 257             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 258                 fprintf(stderr, " (0x%08x / %f)",
 259                         c->uniform_data[reg.index],
 260                         uif(c->uniform_data[reg.index]));
 261         }
 262 }
 263
 264 void
 265 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 266 {
 267         static const char *conditions[] = {
 268                 [QPU_COND_ALWAYS] = "",
 269                 [QPU_COND_NEVER] = ".never",
 270                 [QPU_COND_ZS] = ".zs",
 271                 [QPU_COND_ZC] = ".zc",
 272                 [QPU_COND_NS] = ".ns",
 273                 [QPU_COND_NC] = ".nc",
 274                 [QPU_COND_CS] = ".cs",
 275                 [QPU_COND_CC] = ".cc",
 276         };
 277         fprintf(stderr, "%s%s%s ",
 278                 qir_get_op_name(inst->op),
 279                 conditions[inst->cond],
 280                 inst->sf ? ".sf" : "");
 281
 282         qir_print_reg(c, inst->dst, true);
 283         if (inst->dst.pack) {
 284                 if (inst->dst.pack) {
 285                         if (qir_is_mul(inst))
 286                                 vc4_qpu_disasm_pack_mul(stderr, inst->dst.pack);
 287                         else
 288                                 vc4_qpu_disasm_pack_a(stderr, inst->dst.pack);
 289                 }
 290         }
 291         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 292                 fprintf(stderr, ", ");
 293                 qir_print_reg(c, inst->src[i], false);
 294                 vc4_qpu_disasm_unpack(stderr, inst->src[i].pack);
 295         }
 296 }
 297
 298 void
 299 qir_dump(struct vc4_compile *c)
 300 {
 301         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 302                 qir_dump_inst(c, inst);
 303                 fprintf(stderr, "\n");
 304         }
 305 }
 306
 307 struct qreg
 308 qir_get_temp(struct vc4_compile *c)
 309 {
 310         struct qreg reg;
 311
 312         reg.file = QFILE_TEMP;
 313         reg.index = c->num_temps++;
 314         reg.pack = 0;
 315
 316         if (c->num_temps > c->defs_array_size) {
 317                 uint32_t old_size = c->defs_array_size;
 318                 c->defs_array_size = MAX2(old_size * 2, 16);
 319                 c->defs = reralloc(c, c->defs, struct qinst *,
 320                                    c->defs_array_size);
 321                 memset(&c->defs[old_size], 0,
 322                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 323         }
 324
 325         return reg;
 326 }
 327
 328 struct qinst *
 329 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 330 {
 331         struct qinst *inst = CALLOC_STRUCT(qinst);
 332
 333         inst->op = op;
 334         inst->dst = dst;
 335         inst->src = calloc(2, sizeof(inst->src[0]));
 336         inst->src[0] = src0;
 337         inst->src[1] = src1;
 338         inst->cond = QPU_COND_ALWAYS;
 339
 340         return inst;
 341 }
 342
 343 struct qinst *
 344 qir_inst4(enum qop op, struct qreg dst,
 345           struct qreg a,
 346           struct qreg b,
 347           struct qreg c,
 348           struct qreg d)
 349 {
 350         struct qinst *inst = CALLOC_STRUCT(qinst);
 351
 352         inst->op = op;
 353         inst->dst = dst;
 354         inst->src = calloc(4, sizeof(*inst->src));
 355         inst->src[0] = a;
 356         inst->src[1] = b;
 357         inst->src[2] = c;
 358         inst->src[3] = d;
 359
 360         return inst;
 361 }
 362
 363 void
 364 qir_emit(struct vc4_compile *c, struct qinst *inst)
 365 {
 366         if (inst->dst.file == QFILE_TEMP)
 367                 c->defs[inst->dst.index] = inst;
 368
 369         qir_emit_nodef(c, inst);
 370 }
 371
 372 bool
 373 qir_reg_equals(struct qreg a, struct qreg b)
 374 {
 375         return a.file == b.file && a.index == b.index;
 376 }
 377
 378 struct vc4_compile *
 379 qir_compile_init(void)
 380 {
 381         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 382
 383         list_inithead(&c->instructions);
 384
 385         c->output_position_index = -1;
 386         c->output_color_index = -1;
 387         c->output_point_size_index = -1;
 388         c->output_sample_mask_index = -1;
 389
 390         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 391                                             _mesa_key_pointer_equal);
 392
 393         return c;
 394 }
 395
 396 void
 397 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 398 {
 399         if (qinst->dst.file == QFILE_TEMP)
 400                 c->defs[qinst->dst.index] = NULL;
 401
 402         list_del(&qinst->link);
 403         free(qinst->src);
 404         free(qinst);
 405 }
 406
 407 struct qreg
 408 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 409 {
 410         int pack = reg.pack;
 411
 412         while (reg.file == QFILE_TEMP &&
 413                c->defs[reg.index] &&
 414                (c->defs[reg.index]->op == QOP_MOV ||
 415                 c->defs[reg.index]->op == QOP_FMOV ||
 416                 c->defs[reg.index]->op == QOP_MMOV)&&
 417                !c->defs[reg.index]->dst.pack &&
 418                !c->defs[reg.index]->src[0].pack) {
 419                 reg = c->defs[reg.index]->src[0];
 420         }
 421
 422         reg.pack = pack;
 423         return reg;
 424 }
 425
 426 void
 427 qir_compile_destroy(struct vc4_compile *c)
 428 {
 429         while (!list_empty(&c->instructions)) {
 430                 struct qinst *qinst =
 431                         (struct qinst *)c->instructions.next;
 432                 qir_remove_instruction(c, qinst);
 433         }
 434
 435         ralloc_free(c);
 436 }
 437
 438 const char *
 439 qir_get_stage_name(enum qstage stage)
 440 {
 441         static const char *names[] = {
 442                 [QSTAGE_FRAG] = "FS",
 443                 [QSTAGE_VERT] = "VS",
 444                 [QSTAGE_COORD] = "CS",
 445         };
 446
 447         return names[stage];
 448 }
 449
 450 struct qreg
 451 qir_uniform(struct vc4_compile *c,
 452             enum quniform_contents contents,
 453             uint32_t data)
 454 {
 455         for (int i = 0; i < c->num_uniforms; i++) {
 456                 if (c->uniform_contents[i] == contents &&
 457                     c->uniform_data[i] == data) {
 458                         return (struct qreg) { QFILE_UNIF, i };
 459                 }
 460         }
 461
 462         uint32_t uniform = c->num_uniforms++;
 463         struct qreg u = { QFILE_UNIF, uniform };
 464
 465         if (uniform >= c->uniform_array_size) {
 466                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 467                                              c->uniform_array_size * 2);
 468
 469                 c->uniform_data = reralloc(c, c->uniform_data,
 470                                            uint32_t,
 471                                            c->uniform_array_size);
 472                 c->uniform_contents = reralloc(c, c->uniform_contents,
 473                                                enum quniform_contents,
 474                                                c->uniform_array_size);
 475         }
 476
 477         c->uniform_contents[uniform] = contents;
 478         c->uniform_data[uniform] = data;
 479
 480         return u;
 481 }
 482
 483 void
 484 qir_SF(struct vc4_compile *c, struct qreg src)
 485 {
 486         struct qinst *last_inst = NULL;
 487         if (!list_empty(&c->instructions))
 488                 last_inst = (struct qinst *)c->instructions.prev;
 489
 490         if (src.file != QFILE_TEMP ||
 491             !c->defs[src.index] ||
 492             last_inst != c->defs[src.index] ||
 493             qir_is_multi_instruction(last_inst)) {
 494                 src = qir_MOV(c, src);
 495                 last_inst = (struct qinst *)c->instructions.prev;
 496         }
 497         last_inst->sf = true;
 498 }
 499
 500 #define OPTPASS(func)                                                   \
 501         do {                                                            \
 502                 bool stage_progress = func(c);                          \
 503                 if (stage_progress) {                                   \
 504                         progress = true;                                \
 505                         if (print_opt_debug) {                          \
 506                                 fprintf(stderr,                         \
 507                                         "QIR opt pass %2d: %s progress\n", \
 508                                         pass, #func);                   \
 509                         }                                               \
 510                 }                                                       \
 511         } while (0)
 512
 513 void
 514 qir_optimize(struct vc4_compile *c)
 515 {
 516         bool print_opt_debug = false;
 517         int pass = 1;
 518
 519         while (true) {
 520                 bool progress = false;
 521
 522                 OPTPASS(qir_opt_algebraic);
 523                 OPTPASS(qir_opt_cse);
 524                 OPTPASS(qir_opt_constant_folding);
 525                 OPTPASS(qir_opt_copy_propagation);
 526                 OPTPASS(qir_opt_dead_code);
 527                 OPTPASS(qir_opt_small_immediates);
 528                 OPTPASS(qir_opt_vpm_writes);
 529
 530                 if (!progress)
 531                         break;
 532
 533                 pass++;
 534         }
 535 }