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_SEL_X_0_NS] = { "fsel_x_0_ns", 1, 1, false, true },
  69         [QOP_SEL_X_0_NC] = { "fsel_x_0_nc", 1, 1, false, true },
  70         [QOP_SEL_X_0_ZS] = { "fsel_x_0_zs", 1, 1, false, true },
  71         [QOP_SEL_X_0_ZC] = { "fsel_x_0_zc", 1, 1, false, true },
  72         [QOP_SEL_X_0_CS] = { "fsel_x_0_cs", 1, 1, false, true },
  73         [QOP_SEL_X_0_CC] = { "fsel_x_0_cc", 1, 1, false, true },
  74         [QOP_SEL_X_Y_NS] = { "fsel_x_y_ns", 1, 2, false, true },
  75         [QOP_SEL_X_Y_NC] = { "fsel_x_y_nc", 1, 2, false, true },
  76         [QOP_SEL_X_Y_ZS] = { "fsel_x_y_zs", 1, 2, false, true },
  77         [QOP_SEL_X_Y_ZC] = { "fsel_x_y_zc", 1, 2, false, true },
  78         [QOP_SEL_X_Y_CS] = { "fsel_x_y_cs", 1, 2, false, true },
  79         [QOP_SEL_X_Y_CC] = { "fsel_x_y_cc", 1, 2, false, true },
  80
  81         [QOP_RCP] = { "rcp", 1, 1, false, true },
  82         [QOP_RSQ] = { "rsq", 1, 1, false, true },
  83         [QOP_EXP2] = { "exp2", 1, 2, false, true },
  84         [QOP_LOG2] = { "log2", 1, 2, false, true },
  85         [QOP_TLB_DISCARD_SETUP] = { "discard", 0, 1, true },
  86         [QOP_TLB_STENCIL_SETUP] = { "tlb_stencil_setup", 0, 1, true },
  87         [QOP_TLB_Z_WRITE] = { "tlb_z", 0, 1, true },
  88         [QOP_TLB_COLOR_WRITE] = { "tlb_color", 0, 1, true },
  89         [QOP_TLB_COLOR_WRITE_MS] = { "tlb_color_ms", 0, 1, true },
  90         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0 },
  91         [QOP_MS_MASK] = { "ms_mask", 0, 1, true },
  92         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  93
  94         [QOP_FRAG_X] = { "frag_x", 1, 0 },
  95         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  96         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  97         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  98         [QOP_FRAG_REV_FLAG] = { "frag_rev_flag", 1, 0 },
  99
 100         [QOP_TEX_S] = { "tex_s", 0, 2 },
 101         [QOP_TEX_T] = { "tex_t", 0, 2 },
 102         [QOP_TEX_R] = { "tex_r", 0, 2 },
 103         [QOP_TEX_B] = { "tex_b", 0, 2 },
 104         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
 105         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
 106 };
 107
 108 static const char *
 109 qir_get_op_name(enum qop qop)
 110 {
 111         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 112                 return qir_op_info[qop].name;
 113         else
 114                 return "???";
 115 }
 116
 117 int
 118 qir_get_op_nsrc(enum qop qop)
 119 {
 120         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 121                 return qir_op_info[qop].nsrc;
 122         else
 123                 abort();
 124 }
 125
 126 /**
 127  * Returns whether the instruction has any side effects that must be
 128  * preserved.
 129  */
 130 bool
 131 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 132 {
 133         return qir_op_info[inst->op].has_side_effects;
 134 }
 135
 136 bool
 137 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 138 {
 139         /* We can dead-code eliminate varyings, because we only tell the VS
 140          * about the live ones at the end.  But we have to preserve the
 141          * point/line coordinates reads, because they're generated by
 142          * fixed-function hardware.
 143          */
 144         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 145                 if (inst->src[i].file == QFILE_VARY &&
 146                     c->input_slots[inst->src[i].index].slot == 0xff) {
 147                         return true;
 148                 }
 149
 150                 if (inst->src[i].file == QFILE_VPM)
 151                         return true;
 152         }
 153
 154         if (inst->dst.file == QFILE_VPM)
 155                 return true;
 156
 157         return false;
 158 }
 159
 160 bool
 161 qir_is_multi_instruction(struct qinst *inst)
 162 {
 163         return qir_op_info[inst->op].multi_instruction;
 164 }
 165
 166 bool
 167 qir_is_mul(struct qinst *inst)
 168 {
 169         switch (inst->op) {
 170         case QOP_MMOV:
 171         case QOP_FMUL:
 172         case QOP_MUL24:
 173         case QOP_V8MULD:
 174         case QOP_V8MIN:
 175         case QOP_V8MAX:
 176         case QOP_V8ADDS:
 177         case QOP_V8SUBS:
 178                 return true;
 179         default:
 180                 return false;
 181         }
 182 }
 183
 184 bool
 185 qir_is_float_input(struct qinst *inst)
 186 {
 187         switch (inst->op) {
 188         case QOP_FMOV:
 189         case QOP_FMUL:
 190         case QOP_FADD:
 191         case QOP_FSUB:
 192         case QOP_FMIN:
 193         case QOP_FMAX:
 194         case QOP_FMINABS:
 195         case QOP_FMAXABS:
 196         case QOP_FTOI:
 197                 return true;
 198         default:
 199                 return false;
 200         }
 201 }
 202
 203 bool
 204 qir_is_raw_mov(struct qinst *inst)
 205 {
 206         return ((inst->op == QOP_MOV ||
 207                  inst->op == QOP_FMOV ||
 208                  inst->op == QOP_MMOV) &&
 209                 !inst->dst.pack &&
 210                 !inst->src[0].pack);
 211 }
 212
 213 bool
 214 qir_is_tex(struct qinst *inst)
 215 {
 216         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 217 }
 218
 219 bool
 220 qir_depends_on_flags(struct qinst *inst)
 221 {
 222         switch (inst->op) {
 223         case QOP_SEL_X_0_NS:
 224         case QOP_SEL_X_0_NC:
 225         case QOP_SEL_X_0_ZS:
 226         case QOP_SEL_X_0_ZC:
 227         case QOP_SEL_X_0_CS:
 228         case QOP_SEL_X_0_CC:
 229         case QOP_SEL_X_Y_NS:
 230         case QOP_SEL_X_Y_NC:
 231         case QOP_SEL_X_Y_ZS:
 232         case QOP_SEL_X_Y_ZC:
 233         case QOP_SEL_X_Y_CS:
 234         case QOP_SEL_X_Y_CC:
 235                 return true;
 236         default:
 237                 return false;
 238         }
 239 }
 240
 241 bool
 242 qir_writes_r4(struct qinst *inst)
 243 {
 244         switch (inst->op) {
 245         case QOP_TEX_RESULT:
 246         case QOP_TLB_COLOR_READ:
 247         case QOP_RCP:
 248         case QOP_RSQ:
 249         case QOP_EXP2:
 250         case QOP_LOG2:
 251                 return true;
 252         default:
 253                 return false;
 254         }
 255 }
 256
 257 static void
 258 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 259 {
 260         static const char *files[] = {
 261                 [QFILE_TEMP] = "t",
 262                 [QFILE_VARY] = "v",
 263                 [QFILE_UNIF] = "u",
 264         };
 265
 266         if (reg.file == QFILE_NULL) {
 267                 fprintf(stderr, "null");
 268         } else if (reg.file == QFILE_SMALL_IMM) {
 269                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 270                         fprintf(stderr, "%d", reg.index);
 271                 else
 272                         fprintf(stderr, "%f", uif(reg.index));
 273         } else if (reg.file == QFILE_VPM) {
 274                 if (write) {
 275                         fprintf(stderr, "vpm");
 276                 } else {
 277                         fprintf(stderr, "vpm%d.%d",
 278                                 reg.index / 4, reg.index % 4);
 279                 }
 280         } else {
 281                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 282         }
 283
 284         if (reg.file == QFILE_UNIF &&
 285             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 286                 fprintf(stderr, " (0x%08x / %f)",
 287                         c->uniform_data[reg.index],
 288                         uif(c->uniform_data[reg.index]));
 289         }
 290 }
 291
 292 void
 293 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 294 {
 295         fprintf(stderr, "%s%s ",
 296                 qir_get_op_name(inst->op),
 297                 inst->sf ? ".sf" : "");
 298
 299         qir_print_reg(c, inst->dst, true);
 300         if (inst->dst.pack) {
 301                 if (inst->dst.pack) {
 302                         if (qir_is_mul(inst))
 303                                 vc4_qpu_disasm_pack_mul(stderr, inst->dst.pack);
 304                         else
 305                                 vc4_qpu_disasm_pack_a(stderr, inst->dst.pack);
 306                 }
 307         }
 308         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 309                 fprintf(stderr, ", ");
 310                 qir_print_reg(c, inst->src[i], false);
 311                 vc4_qpu_disasm_unpack(stderr, inst->src[i].pack);
 312         }
 313 }
 314
 315 void
 316 qir_dump(struct vc4_compile *c)
 317 {
 318         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 319                 qir_dump_inst(c, inst);
 320                 fprintf(stderr, "\n");
 321         }
 322 }
 323
 324 struct qreg
 325 qir_get_temp(struct vc4_compile *c)
 326 {
 327         struct qreg reg;
 328
 329         reg.file = QFILE_TEMP;
 330         reg.index = c->num_temps++;
 331         reg.pack = 0;
 332
 333         if (c->num_temps > c->defs_array_size) {
 334                 uint32_t old_size = c->defs_array_size;
 335                 c->defs_array_size = MAX2(old_size * 2, 16);
 336                 c->defs = reralloc(c, c->defs, struct qinst *,
 337                                    c->defs_array_size);
 338                 memset(&c->defs[old_size], 0,
 339                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 340         }
 341
 342         return reg;
 343 }
 344
 345 struct qinst *
 346 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 347 {
 348         struct qinst *inst = CALLOC_STRUCT(qinst);
 349
 350         inst->op = op;
 351         inst->dst = dst;
 352         inst->src = calloc(2, sizeof(inst->src[0]));
 353         inst->src[0] = src0;
 354         inst->src[1] = src1;
 355
 356         return inst;
 357 }
 358
 359 struct qinst *
 360 qir_inst4(enum qop op, struct qreg dst,
 361           struct qreg a,
 362           struct qreg b,
 363           struct qreg c,
 364           struct qreg d)
 365 {
 366         struct qinst *inst = CALLOC_STRUCT(qinst);
 367
 368         inst->op = op;
 369         inst->dst = dst;
 370         inst->src = calloc(4, sizeof(*inst->src));
 371         inst->src[0] = a;
 372         inst->src[1] = b;
 373         inst->src[2] = c;
 374         inst->src[3] = d;
 375
 376         return inst;
 377 }
 378
 379 void
 380 qir_emit(struct vc4_compile *c, struct qinst *inst)
 381 {
 382         if (inst->dst.file == QFILE_TEMP)
 383                 c->defs[inst->dst.index] = inst;
 384
 385         qir_emit_nodef(c, inst);
 386 }
 387
 388 bool
 389 qir_reg_equals(struct qreg a, struct qreg b)
 390 {
 391         return a.file == b.file && a.index == b.index;
 392 }
 393
 394 struct vc4_compile *
 395 qir_compile_init(void)
 396 {
 397         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 398
 399         list_inithead(&c->instructions);
 400
 401         c->output_position_index = -1;
 402         c->output_color_index = -1;
 403         c->output_point_size_index = -1;
 404         c->output_sample_mask_index = -1;
 405
 406         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 407                                             _mesa_key_pointer_equal);
 408
 409         return c;
 410 }
 411
 412 void
 413 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 414 {
 415         if (qinst->dst.file == QFILE_TEMP)
 416                 c->defs[qinst->dst.index] = NULL;
 417
 418         list_del(&qinst->link);
 419         free(qinst->src);
 420         free(qinst);
 421 }
 422
 423 struct qreg
 424 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 425 {
 426         while (reg.file == QFILE_TEMP &&
 427                c->defs[reg.index] &&
 428                c->defs[reg.index]->op == QOP_MOV &&
 429                !c->defs[reg.index]->dst.pack) {
 430                 reg = c->defs[reg.index]->src[0];
 431         }
 432
 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 (struct qreg) { QFILE_UNIF, i };
 469                 }
 470         }
 471
 472         uint32_t uniform = c->num_uniforms++;
 473         struct qreg u = { QFILE_UNIF, uniform };
 474
 475         if (uniform >= c->uniform_array_size) {
 476                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 477                                              c->uniform_array_size * 2);
 478
 479                 c->uniform_data = reralloc(c, c->uniform_data,
 480                                            uint32_t,
 481                                            c->uniform_array_size);
 482                 c->uniform_contents = reralloc(c, c->uniform_contents,
 483                                                enum quniform_contents,
 484                                                c->uniform_array_size);
 485         }
 486
 487         c->uniform_contents[uniform] = contents;
 488         c->uniform_data[uniform] = data;
 489
 490         return u;
 491 }
 492
 493 void
 494 qir_SF(struct vc4_compile *c, struct qreg src)
 495 {
 496         struct qinst *last_inst = NULL;
 497         if (!list_empty(&c->instructions))
 498                 last_inst = (struct qinst *)c->instructions.prev;
 499
 500         if (!last_inst ||
 501             last_inst->dst.file != src.file ||
 502             last_inst->dst.index != src.index ||
 503             qir_is_multi_instruction(last_inst)) {
 504                 src = qir_MOV(c, src);
 505                 last_inst = (struct qinst *)c->instructions.prev;
 506         }
 507         last_inst->sf = true;
 508 }
 509
 510 #define OPTPASS(func)                                                   \
 511         do {                                                            \
 512                 bool stage_progress = func(c);                          \
 513                 if (stage_progress) {                                   \
 514                         progress = true;                                \
 515                         if (print_opt_debug) {                          \
 516                                 fprintf(stderr,                         \
 517                                         "QIR opt pass %2d: %s progress\n", \
 518                                         pass, #func);                   \
 519                         }                                               \
 520                 }                                                       \
 521         } while (0)
 522
 523 void
 524 qir_optimize(struct vc4_compile *c)
 525 {
 526         bool print_opt_debug = false;
 527         int pass = 1;
 528
 529         while (true) {
 530                 bool progress = false;
 531
 532                 OPTPASS(qir_opt_algebraic);
 533                 OPTPASS(qir_opt_cse);
 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_writes);
 539
 540                 if (!progress)
 541                         break;
 542
 543                 pass++;
 544         }
 545 }