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_FADD] = { "fadd", 1, 2 },
  40         [QOP_FSUB] = { "fsub", 1, 2 },
  41         [QOP_FMUL] = { "fmul", 1, 2 },
  42         [QOP_MUL24] = { "mul24", 1, 2 },
  43         [QOP_FMIN] = { "fmin", 1, 2 },
  44         [QOP_FMAX] = { "fmax", 1, 2 },
  45         [QOP_FMINABS] = { "fminabs", 1, 2 },
  46         [QOP_FMAXABS] = { "fmaxabs", 1, 2 },
  47         [QOP_FTOI] = { "ftoi", 1, 1 },
  48         [QOP_ITOF] = { "itof", 1, 1 },
  49         [QOP_ADD] = { "add", 1, 2 },
  50         [QOP_SUB] = { "sub", 1, 2 },
  51         [QOP_SHR] = { "shr", 1, 2 },
  52         [QOP_ASR] = { "asr", 1, 2 },
  53         [QOP_SHL] = { "shl", 1, 2 },
  54         [QOP_MIN] = { "min", 1, 2 },
  55         [QOP_MAX] = { "max", 1, 2 },
  56         [QOP_AND] = { "and", 1, 2 },
  57         [QOP_OR] = { "or", 1, 2 },
  58         [QOP_XOR] = { "xor", 1, 2 },
  59         [QOP_NOT] = { "not", 1, 1 },
  60
  61         [QOP_SEL_X_0_NS] = { "fsel_x_0_ns", 1, 1, false, true },
  62         [QOP_SEL_X_0_NC] = { "fsel_x_0_nc", 1, 1, false, true },
  63         [QOP_SEL_X_0_ZS] = { "fsel_x_0_zs", 1, 1, false, true },
  64         [QOP_SEL_X_0_ZC] = { "fsel_x_0_zc", 1, 1, false, true },
  65         [QOP_SEL_X_Y_NS] = { "fsel_x_y_ns", 1, 2, false, true },
  66         [QOP_SEL_X_Y_NC] = { "fsel_x_y_nc", 1, 2, false, true },
  67         [QOP_SEL_X_Y_ZS] = { "fsel_x_y_zs", 1, 2, false, true },
  68         [QOP_SEL_X_Y_ZC] = { "fsel_x_y_zc", 1, 2, false, true },
  69
  70         [QOP_RCP] = { "rcp", 1, 1, false, true },
  71         [QOP_RSQ] = { "rsq", 1, 1, false, true },
  72         [QOP_EXP2] = { "exp2", 1, 2, false, true },
  73         [QOP_LOG2] = { "log2", 1, 2, false, true },
  74         [QOP_PACK_8888_F] = { "pack_8888_f", 1, 1, false, true },
  75         [QOP_PACK_8A_F] = { "pack_8a_f", 1, 2, false, true },
  76         [QOP_PACK_8B_F] = { "pack_8b_f", 1, 2, false, true },
  77         [QOP_PACK_8C_F] = { "pack_8c_f", 1, 2, false, true },
  78         [QOP_PACK_8D_F] = { "pack_8d_f", 1, 2, false, true },
  79         [QOP_PACK_SCALED] = { "pack_scaled", 1, 2, false, true },
  80         [QOP_TLB_DISCARD_SETUP] = { "discard", 0, 1, true },
  81         [QOP_TLB_STENCIL_SETUP] = { "tlb_stencil_setup", 0, 1, true },
  82         [QOP_TLB_Z_WRITE] = { "tlb_z", 0, 1, true },
  83         [QOP_TLB_COLOR_WRITE] = { "tlb_color", 0, 1, true },
  84         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0 },
  85         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  86
  87         [QOP_FRAG_X] = { "frag_x", 1, 0 },
  88         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  89         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  90         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  91         [QOP_FRAG_REV_FLAG] = { "frag_rev_flag", 1, 0 },
  92
  93         [QOP_TEX_S] = { "tex_s", 0, 2 },
  94         [QOP_TEX_T] = { "tex_t", 0, 2 },
  95         [QOP_TEX_R] = { "tex_r", 0, 2 },
  96         [QOP_TEX_B] = { "tex_b", 0, 2 },
  97         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
  98         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
  99         [QOP_R4_UNPACK_A] = { "r4_unpack_a", 1, 1 },
 100         [QOP_R4_UNPACK_B] = { "r4_unpack_b", 1, 1 },
 101         [QOP_R4_UNPACK_C] = { "r4_unpack_c", 1, 1 },
 102         [QOP_R4_UNPACK_D] = { "r4_unpack_d", 1, 1 },
 103         [QOP_UNPACK_8A_F] = { "unpack_8a_f", 1, 1 },
 104         [QOP_UNPACK_8B_F] = { "unpack_8b_f", 1, 1 },
 105         [QOP_UNPACK_8C_F] = { "unpack_8c_f", 1, 1 },
 106         [QOP_UNPACK_8D_F] = { "unpack_8d_f", 1, 1 },
 107         [QOP_UNPACK_16A_F] = { "unpack_16a_f", 1, 1 },
 108         [QOP_UNPACK_16B_F] = { "unpack_16b_f", 1, 1 },
 109         [QOP_UNPACK_8A_I] = { "unpack_8a_i", 1, 1 },
 110         [QOP_UNPACK_8B_I] = { "unpack_8b_i", 1, 1 },
 111         [QOP_UNPACK_8C_I] = { "unpack_8c_i", 1, 1 },
 112         [QOP_UNPACK_8D_I] = { "unpack_8d_i", 1, 1 },
 113         [QOP_UNPACK_16A_I] = { "unpack_16a_i", 1, 1 },
 114         [QOP_UNPACK_16B_I] = { "unpack_16b_i", 1, 1 },
 115 };
 116
 117 static const char *
 118 qir_get_op_name(enum qop qop)
 119 {
 120         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 121                 return qir_op_info[qop].name;
 122         else
 123                 return "???";
 124 }
 125
 126 int
 127 qir_get_op_nsrc(enum qop qop)
 128 {
 129         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 130                 return qir_op_info[qop].nsrc;
 131         else
 132                 abort();
 133 }
 134
 135 /**
 136  * Returns whether the instruction has any side effects that must be
 137  * preserved.
 138  */
 139 bool
 140 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 141 {
 142         return qir_op_info[inst->op].has_side_effects;
 143 }
 144
 145 bool
 146 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 147 {
 148         /* We can dead-code eliminate varyings, because we only tell the VS
 149          * about the live ones at the end.  But we have to preserve the
 150          * point/line coordinates reads, because they're generated by
 151          * fixed-function hardware.
 152          */
 153         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 154                 if (inst->src[i].file == QFILE_VARY &&
 155                     c->input_semantics[inst->src[i].index].semantic == 0xff) {
 156                         return true;
 157                 }
 158
 159                 if (inst->src[i].file == QFILE_VPM)
 160                         return true;
 161         }
 162
 163         if (inst->dst.file == QFILE_VPM)
 164                 return true;
 165
 166         return false;
 167 }
 168
 169 bool
 170 qir_is_multi_instruction(struct qinst *inst)
 171 {
 172         return qir_op_info[inst->op].multi_instruction;
 173 }
 174
 175 bool
 176 qir_is_tex(struct qinst *inst)
 177 {
 178         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 179 }
 180
 181 bool
 182 qir_depends_on_flags(struct qinst *inst)
 183 {
 184         switch (inst->op) {
 185         case QOP_SEL_X_0_NS:
 186         case QOP_SEL_X_0_NC:
 187         case QOP_SEL_X_0_ZS:
 188         case QOP_SEL_X_0_ZC:
 189         case QOP_SEL_X_Y_NS:
 190         case QOP_SEL_X_Y_NC:
 191         case QOP_SEL_X_Y_ZS:
 192         case QOP_SEL_X_Y_ZC:
 193                 return true;
 194         default:
 195                 return false;
 196         }
 197 }
 198
 199 bool
 200 qir_src_needs_a_file(struct qinst *inst)
 201 {
 202         switch (inst->op) {
 203         case QOP_UNPACK_8A_F:
 204         case QOP_UNPACK_8B_F:
 205         case QOP_UNPACK_8C_F:
 206         case QOP_UNPACK_8D_F:
 207         case QOP_UNPACK_16A_F:
 208         case QOP_UNPACK_16B_F:
 209         case QOP_UNPACK_8A_I:
 210         case QOP_UNPACK_8B_I:
 211         case QOP_UNPACK_8C_I:
 212         case QOP_UNPACK_8D_I:
 213         case QOP_UNPACK_16A_I:
 214         case QOP_UNPACK_16B_I:
 215                 return true;
 216         default:
 217                 return false;
 218         }
 219 }
 220
 221 bool
 222 qir_writes_r4(struct qinst *inst)
 223 {
 224         switch (inst->op) {
 225         case QOP_TEX_RESULT:
 226         case QOP_TLB_COLOR_READ:
 227         case QOP_RCP:
 228         case QOP_RSQ:
 229         case QOP_EXP2:
 230         case QOP_LOG2:
 231                 return true;
 232         default:
 233                 return false;
 234         }
 235 }
 236
 237 bool
 238 qir_reads_r4(struct qinst *inst)
 239 {
 240         switch (inst->op) {
 241         case QOP_R4_UNPACK_A:
 242         case QOP_R4_UNPACK_B:
 243         case QOP_R4_UNPACK_C:
 244         case QOP_R4_UNPACK_D:
 245                 return true;
 246         default:
 247                 return false;
 248         }
 249 }
 250
 251 static void
 252 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 253 {
 254         static const char *files[] = {
 255                 [QFILE_TEMP] = "t",
 256                 [QFILE_VARY] = "v",
 257                 [QFILE_UNIF] = "u",
 258         };
 259
 260         if (reg.file == QFILE_NULL) {
 261                 fprintf(stderr, "null");
 262         } else if (reg.file == QFILE_SMALL_IMM) {
 263                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 264                         fprintf(stderr, "%d", reg.index);
 265                 else
 266                         fprintf(stderr, "%f", uif(reg.index));
 267         } else if (reg.file == QFILE_VPM) {
 268                 if (write) {
 269                         fprintf(stderr, "vpm");
 270                 } else {
 271                         fprintf(stderr, "vpm%d.%d",
 272                                 reg.index / 4, reg.index % 4);
 273                 }
 274         } else {
 275                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 276         }
 277
 278         if (reg.file == QFILE_UNIF &&
 279             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 280                 fprintf(stderr, " (0x%08x / %f)",
 281                         c->uniform_data[reg.index],
 282                         uif(c->uniform_data[reg.index]));
 283         }
 284 }
 285
 286 void
 287 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 288 {
 289         fprintf(stderr, "%s%s ",
 290                 qir_get_op_name(inst->op),
 291                 inst->sf ? ".sf" : "");
 292
 293         qir_print_reg(c, inst->dst, true);
 294         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 295                 fprintf(stderr, ", ");
 296                 qir_print_reg(c, inst->src[i], false);
 297         }
 298 }
 299
 300 void
 301 qir_dump(struct vc4_compile *c)
 302 {
 303         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 304                 qir_dump_inst(c, inst);
 305                 fprintf(stderr, "\n");
 306         }
 307 }
 308
 309 struct qreg
 310 qir_get_temp(struct vc4_compile *c)
 311 {
 312         struct qreg reg;
 313
 314         reg.file = QFILE_TEMP;
 315         reg.index = c->num_temps++;
 316
 317         if (c->num_temps > c->defs_array_size) {
 318                 uint32_t old_size = c->defs_array_size;
 319                 c->defs_array_size = MAX2(old_size * 2, 16);
 320                 c->defs = reralloc(c, c->defs, struct qinst *,
 321                                    c->defs_array_size);
 322                 memset(&c->defs[old_size], 0,
 323                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 324         }
 325
 326         return reg;
 327 }
 328
 329 struct qinst *
 330 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 331 {
 332         struct qinst *inst = CALLOC_STRUCT(qinst);
 333
 334         inst->op = op;
 335         inst->dst = dst;
 336         inst->src = calloc(2, sizeof(inst->src[0]));
 337         inst->src[0] = src0;
 338         inst->src[1] = src1;
 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         list_addtail(&inst->link, &c->instructions);
 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_clipvertex_index = -1;
 387         c->output_color_index = -1;
 388         c->output_point_size_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         while (reg.file == QFILE_TEMP && c->defs[reg.index]->op == QOP_MOV)
 411                 reg = c->defs[reg.index]->src[0];
 412
 413         return reg;
 414 }
 415
 416 void
 417 qir_compile_destroy(struct vc4_compile *c)
 418 {
 419         while (!list_empty(&c->instructions)) {
 420                 struct qinst *qinst =
 421                         (struct qinst *)c->instructions.next;
 422                 qir_remove_instruction(c, qinst);
 423         }
 424
 425         ralloc_free(c);
 426 }
 427
 428 const char *
 429 qir_get_stage_name(enum qstage stage)
 430 {
 431         static const char *names[] = {
 432                 [QSTAGE_FRAG] = "FS",
 433                 [QSTAGE_VERT] = "VS",
 434                 [QSTAGE_COORD] = "CS",
 435         };
 436
 437         return names[stage];
 438 }
 439
 440 struct qreg
 441 qir_uniform(struct vc4_compile *c,
 442             enum quniform_contents contents,
 443             uint32_t data)
 444 {
 445         for (int i = 0; i < c->num_uniforms; i++) {
 446                 if (c->uniform_contents[i] == contents &&
 447                     c->uniform_data[i] == data) {
 448                         return (struct qreg) { QFILE_UNIF, i };
 449                 }
 450         }
 451
 452         uint32_t uniform = c->num_uniforms++;
 453         struct qreg u = { QFILE_UNIF, uniform };
 454
 455         if (uniform >= c->uniform_array_size) {
 456                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 457                                              c->uniform_array_size * 2);
 458
 459                 c->uniform_data = reralloc(c, c->uniform_data,
 460                                            uint32_t,
 461                                            c->uniform_array_size);
 462                 c->uniform_contents = reralloc(c, c->uniform_contents,
 463                                                enum quniform_contents,
 464                                                c->uniform_array_size);
 465         }
 466
 467         c->uniform_contents[uniform] = contents;
 468         c->uniform_data[uniform] = data;
 469
 470         return u;
 471 }
 472
 473 void
 474 qir_SF(struct vc4_compile *c, struct qreg src)
 475 {
 476         struct qinst *last_inst = NULL;
 477         if (!list_empty(&c->instructions))
 478                 last_inst = (struct qinst *)c->instructions.prev;
 479
 480         if (!last_inst ||
 481             last_inst->dst.file != src.file ||
 482             last_inst->dst.index != src.index ||
 483             qir_is_multi_instruction(last_inst)) {
 484                 src = qir_MOV(c, src);
 485                 last_inst = (struct qinst *)c->instructions.prev;
 486         }
 487         last_inst->sf = true;
 488 }
 489
 490 #define OPTPASS(func)                                                   \
 491         do {                                                            \
 492                 bool stage_progress = func(c);                          \
 493                 if (stage_progress) {                                   \
 494                         progress = true;                                \
 495                         if (print_opt_debug) {                          \
 496                                 fprintf(stderr,                         \
 497                                         "QIR opt pass %2d: %s progress\n", \
 498                                         pass, #func);                   \
 499                         }                                               \
 500                 }                                                       \
 501         } while (0)
 502
 503 void
 504 qir_optimize(struct vc4_compile *c)
 505 {
 506         bool print_opt_debug = false;
 507         int pass = 1;
 508
 509         while (true) {
 510                 bool progress = false;
 511
 512                 OPTPASS(qir_opt_algebraic);
 513                 OPTPASS(qir_opt_cse);
 514                 OPTPASS(qir_opt_constant_folding);
 515                 OPTPASS(qir_opt_copy_propagation);
 516                 OPTPASS(qir_opt_dead_code);
 517                 OPTPASS(qir_opt_small_immediates);
 518                 OPTPASS(qir_opt_vpm_writes);
 519
 520                 if (!progress)
 521                         break;
 522
 523                 pass++;
 524         }
 525 }