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 },
  75         [QOP_PACK_8A_F] = { "pack_8a_f", 1, 1 },
  76         [QOP_PACK_8B_F] = { "pack_8b_f", 1, 1 },
  77         [QOP_PACK_8C_F] = { "pack_8c_f", 1, 1 },
  78         [QOP_PACK_8D_F] = { "pack_8d_f", 1, 1 },
  79         [QOP_TLB_DISCARD_SETUP] = { "discard", 0, 1, true },
  80         [QOP_TLB_STENCIL_SETUP] = { "tlb_stencil_setup", 0, 1, true },
  81         [QOP_TLB_Z_WRITE] = { "tlb_z", 0, 1, true },
  82         [QOP_TLB_COLOR_WRITE] = { "tlb_color", 0, 1, true },
  83         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0 },
  84         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  85
  86         [QOP_FRAG_X] = { "frag_x", 1, 0 },
  87         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  88         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  89         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  90         [QOP_FRAG_REV_FLAG] = { "frag_rev_flag", 1, 0 },
  91
  92         [QOP_TEX_S] = { "tex_s", 0, 2 },
  93         [QOP_TEX_T] = { "tex_t", 0, 2 },
  94         [QOP_TEX_R] = { "tex_r", 0, 2 },
  95         [QOP_TEX_B] = { "tex_b", 0, 2 },
  96         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
  97         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
  98         [QOP_UNPACK_8A_F] = { "unpack_8a_f", 1, 1 },
  99         [QOP_UNPACK_8B_F] = { "unpack_8b_f", 1, 1 },
 100         [QOP_UNPACK_8C_F] = { "unpack_8c_f", 1, 1 },
 101         [QOP_UNPACK_8D_F] = { "unpack_8d_f", 1, 1 },
 102         [QOP_UNPACK_16A_F] = { "unpack_16a_f", 1, 1 },
 103         [QOP_UNPACK_16B_F] = { "unpack_16b_f", 1, 1 },
 104         [QOP_UNPACK_8A_I] = { "unpack_8a_i", 1, 1 },
 105         [QOP_UNPACK_8B_I] = { "unpack_8b_i", 1, 1 },
 106         [QOP_UNPACK_8C_I] = { "unpack_8c_i", 1, 1 },
 107         [QOP_UNPACK_8D_I] = { "unpack_8d_i", 1, 1 },
 108         [QOP_UNPACK_16A_I] = { "unpack_16a_i", 1, 1 },
 109         [QOP_UNPACK_16B_I] = { "unpack_16b_i", 1, 1 },
 110 };
 111
 112 static const char *
 113 qir_get_op_name(enum qop qop)
 114 {
 115         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 116                 return qir_op_info[qop].name;
 117         else
 118                 return "???";
 119 }
 120
 121 int
 122 qir_get_op_nsrc(enum qop qop)
 123 {
 124         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 125                 return qir_op_info[qop].nsrc;
 126         else
 127                 abort();
 128 }
 129
 130 /**
 131  * Returns whether the instruction has any side effects that must be
 132  * preserved.
 133  */
 134 bool
 135 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 136 {
 137         return qir_op_info[inst->op].has_side_effects;
 138 }
 139
 140 bool
 141 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 142 {
 143         /* We can dead-code eliminate varyings, because we only tell the VS
 144          * about the live ones at the end.  But we have to preserve the
 145          * point/line coordinates reads, because they're generated by
 146          * fixed-function hardware.
 147          */
 148         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 149                 if (inst->src[i].file == QFILE_VARY &&
 150                     c->input_slots[inst->src[i].index].slot == 0xff) {
 151                         return true;
 152                 }
 153
 154                 if (inst->src[i].file == QFILE_VPM)
 155                         return true;
 156         }
 157
 158         if (inst->dst.file == QFILE_VPM)
 159                 return true;
 160
 161         return false;
 162 }
 163
 164 bool
 165 qir_is_multi_instruction(struct qinst *inst)
 166 {
 167         return qir_op_info[inst->op].multi_instruction;
 168 }
 169
 170 bool
 171 qir_is_mul(struct qinst *inst)
 172 {
 173         switch (inst->op) {
 174         case QOP_FMUL:
 175         case QOP_MUL24:
 176                 return true;
 177         default:
 178                 return false;
 179         }
 180 }
 181
 182 bool
 183 qir_is_tex(struct qinst *inst)
 184 {
 185         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 186 }
 187
 188 bool
 189 qir_depends_on_flags(struct qinst *inst)
 190 {
 191         switch (inst->op) {
 192         case QOP_SEL_X_0_NS:
 193         case QOP_SEL_X_0_NC:
 194         case QOP_SEL_X_0_ZS:
 195         case QOP_SEL_X_0_ZC:
 196         case QOP_SEL_X_Y_NS:
 197         case QOP_SEL_X_Y_NC:
 198         case QOP_SEL_X_Y_ZS:
 199         case QOP_SEL_X_Y_ZC:
 200                 return true;
 201         default:
 202                 return false;
 203         }
 204 }
 205
 206 bool
 207 qir_src_needs_a_file(struct qinst *inst)
 208 {
 209         switch (inst->op) {
 210         case QOP_UNPACK_8A_F:
 211         case QOP_UNPACK_8B_F:
 212         case QOP_UNPACK_8C_F:
 213         case QOP_UNPACK_8D_F:
 214         case QOP_UNPACK_16A_F:
 215         case QOP_UNPACK_16B_F:
 216         case QOP_UNPACK_8A_I:
 217         case QOP_UNPACK_8B_I:
 218         case QOP_UNPACK_8C_I:
 219         case QOP_UNPACK_8D_I:
 220         case QOP_UNPACK_16A_I:
 221         case QOP_UNPACK_16B_I:
 222                 return true;
 223         default:
 224                 return false;
 225         }
 226 }
 227
 228 bool
 229 qir_writes_r4(struct qinst *inst)
 230 {
 231         switch (inst->op) {
 232         case QOP_TEX_RESULT:
 233         case QOP_TLB_COLOR_READ:
 234         case QOP_RCP:
 235         case QOP_RSQ:
 236         case QOP_EXP2:
 237         case QOP_LOG2:
 238                 return true;
 239         default:
 240                 return false;
 241         }
 242 }
 243
 244 static void
 245 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 246 {
 247         static const char *files[] = {
 248                 [QFILE_TEMP] = "t",
 249                 [QFILE_VARY] = "v",
 250                 [QFILE_UNIF] = "u",
 251         };
 252
 253         if (reg.file == QFILE_NULL) {
 254                 fprintf(stderr, "null");
 255         } else if (reg.file == QFILE_SMALL_IMM) {
 256                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 257                         fprintf(stderr, "%d", reg.index);
 258                 else
 259                         fprintf(stderr, "%f", uif(reg.index));
 260         } else if (reg.file == QFILE_VPM) {
 261                 if (write) {
 262                         fprintf(stderr, "vpm");
 263                 } else {
 264                         fprintf(stderr, "vpm%d.%d",
 265                                 reg.index / 4, reg.index % 4);
 266                 }
 267         } else {
 268                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 269         }
 270
 271         if (reg.file == QFILE_UNIF &&
 272             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 273                 fprintf(stderr, " (0x%08x / %f)",
 274                         c->uniform_data[reg.index],
 275                         uif(c->uniform_data[reg.index]));
 276         }
 277 }
 278
 279 void
 280 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 281 {
 282         fprintf(stderr, "%s%s ",
 283                 qir_get_op_name(inst->op),
 284                 inst->sf ? ".sf" : "");
 285
 286         qir_print_reg(c, inst->dst, true);
 287         if (inst->dst.pack) {
 288                 if (inst->dst.pack) {
 289                         if (qir_is_mul(inst))
 290                                 vc4_qpu_disasm_pack_mul(stderr, inst->dst.pack);
 291                         else
 292                                 vc4_qpu_disasm_pack_a(stderr, inst->dst.pack);
 293                 }
 294         }
 295         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 296                 fprintf(stderr, ", ");
 297                 qir_print_reg(c, inst->src[i], false);
 298         }
 299 }
 300
 301 void
 302 qir_dump(struct vc4_compile *c)
 303 {
 304         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 305                 qir_dump_inst(c, inst);
 306                 fprintf(stderr, "\n");
 307         }
 308 }
 309
 310 struct qreg
 311 qir_get_temp(struct vc4_compile *c)
 312 {
 313         struct qreg reg;
 314
 315         reg.file = QFILE_TEMP;
 316         reg.index = c->num_temps++;
 317         reg.pack = 0;
 318
 319         if (c->num_temps > c->defs_array_size) {
 320                 uint32_t old_size = c->defs_array_size;
 321                 c->defs_array_size = MAX2(old_size * 2, 16);
 322                 c->defs = reralloc(c, c->defs, struct qinst *,
 323                                    c->defs_array_size);
 324                 memset(&c->defs[old_size], 0,
 325                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 326         }
 327
 328         return reg;
 329 }
 330
 331 struct qinst *
 332 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 333 {
 334         struct qinst *inst = CALLOC_STRUCT(qinst);
 335
 336         inst->op = op;
 337         inst->dst = dst;
 338         inst->src = calloc(2, sizeof(inst->src[0]));
 339         inst->src[0] = src0;
 340         inst->src[1] = src1;
 341
 342         return inst;
 343 }
 344
 345 struct qinst *
 346 qir_inst4(enum qop op, struct qreg dst,
 347           struct qreg a,
 348           struct qreg b,
 349           struct qreg c,
 350           struct qreg d)
 351 {
 352         struct qinst *inst = CALLOC_STRUCT(qinst);
 353
 354         inst->op = op;
 355         inst->dst = dst;
 356         inst->src = calloc(4, sizeof(*inst->src));
 357         inst->src[0] = a;
 358         inst->src[1] = b;
 359         inst->src[2] = c;
 360         inst->src[3] = d;
 361
 362         return inst;
 363 }
 364
 365 void
 366 qir_emit(struct vc4_compile *c, struct qinst *inst)
 367 {
 368         if (inst->dst.file == QFILE_TEMP)
 369                 c->defs[inst->dst.index] = inst;
 370
 371         qir_emit_nodef(c, inst);
 372 }
 373
 374 bool
 375 qir_reg_equals(struct qreg a, struct qreg b)
 376 {
 377         return a.file == b.file && a.index == b.index;
 378 }
 379
 380 struct vc4_compile *
 381 qir_compile_init(void)
 382 {
 383         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 384
 385         list_inithead(&c->instructions);
 386
 387         c->output_position_index = -1;
 388         c->output_color_index = -1;
 389         c->output_point_size_index = -1;
 390
 391         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 392                                             _mesa_key_pointer_equal);
 393
 394         return c;
 395 }
 396
 397 void
 398 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 399 {
 400         if (qinst->dst.file == QFILE_TEMP)
 401                 c->defs[qinst->dst.index] = NULL;
 402
 403         list_del(&qinst->link);
 404         free(qinst->src);
 405         free(qinst);
 406 }
 407
 408 struct qreg
 409 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 410 {
 411         while (reg.file == QFILE_TEMP &&
 412                c->defs[reg.index] &&
 413                c->defs[reg.index]->op == QOP_MOV) {
 414                 reg = c->defs[reg.index]->src[0];
 415         }
 416
 417         return reg;
 418 }
 419
 420 void
 421 qir_compile_destroy(struct vc4_compile *c)
 422 {
 423         while (!list_empty(&c->instructions)) {
 424                 struct qinst *qinst =
 425                         (struct qinst *)c->instructions.next;
 426                 qir_remove_instruction(c, qinst);
 427         }
 428
 429         ralloc_free(c);
 430 }
 431
 432 const char *
 433 qir_get_stage_name(enum qstage stage)
 434 {
 435         static const char *names[] = {
 436                 [QSTAGE_FRAG] = "FS",
 437                 [QSTAGE_VERT] = "VS",
 438                 [QSTAGE_COORD] = "CS",
 439         };
 440
 441         return names[stage];
 442 }
 443
 444 struct qreg
 445 qir_uniform(struct vc4_compile *c,
 446             enum quniform_contents contents,
 447             uint32_t data)
 448 {
 449         for (int i = 0; i < c->num_uniforms; i++) {
 450                 if (c->uniform_contents[i] == contents &&
 451                     c->uniform_data[i] == data) {
 452                         return (struct qreg) { QFILE_UNIF, i };
 453                 }
 454         }
 455
 456         uint32_t uniform = c->num_uniforms++;
 457         struct qreg u = { QFILE_UNIF, uniform };
 458
 459         if (uniform >= c->uniform_array_size) {
 460                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 461                                              c->uniform_array_size * 2);
 462
 463                 c->uniform_data = reralloc(c, c->uniform_data,
 464                                            uint32_t,
 465                                            c->uniform_array_size);
 466                 c->uniform_contents = reralloc(c, c->uniform_contents,
 467                                                enum quniform_contents,
 468                                                c->uniform_array_size);
 469         }
 470
 471         c->uniform_contents[uniform] = contents;
 472         c->uniform_data[uniform] = data;
 473
 474         return u;
 475 }
 476
 477 void
 478 qir_SF(struct vc4_compile *c, struct qreg src)
 479 {
 480         struct qinst *last_inst = NULL;
 481         if (!list_empty(&c->instructions))
 482                 last_inst = (struct qinst *)c->instructions.prev;
 483
 484         if (!last_inst ||
 485             last_inst->dst.file != src.file ||
 486             last_inst->dst.index != src.index ||
 487             qir_is_multi_instruction(last_inst)) {
 488                 src = qir_MOV(c, src);
 489                 last_inst = (struct qinst *)c->instructions.prev;
 490         }
 491         last_inst->sf = true;
 492 }
 493
 494 #define OPTPASS(func)                                                   \
 495         do {                                                            \
 496                 bool stage_progress = func(c);                          \
 497                 if (stage_progress) {                                   \
 498                         progress = true;                                \
 499                         if (print_opt_debug) {                          \
 500                                 fprintf(stderr,                         \
 501                                         "QIR opt pass %2d: %s progress\n", \
 502                                         pass, #func);                   \
 503                         }                                               \
 504                 }                                                       \
 505         } while (0)
 506
 507 void
 508 qir_optimize(struct vc4_compile *c)
 509 {
 510         bool print_opt_debug = false;
 511         int pass = 1;
 512
 513         while (true) {
 514                 bool progress = false;
 515
 516                 OPTPASS(qir_opt_algebraic);
 517                 OPTPASS(qir_opt_cse);
 518                 OPTPASS(qir_opt_constant_folding);
 519                 OPTPASS(qir_opt_copy_propagation);
 520                 OPTPASS(qir_opt_dead_code);
 521                 OPTPASS(qir_opt_small_immediates);
 522                 OPTPASS(qir_opt_vpm_writes);
 523
 524                 if (!progress)
 525                         break;
 526
 527                 pass++;
 528         }
 529 }