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_PACK_16A_I] = { "pack_16a_i", 1, 1 },
  80         [QOP_PACK_16B_I] = { "pack_16b_i", 1, 1 },
  81         [QOP_TLB_DISCARD_SETUP] = { "discard", 0, 1, true },
  82         [QOP_TLB_STENCIL_SETUP] = { "tlb_stencil_setup", 0, 1, true },
  83         [QOP_TLB_Z_WRITE] = { "tlb_z", 0, 1, true },
  84         [QOP_TLB_COLOR_WRITE] = { "tlb_color", 0, 1, true },
  85         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0 },
  86         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  87
  88         [QOP_FRAG_X] = { "frag_x", 1, 0 },
  89         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  90         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  91         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  92         [QOP_FRAG_REV_FLAG] = { "frag_rev_flag", 1, 0 },
  93
  94         [QOP_TEX_S] = { "tex_s", 0, 2 },
  95         [QOP_TEX_T] = { "tex_t", 0, 2 },
  96         [QOP_TEX_R] = { "tex_r", 0, 2 },
  97         [QOP_TEX_B] = { "tex_b", 0, 2 },
  98         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
  99         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
 100         [QOP_UNPACK_8A_F] = { "unpack_8a_f", 1, 1 },
 101         [QOP_UNPACK_8B_F] = { "unpack_8b_f", 1, 1 },
 102         [QOP_UNPACK_8C_F] = { "unpack_8c_f", 1, 1 },
 103         [QOP_UNPACK_8D_F] = { "unpack_8d_f", 1, 1 },
 104         [QOP_UNPACK_16A_F] = { "unpack_16a_f", 1, 1 },
 105         [QOP_UNPACK_16B_F] = { "unpack_16b_f", 1, 1 },
 106         [QOP_UNPACK_8A_I] = { "unpack_8a_i", 1, 1 },
 107         [QOP_UNPACK_8B_I] = { "unpack_8b_i", 1, 1 },
 108         [QOP_UNPACK_8C_I] = { "unpack_8c_i", 1, 1 },
 109         [QOP_UNPACK_8D_I] = { "unpack_8d_i", 1, 1 },
 110         [QOP_UNPACK_16A_I] = { "unpack_16a_i", 1, 1 },
 111         [QOP_UNPACK_16B_I] = { "unpack_16b_i", 1, 1 },
 112 };
 113
 114 static const char *
 115 qir_get_op_name(enum qop qop)
 116 {
 117         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 118                 return qir_op_info[qop].name;
 119         else
 120                 return "???";
 121 }
 122
 123 int
 124 qir_get_op_nsrc(enum qop qop)
 125 {
 126         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 127                 return qir_op_info[qop].nsrc;
 128         else
 129                 abort();
 130 }
 131
 132 /**
 133  * Returns whether the instruction has any side effects that must be
 134  * preserved.
 135  */
 136 bool
 137 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 138 {
 139         return qir_op_info[inst->op].has_side_effects;
 140 }
 141
 142 bool
 143 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 144 {
 145         /* We can dead-code eliminate varyings, because we only tell the VS
 146          * about the live ones at the end.  But we have to preserve the
 147          * point/line coordinates reads, because they're generated by
 148          * fixed-function hardware.
 149          */
 150         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 151                 if (inst->src[i].file == QFILE_VARY &&
 152                     c->input_semantics[inst->src[i].index].semantic == 0xff) {
 153                         return true;
 154                 }
 155
 156                 if (inst->src[i].file == QFILE_VPM)
 157                         return true;
 158         }
 159
 160         if (inst->dst.file == QFILE_VPM)
 161                 return true;
 162
 163         return false;
 164 }
 165
 166 bool
 167 qir_is_multi_instruction(struct qinst *inst)
 168 {
 169         return qir_op_info[inst->op].multi_instruction;
 170 }
 171
 172 bool
 173 qir_is_mul(struct qinst *inst)
 174 {
 175         switch (inst->op) {
 176         case QOP_FMUL:
 177         case QOP_MUL24:
 178                 return true;
 179         default:
 180                 return false;
 181         }
 182 }
 183
 184 bool
 185 qir_is_tex(struct qinst *inst)
 186 {
 187         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 188 }
 189
 190 bool
 191 qir_depends_on_flags(struct qinst *inst)
 192 {
 193         switch (inst->op) {
 194         case QOP_SEL_X_0_NS:
 195         case QOP_SEL_X_0_NC:
 196         case QOP_SEL_X_0_ZS:
 197         case QOP_SEL_X_0_ZC:
 198         case QOP_SEL_X_Y_NS:
 199         case QOP_SEL_X_Y_NC:
 200         case QOP_SEL_X_Y_ZS:
 201         case QOP_SEL_X_Y_ZC:
 202                 return true;
 203         default:
 204                 return false;
 205         }
 206 }
 207
 208 bool
 209 qir_src_needs_a_file(struct qinst *inst)
 210 {
 211         switch (inst->op) {
 212         case QOP_UNPACK_8A_F:
 213         case QOP_UNPACK_8B_F:
 214         case QOP_UNPACK_8C_F:
 215         case QOP_UNPACK_8D_F:
 216         case QOP_UNPACK_16A_F:
 217         case QOP_UNPACK_16B_F:
 218         case QOP_UNPACK_8A_I:
 219         case QOP_UNPACK_8B_I:
 220         case QOP_UNPACK_8C_I:
 221         case QOP_UNPACK_8D_I:
 222         case QOP_UNPACK_16A_I:
 223         case QOP_UNPACK_16B_I:
 224                 return true;
 225         default:
 226                 return false;
 227         }
 228 }
 229
 230 bool
 231 qir_writes_r4(struct qinst *inst)
 232 {
 233         switch (inst->op) {
 234         case QOP_TEX_RESULT:
 235         case QOP_TLB_COLOR_READ:
 236         case QOP_RCP:
 237         case QOP_RSQ:
 238         case QOP_EXP2:
 239         case QOP_LOG2:
 240                 return true;
 241         default:
 242                 return false;
 243         }
 244 }
 245
 246 static void
 247 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 248 {
 249         static const char *files[] = {
 250                 [QFILE_TEMP] = "t",
 251                 [QFILE_VARY] = "v",
 252                 [QFILE_UNIF] = "u",
 253         };
 254
 255         if (reg.file == QFILE_NULL) {
 256                 fprintf(stderr, "null");
 257         } else if (reg.file == QFILE_SMALL_IMM) {
 258                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 259                         fprintf(stderr, "%d", reg.index);
 260                 else
 261                         fprintf(stderr, "%f", uif(reg.index));
 262         } else if (reg.file == QFILE_VPM) {
 263                 if (write) {
 264                         fprintf(stderr, "vpm");
 265                 } else {
 266                         fprintf(stderr, "vpm%d.%d",
 267                                 reg.index / 4, reg.index % 4);
 268                 }
 269         } else {
 270                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 271         }
 272
 273         if (reg.file == QFILE_UNIF &&
 274             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 275                 fprintf(stderr, " (0x%08x / %f)",
 276                         c->uniform_data[reg.index],
 277                         uif(c->uniform_data[reg.index]));
 278         }
 279 }
 280
 281 void
 282 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 283 {
 284         fprintf(stderr, "%s%s ",
 285                 qir_get_op_name(inst->op),
 286                 inst->sf ? ".sf" : "");
 287
 288         qir_print_reg(c, inst->dst, true);
 289         if (inst->dst.pack) {
 290                 if (inst->dst.pack) {
 291                         if (qir_is_mul(inst))
 292                                 vc4_qpu_disasm_pack_mul(stderr, inst->dst.pack);
 293                         else
 294                                 vc4_qpu_disasm_pack_a(stderr, inst->dst.pack);
 295                 }
 296         }
 297         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 298                 fprintf(stderr, ", ");
 299                 qir_print_reg(c, inst->src[i], false);
 300         }
 301 }
 302
 303 void
 304 qir_dump(struct vc4_compile *c)
 305 {
 306         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 307                 qir_dump_inst(c, inst);
 308                 fprintf(stderr, "\n");
 309         }
 310 }
 311
 312 struct qreg
 313 qir_get_temp(struct vc4_compile *c)
 314 {
 315         struct qreg reg;
 316
 317         reg.file = QFILE_TEMP;
 318         reg.index = c->num_temps++;
 319
 320         if (c->num_temps > c->defs_array_size) {
 321                 uint32_t old_size = c->defs_array_size;
 322                 c->defs_array_size = MAX2(old_size * 2, 16);
 323                 c->defs = reralloc(c, c->defs, struct qinst *,
 324                                    c->defs_array_size);
 325                 memset(&c->defs[old_size], 0,
 326                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 327         }
 328
 329         return reg;
 330 }
 331
 332 struct qinst *
 333 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 334 {
 335         struct qinst *inst = CALLOC_STRUCT(qinst);
 336
 337         inst->op = op;
 338         inst->dst = dst;
 339         inst->src = calloc(2, sizeof(inst->src[0]));
 340         inst->src[0] = src0;
 341         inst->src[1] = src1;
 342
 343         return inst;
 344 }
 345
 346 struct qinst *
 347 qir_inst4(enum qop op, struct qreg dst,
 348           struct qreg a,
 349           struct qreg b,
 350           struct qreg c,
 351           struct qreg d)
 352 {
 353         struct qinst *inst = CALLOC_STRUCT(qinst);
 354
 355         inst->op = op;
 356         inst->dst = dst;
 357         inst->src = calloc(4, sizeof(*inst->src));
 358         inst->src[0] = a;
 359         inst->src[1] = b;
 360         inst->src[2] = c;
 361         inst->src[3] = d;
 362
 363         return inst;
 364 }
 365
 366 void
 367 qir_emit(struct vc4_compile *c, struct qinst *inst)
 368 {
 369         if (inst->dst.file == QFILE_TEMP)
 370                 c->defs[inst->dst.index] = inst;
 371
 372         qir_emit_nodef(c, inst);
 373 }
 374
 375 bool
 376 qir_reg_equals(struct qreg a, struct qreg b)
 377 {
 378         return a.file == b.file && a.index == b.index;
 379 }
 380
 381 struct vc4_compile *
 382 qir_compile_init(void)
 383 {
 384         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 385
 386         list_inithead(&c->instructions);
 387
 388         c->output_position_index = -1;
 389         c->output_clipvertex_index = -1;
 390         c->output_color_index = -1;
 391         c->output_point_size_index = -1;
 392
 393         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 394                                             _mesa_key_pointer_equal);
 395
 396         return c;
 397 }
 398
 399 void
 400 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 401 {
 402         if (qinst->dst.file == QFILE_TEMP)
 403                 c->defs[qinst->dst.index] = NULL;
 404
 405         list_del(&qinst->link);
 406         free(qinst->src);
 407         free(qinst);
 408 }
 409
 410 struct qreg
 411 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 412 {
 413         while (reg.file == QFILE_TEMP &&
 414                c->defs[reg.index] &&
 415                c->defs[reg.index]->op == QOP_MOV) {
 416                 reg = c->defs[reg.index]->src[0];
 417         }
 418
 419         return reg;
 420 }
 421
 422 void
 423 qir_compile_destroy(struct vc4_compile *c)
 424 {
 425         while (!list_empty(&c->instructions)) {
 426                 struct qinst *qinst =
 427                         (struct qinst *)c->instructions.next;
 428                 qir_remove_instruction(c, qinst);
 429         }
 430
 431         ralloc_free(c);
 432 }
 433
 434 const char *
 435 qir_get_stage_name(enum qstage stage)
 436 {
 437         static const char *names[] = {
 438                 [QSTAGE_FRAG] = "FS",
 439                 [QSTAGE_VERT] = "VS",
 440                 [QSTAGE_COORD] = "CS",
 441         };
 442
 443         return names[stage];
 444 }
 445
 446 struct qreg
 447 qir_uniform(struct vc4_compile *c,
 448             enum quniform_contents contents,
 449             uint32_t data)
 450 {
 451         for (int i = 0; i < c->num_uniforms; i++) {
 452                 if (c->uniform_contents[i] == contents &&
 453                     c->uniform_data[i] == data) {
 454                         return (struct qreg) { QFILE_UNIF, i };
 455                 }
 456         }
 457
 458         uint32_t uniform = c->num_uniforms++;
 459         struct qreg u = { QFILE_UNIF, uniform };
 460
 461         if (uniform >= c->uniform_array_size) {
 462                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 463                                              c->uniform_array_size * 2);
 464
 465                 c->uniform_data = reralloc(c, c->uniform_data,
 466                                            uint32_t,
 467                                            c->uniform_array_size);
 468                 c->uniform_contents = reralloc(c, c->uniform_contents,
 469                                                enum quniform_contents,
 470                                                c->uniform_array_size);
 471         }
 472
 473         c->uniform_contents[uniform] = contents;
 474         c->uniform_data[uniform] = data;
 475
 476         return u;
 477 }
 478
 479 void
 480 qir_SF(struct vc4_compile *c, struct qreg src)
 481 {
 482         struct qinst *last_inst = NULL;
 483         if (!list_empty(&c->instructions))
 484                 last_inst = (struct qinst *)c->instructions.prev;
 485
 486         if (!last_inst ||
 487             last_inst->dst.file != src.file ||
 488             last_inst->dst.index != src.index ||
 489             qir_is_multi_instruction(last_inst)) {
 490                 src = qir_MOV(c, src);
 491                 last_inst = (struct qinst *)c->instructions.prev;
 492         }
 493         last_inst->sf = true;
 494 }
 495
 496 #define OPTPASS(func)                                                   \
 497         do {                                                            \
 498                 bool stage_progress = func(c);                          \
 499                 if (stage_progress) {                                   \
 500                         progress = true;                                \
 501                         if (print_opt_debug) {                          \
 502                                 fprintf(stderr,                         \
 503                                         "QIR opt pass %2d: %s progress\n", \
 504                                         pass, #func);                   \
 505                         }                                               \
 506                 }                                                       \
 507         } while (0)
 508
 509 void
 510 qir_optimize(struct vc4_compile *c)
 511 {
 512         bool print_opt_debug = false;
 513         int pass = 1;
 514
 515         while (true) {
 516                 bool progress = false;
 517
 518                 OPTPASS(qir_opt_algebraic);
 519                 OPTPASS(qir_opt_cse);
 520                 OPTPASS(qir_opt_constant_folding);
 521                 OPTPASS(qir_opt_copy_propagation);
 522                 OPTPASS(qir_opt_dead_code);
 523                 OPTPASS(qir_opt_small_immediates);
 524                 OPTPASS(qir_opt_vpm_writes);
 525
 526                 if (!progress)
 527                         break;
 528
 529                 pass++;
 530         }
 531 }