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_tex(struct qinst *inst)
 174 {
 175         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 176 }
 177
 178 bool
 179 qir_depends_on_flags(struct qinst *inst)
 180 {
 181         switch (inst->op) {
 182         case QOP_SEL_X_0_NS:
 183         case QOP_SEL_X_0_NC:
 184         case QOP_SEL_X_0_ZS:
 185         case QOP_SEL_X_0_ZC:
 186         case QOP_SEL_X_Y_NS:
 187         case QOP_SEL_X_Y_NC:
 188         case QOP_SEL_X_Y_ZS:
 189         case QOP_SEL_X_Y_ZC:
 190                 return true;
 191         default:
 192                 return false;
 193         }
 194 }
 195
 196 bool
 197 qir_src_needs_a_file(struct qinst *inst)
 198 {
 199         switch (inst->op) {
 200         case QOP_UNPACK_8A_F:
 201         case QOP_UNPACK_8B_F:
 202         case QOP_UNPACK_8C_F:
 203         case QOP_UNPACK_8D_F:
 204         case QOP_UNPACK_16A_F:
 205         case QOP_UNPACK_16B_F:
 206         case QOP_UNPACK_8A_I:
 207         case QOP_UNPACK_8B_I:
 208         case QOP_UNPACK_8C_I:
 209         case QOP_UNPACK_8D_I:
 210         case QOP_UNPACK_16A_I:
 211         case QOP_UNPACK_16B_I:
 212                 return true;
 213         default:
 214                 return false;
 215         }
 216 }
 217
 218 bool
 219 qir_writes_r4(struct qinst *inst)
 220 {
 221         switch (inst->op) {
 222         case QOP_TEX_RESULT:
 223         case QOP_TLB_COLOR_READ:
 224         case QOP_RCP:
 225         case QOP_RSQ:
 226         case QOP_EXP2:
 227         case QOP_LOG2:
 228                 return true;
 229         default:
 230                 return false;
 231         }
 232 }
 233
 234 static void
 235 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 236 {
 237         static const char *files[] = {
 238                 [QFILE_TEMP] = "t",
 239                 [QFILE_VARY] = "v",
 240                 [QFILE_UNIF] = "u",
 241         };
 242
 243         if (reg.file == QFILE_NULL) {
 244                 fprintf(stderr, "null");
 245         } else if (reg.file == QFILE_SMALL_IMM) {
 246                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 247                         fprintf(stderr, "%d", reg.index);
 248                 else
 249                         fprintf(stderr, "%f", uif(reg.index));
 250         } else if (reg.file == QFILE_VPM) {
 251                 if (write) {
 252                         fprintf(stderr, "vpm");
 253                 } else {
 254                         fprintf(stderr, "vpm%d.%d",
 255                                 reg.index / 4, reg.index % 4);
 256                 }
 257         } else {
 258                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 259         }
 260
 261         if (reg.file == QFILE_UNIF &&
 262             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 263                 fprintf(stderr, " (0x%08x / %f)",
 264                         c->uniform_data[reg.index],
 265                         uif(c->uniform_data[reg.index]));
 266         }
 267 }
 268
 269 void
 270 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 271 {
 272         fprintf(stderr, "%s%s ",
 273                 qir_get_op_name(inst->op),
 274                 inst->sf ? ".sf" : "");
 275
 276         qir_print_reg(c, inst->dst, true);
 277         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 278                 fprintf(stderr, ", ");
 279                 qir_print_reg(c, inst->src[i], false);
 280         }
 281 }
 282
 283 void
 284 qir_dump(struct vc4_compile *c)
 285 {
 286         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 287                 qir_dump_inst(c, inst);
 288                 fprintf(stderr, "\n");
 289         }
 290 }
 291
 292 struct qreg
 293 qir_get_temp(struct vc4_compile *c)
 294 {
 295         struct qreg reg;
 296
 297         reg.file = QFILE_TEMP;
 298         reg.index = c->num_temps++;
 299
 300         if (c->num_temps > c->defs_array_size) {
 301                 uint32_t old_size = c->defs_array_size;
 302                 c->defs_array_size = MAX2(old_size * 2, 16);
 303                 c->defs = reralloc(c, c->defs, struct qinst *,
 304                                    c->defs_array_size);
 305                 memset(&c->defs[old_size], 0,
 306                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 307         }
 308
 309         return reg;
 310 }
 311
 312 struct qinst *
 313 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 314 {
 315         struct qinst *inst = CALLOC_STRUCT(qinst);
 316
 317         inst->op = op;
 318         inst->dst = dst;
 319         inst->src = calloc(2, sizeof(inst->src[0]));
 320         inst->src[0] = src0;
 321         inst->src[1] = src1;
 322
 323         return inst;
 324 }
 325
 326 struct qinst *
 327 qir_inst4(enum qop op, struct qreg dst,
 328           struct qreg a,
 329           struct qreg b,
 330           struct qreg c,
 331           struct qreg d)
 332 {
 333         struct qinst *inst = CALLOC_STRUCT(qinst);
 334
 335         inst->op = op;
 336         inst->dst = dst;
 337         inst->src = calloc(4, sizeof(*inst->src));
 338         inst->src[0] = a;
 339         inst->src[1] = b;
 340         inst->src[2] = c;
 341         inst->src[3] = d;
 342
 343         return inst;
 344 }
 345
 346 void
 347 qir_emit(struct vc4_compile *c, struct qinst *inst)
 348 {
 349         if (inst->dst.file == QFILE_TEMP)
 350                 c->defs[inst->dst.index] = inst;
 351
 352         list_addtail(&inst->link, &c->instructions);
 353 }
 354
 355 bool
 356 qir_reg_equals(struct qreg a, struct qreg b)
 357 {
 358         return a.file == b.file && a.index == b.index;
 359 }
 360
 361 struct vc4_compile *
 362 qir_compile_init(void)
 363 {
 364         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 365
 366         list_inithead(&c->instructions);
 367
 368         c->output_position_index = -1;
 369         c->output_clipvertex_index = -1;
 370         c->output_color_index = -1;
 371         c->output_point_size_index = -1;
 372
 373         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 374                                             _mesa_key_pointer_equal);
 375
 376         return c;
 377 }
 378
 379 void
 380 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 381 {
 382         if (qinst->dst.file == QFILE_TEMP)
 383                 c->defs[qinst->dst.index] = NULL;
 384
 385         list_del(&qinst->link);
 386         free(qinst->src);
 387         free(qinst);
 388 }
 389
 390 struct qreg
 391 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 392 {
 393         while (reg.file == QFILE_TEMP &&
 394                c->defs[reg.index] &&
 395                c->defs[reg.index]->op == QOP_MOV) {
 396                 reg = c->defs[reg.index]->src[0];
 397         }
 398
 399         return reg;
 400 }
 401
 402 void
 403 qir_compile_destroy(struct vc4_compile *c)
 404 {
 405         while (!list_empty(&c->instructions)) {
 406                 struct qinst *qinst =
 407                         (struct qinst *)c->instructions.next;
 408                 qir_remove_instruction(c, qinst);
 409         }
 410
 411         ralloc_free(c);
 412 }
 413
 414 const char *
 415 qir_get_stage_name(enum qstage stage)
 416 {
 417         static const char *names[] = {
 418                 [QSTAGE_FRAG] = "FS",
 419                 [QSTAGE_VERT] = "VS",
 420                 [QSTAGE_COORD] = "CS",
 421         };
 422
 423         return names[stage];
 424 }
 425
 426 struct qreg
 427 qir_uniform(struct vc4_compile *c,
 428             enum quniform_contents contents,
 429             uint32_t data)
 430 {
 431         for (int i = 0; i < c->num_uniforms; i++) {
 432                 if (c->uniform_contents[i] == contents &&
 433                     c->uniform_data[i] == data) {
 434                         return (struct qreg) { QFILE_UNIF, i };
 435                 }
 436         }
 437
 438         uint32_t uniform = c->num_uniforms++;
 439         struct qreg u = { QFILE_UNIF, uniform };
 440
 441         if (uniform >= c->uniform_array_size) {
 442                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 443                                              c->uniform_array_size * 2);
 444
 445                 c->uniform_data = reralloc(c, c->uniform_data,
 446                                            uint32_t,
 447                                            c->uniform_array_size);
 448                 c->uniform_contents = reralloc(c, c->uniform_contents,
 449                                                enum quniform_contents,
 450                                                c->uniform_array_size);
 451         }
 452
 453         c->uniform_contents[uniform] = contents;
 454         c->uniform_data[uniform] = data;
 455
 456         return u;
 457 }
 458
 459 void
 460 qir_SF(struct vc4_compile *c, struct qreg src)
 461 {
 462         struct qinst *last_inst = NULL;
 463         if (!list_empty(&c->instructions))
 464                 last_inst = (struct qinst *)c->instructions.prev;
 465
 466         if (!last_inst ||
 467             last_inst->dst.file != src.file ||
 468             last_inst->dst.index != src.index ||
 469             qir_is_multi_instruction(last_inst)) {
 470                 src = qir_MOV(c, src);
 471                 last_inst = (struct qinst *)c->instructions.prev;
 472         }
 473         last_inst->sf = true;
 474 }
 475
 476 #define OPTPASS(func)                                                   \
 477         do {                                                            \
 478                 bool stage_progress = func(c);                          \
 479                 if (stage_progress) {                                   \
 480                         progress = true;                                \
 481                         if (print_opt_debug) {                          \
 482                                 fprintf(stderr,                         \
 483                                         "QIR opt pass %2d: %s progress\n", \
 484                                         pass, #func);                   \
 485                         }                                               \
 486                 }                                                       \
 487         } while (0)
 488
 489 void
 490 qir_optimize(struct vc4_compile *c)
 491 {
 492         bool print_opt_debug = false;
 493         int pass = 1;
 494
 495         while (true) {
 496                 bool progress = false;
 497
 498                 OPTPASS(qir_opt_algebraic);
 499                 OPTPASS(qir_opt_cse);
 500                 OPTPASS(qir_opt_constant_folding);
 501                 OPTPASS(qir_opt_copy_propagation);
 502                 OPTPASS(qir_opt_dead_code);
 503                 OPTPASS(qir_opt_small_immediates);
 504                 OPTPASS(qir_opt_vpm_writes);
 505
 506                 if (!progress)
 507                         break;
 508
 509                 pass++;
 510         }
 511 }