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/simple_list.h"
  26 #include "util/ralloc.h"
  27
  28 #include "vc4_qir.h"
  29 #include "vc4_qpu.h"
  30
  31 struct qir_op_info {
  32         const char *name;
  33         uint8_t ndst, nsrc;
  34         bool has_side_effects;
  35         bool multi_instruction;
  36 };
  37
  38 static const struct qir_op_info qir_op_info[] = {
  39         [QOP_MOV] = { "mov", 1, 1 },
  40         [QOP_FADD] = { "fadd", 1, 2 },
  41         [QOP_FSUB] = { "fsub", 1, 2 },
  42         [QOP_FMUL] = { "fmul", 1, 2 },
  43         [QOP_MUL24] = { "mul24", 1, 2 },
  44         [QOP_FMIN] = { "fmin", 1, 2 },
  45         [QOP_FMAX] = { "fmax", 1, 2 },
  46         [QOP_FMINABS] = { "fminabs", 1, 2 },
  47         [QOP_FMAXABS] = { "fmaxabs", 1, 2 },
  48         [QOP_FTOI] = { "ftoi", 1, 1 },
  49         [QOP_ITOF] = { "itof", 1, 1 },
  50         [QOP_ADD] = { "add", 1, 2 },
  51         [QOP_SUB] = { "sub", 1, 2 },
  52         [QOP_SHR] = { "shr", 1, 2 },
  53         [QOP_ASR] = { "asr", 1, 2 },
  54         [QOP_SHL] = { "shl", 1, 2 },
  55         [QOP_MIN] = { "min", 1, 2 },
  56         [QOP_MAX] = { "max", 1, 2 },
  57         [QOP_AND] = { "and", 1, 2 },
  58         [QOP_OR] = { "or", 1, 2 },
  59         [QOP_XOR] = { "xor", 1, 2 },
  60         [QOP_NOT] = { "not", 1, 1 },
  61
  62         [QOP_SEL_X_0_NS] = { "fsel_x_0_ns", 1, 1, false, true },
  63         [QOP_SEL_X_0_NC] = { "fsel_x_0_nc", 1, 1, false, true },
  64         [QOP_SEL_X_0_ZS] = { "fsel_x_0_zs", 1, 1, false, true },
  65         [QOP_SEL_X_0_ZC] = { "fsel_x_0_zc", 1, 1, false, true },
  66         [QOP_SEL_X_Y_NS] = { "fsel_x_y_ns", 1, 2, false, true },
  67         [QOP_SEL_X_Y_NC] = { "fsel_x_y_nc", 1, 2, false, true },
  68         [QOP_SEL_X_Y_ZS] = { "fsel_x_y_zs", 1, 2, false, true },
  69         [QOP_SEL_X_Y_ZC] = { "fsel_x_y_zc", 1, 2, false, true },
  70
  71         [QOP_RCP] = { "rcp", 1, 1, false, true },
  72         [QOP_RSQ] = { "rsq", 1, 1, false, true },
  73         [QOP_EXP2] = { "exp2", 1, 2, false, true },
  74         [QOP_LOG2] = { "log2", 1, 2, false, true },
  75         [QOP_PACK_8888_F] = { "pack_8888_f", 1, 1, false, true },
  76         [QOP_PACK_8A_F] = { "pack_8a_f", 1, 2, false, true },
  77         [QOP_PACK_8B_F] = { "pack_8b_f", 1, 2, false, true },
  78         [QOP_PACK_8C_F] = { "pack_8c_f", 1, 2, false, true },
  79         [QOP_PACK_8D_F] = { "pack_8d_f", 1, 2, false, true },
  80         [QOP_PACK_SCALED] = { "pack_scaled", 1, 2, false, true },
  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_R4_UNPACK_A] = { "r4_unpack_a", 1, 1 },
 101         [QOP_R4_UNPACK_B] = { "r4_unpack_b", 1, 1 },
 102         [QOP_R4_UNPACK_C] = { "r4_unpack_c", 1, 1 },
 103         [QOP_R4_UNPACK_D] = { "r4_unpack_d", 1, 1 },
 104         [QOP_UNPACK_8A_F] = { "unpack_8a_f", 1, 1 },
 105         [QOP_UNPACK_8B_F] = { "unpack_8b_f", 1, 1 },
 106         [QOP_UNPACK_8C_F] = { "unpack_8c_f", 1, 1 },
 107         [QOP_UNPACK_8D_F] = { "unpack_8d_f", 1, 1 },
 108         [QOP_UNPACK_16A_F] = { "unpack_16a_f", 1, 1 },
 109         [QOP_UNPACK_16B_F] = { "unpack_16b_f", 1, 1 },
 110         [QOP_UNPACK_8A_I] = { "unpack_8a_i", 1, 1 },
 111         [QOP_UNPACK_8B_I] = { "unpack_8b_i", 1, 1 },
 112         [QOP_UNPACK_8C_I] = { "unpack_8c_i", 1, 1 },
 113         [QOP_UNPACK_8D_I] = { "unpack_8d_i", 1, 1 },
 114         [QOP_UNPACK_16A_I] = { "unpack_16a_i", 1, 1 },
 115         [QOP_UNPACK_16B_I] = { "unpack_16b_i", 1, 1 },
 116 };
 117
 118 static const char *
 119 qir_get_op_name(enum qop qop)
 120 {
 121         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 122                 return qir_op_info[qop].name;
 123         else
 124                 return "???";
 125 }
 126
 127 int
 128 qir_get_op_nsrc(enum qop qop)
 129 {
 130         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 131                 return qir_op_info[qop].nsrc;
 132         else
 133                 abort();
 134 }
 135
 136 /**
 137  * Returns whether the instruction has any side effects that must be
 138  * preserved.
 139  */
 140 bool
 141 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 142 {
 143         return qir_op_info[inst->op].has_side_effects;
 144 }
 145
 146 bool
 147 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 148 {
 149         /* We can dead-code eliminate varyings, because we only tell the VS
 150          * about the live ones at the end.  But we have to preserve the
 151          * point/line coordinates reads, because they're generated by
 152          * fixed-function hardware.
 153          */
 154         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 155                 if (inst->src[i].file == QFILE_VARY &&
 156                     c->input_semantics[inst->src[i].index].semantic == 0xff) {
 157                         return true;
 158                 }
 159
 160                 if (inst->src[i].file == QFILE_VPM)
 161                         return true;
 162         }
 163
 164         if (inst->dst.file == QFILE_VPM)
 165                 return true;
 166
 167         return false;
 168 }
 169
 170 bool
 171 qir_is_multi_instruction(struct qinst *inst)
 172 {
 173         return qir_op_info[inst->op].multi_instruction;
 174 }
 175
 176 bool
 177 qir_is_tex(struct qinst *inst)
 178 {
 179         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 180 }
 181
 182 bool
 183 qir_depends_on_flags(struct qinst *inst)
 184 {
 185         switch (inst->op) {
 186         case QOP_SEL_X_0_NS:
 187         case QOP_SEL_X_0_NC:
 188         case QOP_SEL_X_0_ZS:
 189         case QOP_SEL_X_0_ZC:
 190         case QOP_SEL_X_Y_NS:
 191         case QOP_SEL_X_Y_NC:
 192         case QOP_SEL_X_Y_ZS:
 193         case QOP_SEL_X_Y_ZC:
 194                 return true;
 195         default:
 196                 return false;
 197         }
 198 }
 199
 200 bool
 201 qir_src_needs_a_file(struct qinst *inst)
 202 {
 203         switch (inst->op) {
 204         case QOP_UNPACK_8A_F:
 205         case QOP_UNPACK_8B_F:
 206         case QOP_UNPACK_8C_F:
 207         case QOP_UNPACK_8D_F:
 208         case QOP_UNPACK_16A_F:
 209         case QOP_UNPACK_16B_F:
 210         case QOP_UNPACK_8A_I:
 211         case QOP_UNPACK_8B_I:
 212         case QOP_UNPACK_8C_I:
 213         case QOP_UNPACK_8D_I:
 214         case QOP_UNPACK_16A_I:
 215         case QOP_UNPACK_16B_I:
 216                 return true;
 217         default:
 218                 return false;
 219         }
 220 }
 221
 222 bool
 223 qir_writes_r4(struct qinst *inst)
 224 {
 225         switch (inst->op) {
 226         case QOP_TEX_RESULT:
 227         case QOP_TLB_COLOR_READ:
 228         case QOP_RCP:
 229         case QOP_RSQ:
 230         case QOP_EXP2:
 231         case QOP_LOG2:
 232                 return true;
 233         default:
 234                 return false;
 235         }
 236 }
 237
 238 bool
 239 qir_reads_r4(struct qinst *inst)
 240 {
 241         switch (inst->op) {
 242         case QOP_R4_UNPACK_A:
 243         case QOP_R4_UNPACK_B:
 244         case QOP_R4_UNPACK_C:
 245         case QOP_R4_UNPACK_D:
 246                 return true;
 247         default:
 248                 return false;
 249         }
 250 }
 251
 252 static void
 253 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 254 {
 255         static const char *files[] = {
 256                 [QFILE_TEMP] = "t",
 257                 [QFILE_VARY] = "v",
 258                 [QFILE_UNIF] = "u",
 259         };
 260
 261         if (reg.file == QFILE_NULL) {
 262                 fprintf(stderr, "null");
 263         } else if (reg.file == QFILE_SMALL_IMM) {
 264                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 265                         fprintf(stderr, "%d", reg.index);
 266                 else
 267                         fprintf(stderr, "%f", uif(reg.index));
 268         } else if (reg.file == QFILE_VPM) {
 269                 if (write) {
 270                         fprintf(stderr, "vpm");
 271                 } else {
 272                         fprintf(stderr, "vpm%d.%d",
 273                                 reg.index / 4, reg.index % 4);
 274                 }
 275         } else {
 276                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 277         }
 278
 279         if (reg.file == QFILE_UNIF &&
 280             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 281                 fprintf(stderr, " (0x%08x / %f)",
 282                         c->uniform_data[reg.index],
 283                         uif(c->uniform_data[reg.index]));
 284         }
 285 }
 286
 287 void
 288 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 289 {
 290         fprintf(stderr, "%s%s ",
 291                 qir_get_op_name(inst->op),
 292                 inst->sf ? ".sf" : "");
 293
 294         qir_print_reg(c, inst->dst, true);
 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         struct simple_node *node;
 305
 306         foreach(node, &c->instructions) {
 307                 struct qinst *inst = (struct qinst *)node;
 308                 qir_dump_inst(c, inst);
 309                 fprintf(stderr, "\n");
 310         }
 311 }
 312
 313 struct qreg
 314 qir_get_temp(struct vc4_compile *c)
 315 {
 316         struct qreg reg;
 317
 318         reg.file = QFILE_TEMP;
 319         reg.index = c->num_temps++;
 320
 321         return reg;
 322 }
 323
 324 struct qinst *
 325 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 326 {
 327         struct qinst *inst = CALLOC_STRUCT(qinst);
 328
 329         inst->op = op;
 330         inst->dst = dst;
 331         inst->src = calloc(2, sizeof(inst->src[0]));
 332         inst->src[0] = src0;
 333         inst->src[1] = src1;
 334
 335         return inst;
 336 }
 337
 338 struct qinst *
 339 qir_inst4(enum qop op, struct qreg dst,
 340           struct qreg a,
 341           struct qreg b,
 342           struct qreg c,
 343           struct qreg d)
 344 {
 345         struct qinst *inst = CALLOC_STRUCT(qinst);
 346
 347         inst->op = op;
 348         inst->dst = dst;
 349         inst->src = calloc(4, sizeof(*inst->src));
 350         inst->src[0] = a;
 351         inst->src[1] = b;
 352         inst->src[2] = c;
 353         inst->src[3] = d;
 354
 355         return inst;
 356 }
 357
 358 void
 359 qir_emit(struct vc4_compile *c, struct qinst *inst)
 360 {
 361         insert_at_tail(&c->instructions, &inst->link);
 362 }
 363
 364 bool
 365 qir_reg_equals(struct qreg a, struct qreg b)
 366 {
 367         return a.file == b.file && a.index == b.index;
 368 }
 369
 370 struct vc4_compile *
 371 qir_compile_init(void)
 372 {
 373         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 374
 375         make_empty_list(&c->instructions);
 376
 377         c->output_position_index = -1;
 378         c->output_clipvertex_index = -1;
 379         c->output_color_index = -1;
 380         c->output_point_size_index = -1;
 381
 382         return c;
 383 }
 384
 385 void
 386 qir_remove_instruction(struct qinst *qinst)
 387 {
 388         remove_from_list(&qinst->link);
 389         free(qinst->src);
 390         free(qinst);
 391 }
 392
 393 struct qreg
 394 qir_follow_movs(struct qinst **defs, struct qreg reg)
 395 {
 396         while (reg.file == QFILE_TEMP && defs[reg.index]->op == QOP_MOV)
 397                 reg = defs[reg.index]->src[0];
 398
 399         return reg;
 400 }
 401
 402 void
 403 qir_compile_destroy(struct vc4_compile *c)
 404 {
 405         while (!is_empty_list(&c->instructions)) {
 406                 struct qinst *qinst =
 407                         (struct qinst *)first_elem(&c->instructions);
 408                 qir_remove_instruction(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 (!is_empty_list(&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_copy_propagation);
 501                 OPTPASS(qir_opt_dead_code);
 502                 OPTPASS(qir_opt_small_immediates);
 503                 OPTPASS(qir_opt_vpm_writes);
 504
 505                 if (!progress)
 506                         break;
 507
 508                 pass++;
 509         }
 510 }