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_UNPACK_8A_F] = { "unpack_8a_f", 1, 1 },
 100         [QOP_UNPACK_8B_F] = { "unpack_8b_f", 1, 1 },
 101         [QOP_UNPACK_8C_F] = { "unpack_8c_f", 1, 1 },
 102         [QOP_UNPACK_8D_F] = { "unpack_8d_f", 1, 1 },
 103         [QOP_UNPACK_16A_F] = { "unpack_16a_f", 1, 1 },
 104         [QOP_UNPACK_16B_F] = { "unpack_16b_f", 1, 1 },
 105         [QOP_UNPACK_8A_I] = { "unpack_8a_i", 1, 1 },
 106         [QOP_UNPACK_8B_I] = { "unpack_8b_i", 1, 1 },
 107         [QOP_UNPACK_8C_I] = { "unpack_8c_i", 1, 1 },
 108         [QOP_UNPACK_8D_I] = { "unpack_8d_i", 1, 1 },
 109         [QOP_UNPACK_16A_I] = { "unpack_16a_i", 1, 1 },
 110         [QOP_UNPACK_16B_I] = { "unpack_16b_i", 1, 1 },
 111 };
 112
 113 static const char *
 114 qir_get_op_name(enum qop qop)
 115 {
 116         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 117                 return qir_op_info[qop].name;
 118         else
 119                 return "???";
 120 }
 121
 122 int
 123 qir_get_op_nsrc(enum qop qop)
 124 {
 125         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 126                 return qir_op_info[qop].nsrc;
 127         else
 128                 abort();
 129 }
 130
 131 /**
 132  * Returns whether the instruction has any side effects that must be
 133  * preserved.
 134  */
 135 bool
 136 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 137 {
 138         return qir_op_info[inst->op].has_side_effects;
 139 }
 140
 141 bool
 142 qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
 143 {
 144         /* We can dead-code eliminate varyings, because we only tell the VS
 145          * about the live ones at the end.  But we have to preserve the
 146          * point/line coordinates reads, because they're generated by
 147          * fixed-function hardware.
 148          */
 149         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 150                 if (inst->src[i].file == QFILE_VARY &&
 151                     c->input_semantics[inst->src[i].index].semantic == 0xff) {
 152                         return true;
 153                 }
 154
 155                 if (inst->src[i].file == QFILE_VPM)
 156                         return true;
 157         }
 158
 159         if (inst->dst.file == QFILE_VPM)
 160                 return true;
 161
 162         return false;
 163 }
 164
 165 bool
 166 qir_is_multi_instruction(struct qinst *inst)
 167 {
 168         return qir_op_info[inst->op].multi_instruction;
 169 }
 170
 171 bool
 172 qir_is_tex(struct qinst *inst)
 173 {
 174         return inst->op >= QOP_TEX_S && inst->op <= QOP_TEX_DIRECT;
 175 }
 176
 177 bool
 178 qir_depends_on_flags(struct qinst *inst)
 179 {
 180         switch (inst->op) {
 181         case QOP_SEL_X_0_NS:
 182         case QOP_SEL_X_0_NC:
 183         case QOP_SEL_X_0_ZS:
 184         case QOP_SEL_X_0_ZC:
 185         case QOP_SEL_X_Y_NS:
 186         case QOP_SEL_X_Y_NC:
 187         case QOP_SEL_X_Y_ZS:
 188         case QOP_SEL_X_Y_ZC:
 189                 return true;
 190         default:
 191                 return false;
 192         }
 193 }
 194
 195 bool
 196 qir_src_needs_a_file(struct qinst *inst)
 197 {
 198         switch (inst->op) {
 199         case QOP_UNPACK_8A_F:
 200         case QOP_UNPACK_8B_F:
 201         case QOP_UNPACK_8C_F:
 202         case QOP_UNPACK_8D_F:
 203         case QOP_UNPACK_16A_F:
 204         case QOP_UNPACK_16B_F:
 205         case QOP_UNPACK_8A_I:
 206         case QOP_UNPACK_8B_I:
 207         case QOP_UNPACK_8C_I:
 208         case QOP_UNPACK_8D_I:
 209         case QOP_UNPACK_16A_I:
 210         case QOP_UNPACK_16B_I:
 211                 return true;
 212         default:
 213                 return false;
 214         }
 215 }
 216
 217 bool
 218 qir_writes_r4(struct qinst *inst)
 219 {
 220         switch (inst->op) {
 221         case QOP_TEX_RESULT:
 222         case QOP_TLB_COLOR_READ:
 223         case QOP_RCP:
 224         case QOP_RSQ:
 225         case QOP_EXP2:
 226         case QOP_LOG2:
 227                 return true;
 228         default:
 229                 return false;
 230         }
 231 }
 232
 233 static void
 234 qir_print_reg(struct vc4_compile *c, struct qreg reg, bool write)
 235 {
 236         static const char *files[] = {
 237                 [QFILE_TEMP] = "t",
 238                 [QFILE_VARY] = "v",
 239                 [QFILE_UNIF] = "u",
 240         };
 241
 242         if (reg.file == QFILE_NULL) {
 243                 fprintf(stderr, "null");
 244         } else if (reg.file == QFILE_SMALL_IMM) {
 245                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 246                         fprintf(stderr, "%d", reg.index);
 247                 else
 248                         fprintf(stderr, "%f", uif(reg.index));
 249         } else if (reg.file == QFILE_VPM) {
 250                 if (write) {
 251                         fprintf(stderr, "vpm");
 252                 } else {
 253                         fprintf(stderr, "vpm%d.%d",
 254                                 reg.index / 4, reg.index % 4);
 255                 }
 256         } else {
 257                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 258         }
 259
 260         if (reg.file == QFILE_UNIF &&
 261             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 262                 fprintf(stderr, " (0x%08x / %f)",
 263                         c->uniform_data[reg.index],
 264                         uif(c->uniform_data[reg.index]));
 265         }
 266 }
 267
 268 void
 269 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 270 {
 271         fprintf(stderr, "%s%s ",
 272                 qir_get_op_name(inst->op),
 273                 inst->sf ? ".sf" : "");
 274
 275         qir_print_reg(c, inst->dst, true);
 276         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 277                 fprintf(stderr, ", ");
 278                 qir_print_reg(c, inst->src[i], false);
 279         }
 280 }
 281
 282 void
 283 qir_dump(struct vc4_compile *c)
 284 {
 285         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 286                 qir_dump_inst(c, inst);
 287                 fprintf(stderr, "\n");
 288         }
 289 }
 290
 291 struct qreg
 292 qir_get_temp(struct vc4_compile *c)
 293 {
 294         struct qreg reg;
 295
 296         reg.file = QFILE_TEMP;
 297         reg.index = c->num_temps++;
 298
 299         if (c->num_temps > c->defs_array_size) {
 300                 uint32_t old_size = c->defs_array_size;
 301                 c->defs_array_size = MAX2(old_size * 2, 16);
 302                 c->defs = reralloc(c, c->defs, struct qinst *,
 303                                    c->defs_array_size);
 304                 memset(&c->defs[old_size], 0,
 305                        sizeof(c->defs[0]) * (c->defs_array_size - old_size));
 306         }
 307
 308         return reg;
 309 }
 310
 311 struct qinst *
 312 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 313 {
 314         struct qinst *inst = CALLOC_STRUCT(qinst);
 315
 316         inst->op = op;
 317         inst->dst = dst;
 318         inst->src = calloc(2, sizeof(inst->src[0]));
 319         inst->src[0] = src0;
 320         inst->src[1] = src1;
 321
 322         return inst;
 323 }
 324
 325 struct qinst *
 326 qir_inst4(enum qop op, struct qreg dst,
 327           struct qreg a,
 328           struct qreg b,
 329           struct qreg c,
 330           struct qreg d)
 331 {
 332         struct qinst *inst = CALLOC_STRUCT(qinst);
 333
 334         inst->op = op;
 335         inst->dst = dst;
 336         inst->src = calloc(4, sizeof(*inst->src));
 337         inst->src[0] = a;
 338         inst->src[1] = b;
 339         inst->src[2] = c;
 340         inst->src[3] = d;
 341
 342         return inst;
 343 }
 344
 345 void
 346 qir_emit(struct vc4_compile *c, struct qinst *inst)
 347 {
 348         if (inst->dst.file == QFILE_TEMP)
 349                 c->defs[inst->dst.index] = inst;
 350
 351         list_addtail(&inst->link, &c->instructions);
 352 }
 353
 354 bool
 355 qir_reg_equals(struct qreg a, struct qreg b)
 356 {
 357         return a.file == b.file && a.index == b.index;
 358 }
 359
 360 struct vc4_compile *
 361 qir_compile_init(void)
 362 {
 363         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 364
 365         list_inithead(&c->instructions);
 366
 367         c->output_position_index = -1;
 368         c->output_clipvertex_index = -1;
 369         c->output_color_index = -1;
 370         c->output_point_size_index = -1;
 371
 372         c->def_ht = _mesa_hash_table_create(c, _mesa_hash_pointer,
 373                                             _mesa_key_pointer_equal);
 374
 375         return c;
 376 }
 377
 378 void
 379 qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst)
 380 {
 381         if (qinst->dst.file == QFILE_TEMP)
 382                 c->defs[qinst->dst.index] = NULL;
 383
 384         list_del(&qinst->link);
 385         free(qinst->src);
 386         free(qinst);
 387 }
 388
 389 struct qreg
 390 qir_follow_movs(struct vc4_compile *c, struct qreg reg)
 391 {
 392         while (reg.file == QFILE_TEMP && c->defs[reg.index]->op == QOP_MOV)
 393                 reg = c->defs[reg.index]->src[0];
 394
 395         return reg;
 396 }
 397
 398 void
 399 qir_compile_destroy(struct vc4_compile *c)
 400 {
 401         while (!list_empty(&c->instructions)) {
 402                 struct qinst *qinst =
 403                         (struct qinst *)c->instructions.next;
 404                 qir_remove_instruction(c, qinst);
 405         }
 406
 407         ralloc_free(c);
 408 }
 409
 410 const char *
 411 qir_get_stage_name(enum qstage stage)
 412 {
 413         static const char *names[] = {
 414                 [QSTAGE_FRAG] = "FS",
 415                 [QSTAGE_VERT] = "VS",
 416                 [QSTAGE_COORD] = "CS",
 417         };
 418
 419         return names[stage];
 420 }
 421
 422 struct qreg
 423 qir_uniform(struct vc4_compile *c,
 424             enum quniform_contents contents,
 425             uint32_t data)
 426 {
 427         for (int i = 0; i < c->num_uniforms; i++) {
 428                 if (c->uniform_contents[i] == contents &&
 429                     c->uniform_data[i] == data) {
 430                         return (struct qreg) { QFILE_UNIF, i };
 431                 }
 432         }
 433
 434         uint32_t uniform = c->num_uniforms++;
 435         struct qreg u = { QFILE_UNIF, uniform };
 436
 437         if (uniform >= c->uniform_array_size) {
 438                 c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
 439                                              c->uniform_array_size * 2);
 440
 441                 c->uniform_data = reralloc(c, c->uniform_data,
 442                                            uint32_t,
 443                                            c->uniform_array_size);
 444                 c->uniform_contents = reralloc(c, c->uniform_contents,
 445                                                enum quniform_contents,
 446                                                c->uniform_array_size);
 447         }
 448
 449         c->uniform_contents[uniform] = contents;
 450         c->uniform_data[uniform] = data;
 451
 452         return u;
 453 }
 454
 455 void
 456 qir_SF(struct vc4_compile *c, struct qreg src)
 457 {
 458         struct qinst *last_inst = NULL;
 459         if (!list_empty(&c->instructions))
 460                 last_inst = (struct qinst *)c->instructions.prev;
 461
 462         if (!last_inst ||
 463             last_inst->dst.file != src.file ||
 464             last_inst->dst.index != src.index ||
 465             qir_is_multi_instruction(last_inst)) {
 466                 src = qir_MOV(c, src);
 467                 last_inst = (struct qinst *)c->instructions.prev;
 468         }
 469         last_inst->sf = true;
 470 }
 471
 472 #define OPTPASS(func)                                                   \
 473         do {                                                            \
 474                 bool stage_progress = func(c);                          \
 475                 if (stage_progress) {                                   \
 476                         progress = true;                                \
 477                         if (print_opt_debug) {                          \
 478                                 fprintf(stderr,                         \
 479                                         "QIR opt pass %2d: %s progress\n", \
 480                                         pass, #func);                   \
 481                         }                                               \
 482                 }                                                       \
 483         } while (0)
 484
 485 void
 486 qir_optimize(struct vc4_compile *c)
 487 {
 488         bool print_opt_debug = false;
 489         int pass = 1;
 490
 491         while (true) {
 492                 bool progress = false;
 493
 494                 OPTPASS(qir_opt_algebraic);
 495                 OPTPASS(qir_opt_cse);
 496                 OPTPASS(qir_opt_constant_folding);
 497                 OPTPASS(qir_opt_copy_propagation);
 498                 OPTPASS(qir_opt_dead_code);
 499                 OPTPASS(qir_opt_small_immediates);
 500                 OPTPASS(qir_opt_vpm_writes);
 501
 502                 if (!progress)
 503                         break;
 504
 505                 pass++;
 506         }
 507 }