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_depends_on_flags(struct qinst *inst)
 178 {
 179         switch (inst->op) {
 180         case QOP_SEL_X_0_NS:
 181         case QOP_SEL_X_0_NC:
 182         case QOP_SEL_X_0_ZS:
 183         case QOP_SEL_X_0_ZC:
 184         case QOP_SEL_X_Y_NS:
 185         case QOP_SEL_X_Y_NC:
 186         case QOP_SEL_X_Y_ZS:
 187         case QOP_SEL_X_Y_ZC:
 188                 return true;
 189         default:
 190                 return false;
 191         }
 192 }
 193
 194 bool
 195 qir_src_needs_a_file(struct qinst *inst)
 196 {
 197         switch (inst->op) {
 198         case QOP_UNPACK_8A_F:
 199         case QOP_UNPACK_8B_F:
 200         case QOP_UNPACK_8C_F:
 201         case QOP_UNPACK_8D_F:
 202         case QOP_UNPACK_16A_F:
 203         case QOP_UNPACK_16B_F:
 204         case QOP_UNPACK_8A_I:
 205         case QOP_UNPACK_8B_I:
 206         case QOP_UNPACK_8C_I:
 207         case QOP_UNPACK_8D_I:
 208         case QOP_UNPACK_16A_I:
 209         case QOP_UNPACK_16B_I:
 210                 return true;
 211         default:
 212                 return false;
 213         }
 214 }
 215
 216 bool
 217 qir_writes_r4(struct qinst *inst)
 218 {
 219         switch (inst->op) {
 220         case QOP_TEX_RESULT:
 221         case QOP_TLB_COLOR_READ:
 222         case QOP_RCP:
 223         case QOP_RSQ:
 224         case QOP_EXP2:
 225         case QOP_LOG2:
 226                 return true;
 227         default:
 228                 return false;
 229         }
 230 }
 231
 232 bool
 233 qir_reads_r4(struct qinst *inst)
 234 {
 235         switch (inst->op) {
 236         case QOP_R4_UNPACK_A:
 237         case QOP_R4_UNPACK_B:
 238         case QOP_R4_UNPACK_C:
 239         case QOP_R4_UNPACK_D:
 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         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 290                 fprintf(stderr, ", ");
 291                 qir_print_reg(c, inst->src[i], false);
 292         }
 293 }
 294
 295 void
 296 qir_dump(struct vc4_compile *c)
 297 {
 298         struct simple_node *node;
 299
 300         foreach(node, &c->instructions) {
 301                 struct qinst *inst = (struct qinst *)node;
 302                 qir_dump_inst(c, inst);
 303                 fprintf(stderr, "\n");
 304         }
 305 }
 306
 307 struct qreg
 308 qir_get_temp(struct vc4_compile *c)
 309 {
 310         struct qreg reg;
 311
 312         reg.file = QFILE_TEMP;
 313         reg.index = c->num_temps++;
 314
 315         return reg;
 316 }
 317
 318 struct qinst *
 319 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 320 {
 321         struct qinst *inst = CALLOC_STRUCT(qinst);
 322
 323         inst->op = op;
 324         inst->dst = dst;
 325         inst->src = calloc(2, sizeof(inst->src[0]));
 326         inst->src[0] = src0;
 327         inst->src[1] = src1;
 328
 329         return inst;
 330 }
 331
 332 struct qinst *
 333 qir_inst4(enum qop op, struct qreg dst,
 334           struct qreg a,
 335           struct qreg b,
 336           struct qreg c,
 337           struct qreg d)
 338 {
 339         struct qinst *inst = CALLOC_STRUCT(qinst);
 340
 341         inst->op = op;
 342         inst->dst = dst;
 343         inst->src = calloc(4, sizeof(*inst->src));
 344         inst->src[0] = a;
 345         inst->src[1] = b;
 346         inst->src[2] = c;
 347         inst->src[3] = d;
 348
 349         return inst;
 350 }
 351
 352 void
 353 qir_emit(struct vc4_compile *c, struct qinst *inst)
 354 {
 355         insert_at_tail(&c->instructions, &inst->link);
 356 }
 357
 358 bool
 359 qir_reg_equals(struct qreg a, struct qreg b)
 360 {
 361         return a.file == b.file && a.index == b.index;
 362 }
 363
 364 struct vc4_compile *
 365 qir_compile_init(void)
 366 {
 367         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 368
 369         make_empty_list(&c->instructions);
 370
 371         c->output_position_index = -1;
 372         c->output_clipvertex_index = -1;
 373         c->output_color_index = -1;
 374         c->output_point_size_index = -1;
 375
 376         return c;
 377 }
 378
 379 void
 380 qir_remove_instruction(struct qinst *qinst)
 381 {
 382         remove_from_list(&qinst->link);
 383         free(qinst->src);
 384         free(qinst);
 385 }
 386
 387 struct qreg
 388 qir_follow_movs(struct qinst **defs, struct qreg reg)
 389 {
 390         while (reg.file == QFILE_TEMP && defs[reg.index]->op == QOP_MOV)
 391                 reg = defs[reg.index]->src[0];
 392
 393         return reg;
 394 }
 395
 396 void
 397 qir_compile_destroy(struct vc4_compile *c)
 398 {
 399         while (!is_empty_list(&c->instructions)) {
 400                 struct qinst *qinst =
 401                         (struct qinst *)first_elem(&c->instructions);
 402                 qir_remove_instruction(qinst);
 403         }
 404
 405         ralloc_free(c);
 406 }
 407
 408 const char *
 409 qir_get_stage_name(enum qstage stage)
 410 {
 411         static const char *names[] = {
 412                 [QSTAGE_FRAG] = "FS",
 413                 [QSTAGE_VERT] = "VS",
 414                 [QSTAGE_COORD] = "CS",
 415         };
 416
 417         return names[stage];
 418 }
 419
 420 void
 421 qir_SF(struct vc4_compile *c, struct qreg src)
 422 {
 423         assert(!is_empty_list(&c->instructions));
 424         struct qinst *last_inst = (struct qinst *)c->instructions.prev;
 425         if (last_inst->dst.file != src.file ||
 426             last_inst->dst.index != src.index ||
 427             qir_is_multi_instruction(last_inst)) {
 428                 src = qir_MOV(c, src);
 429                 last_inst = (struct qinst *)c->instructions.prev;
 430         }
 431         last_inst->sf = true;
 432 }
 433
 434 #define OPTPASS(func)                                                   \
 435         do {                                                            \
 436                 bool stage_progress = func(c);                          \
 437                 if (stage_progress) {                                   \
 438                         progress = true;                                \
 439                         if (print_opt_debug) {                          \
 440                                 fprintf(stderr,                         \
 441                                         "QIR opt pass %2d: %s progress\n", \
 442                                         pass, #func);                   \
 443                         }                                               \
 444                 }                                                       \
 445         } while (0)
 446
 447 void
 448 qir_optimize(struct vc4_compile *c)
 449 {
 450         bool print_opt_debug = false;
 451         int pass = 1;
 452
 453         while (true) {
 454                 bool progress = false;
 455
 456                 OPTPASS(qir_opt_algebraic);
 457                 OPTPASS(qir_opt_cse);
 458                 OPTPASS(qir_opt_copy_propagation);
 459                 OPTPASS(qir_opt_dead_code);
 460                 OPTPASS(qir_opt_small_immediates);
 461                 OPTPASS(qir_opt_vpm_writes);
 462
 463                 if (!progress)
 464                         break;
 465
 466                 pass++;
 467         }
 468 }