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/u_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_SF] = { "sf", 0, 1 },
  63         [QOP_SEL_X_0_NS] = { "fsel_x_0_ns", 1, 1, false, true },
  64         [QOP_SEL_X_0_NC] = { "fsel_x_0_nc", 1, 1, false, true },
  65         [QOP_SEL_X_0_ZS] = { "fsel_x_0_zs", 1, 1, false, true },
  66         [QOP_SEL_X_0_ZC] = { "fsel_x_0_zc", 1, 1, false, true },
  67         [QOP_SEL_X_Y_NS] = { "fsel_x_y_ns", 1, 2, false, true },
  68         [QOP_SEL_X_Y_NC] = { "fsel_x_y_nc", 1, 2, false, true },
  69         [QOP_SEL_X_Y_ZS] = { "fsel_x_y_zs", 1, 2, false, true },
  70         [QOP_SEL_X_Y_ZC] = { "fsel_x_y_zc", 1, 2, false, true },
  71
  72         [QOP_RCP] = { "rcp", 1, 1, false, true },
  73         [QOP_RSQ] = { "rsq", 1, 1, false, true },
  74         [QOP_EXP2] = { "exp2", 1, 2, false, true },
  75         [QOP_LOG2] = { "log2", 1, 2, false, true },
  76         [QOP_PACK_COLORS] = { "pack_colors", 1, 4, false, true },
  77         [QOP_PACK_SCALED] = { "pack_scaled", 1, 2, false, true },
  78         [QOP_VPM_READ] = { "vpm_read", 0, 1, true },
  79         [QOP_TLB_DISCARD_SETUP] = { "discard", 0, 1, true },
  80         [QOP_TLB_STENCIL_SETUP] = { "tlb_stencil_setup", 0, 1, true },
  81         [QOP_TLB_Z_WRITE] = { "tlb_z", 0, 1, true },
  82         [QOP_TLB_COLOR_WRITE] = { "tlb_color", 0, 1, true },
  83         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0 },
  84         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  85
  86         [QOP_FRAG_X] = { "frag_x", 1, 0 },
  87         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  88         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  89         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  90         [QOP_FRAG_REV_FLAG] = { "frag_rev_flag", 1, 0 },
  91
  92         [QOP_TEX_S] = { "tex_s", 0, 2 },
  93         [QOP_TEX_T] = { "tex_t", 0, 2 },
  94         [QOP_TEX_R] = { "tex_r", 0, 2 },
  95         [QOP_TEX_B] = { "tex_b", 0, 2 },
  96         [QOP_TEX_DIRECT] = { "tex_direct", 0, 2 },
  97         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
  98         [QOP_R4_UNPACK_A] = { "r4_unpack_a", 1, 1 },
  99         [QOP_R4_UNPACK_B] = { "r4_unpack_b", 1, 1 },
 100         [QOP_R4_UNPACK_C] = { "r4_unpack_c", 1, 1 },
 101         [QOP_R4_UNPACK_D] = { "r4_unpack_d", 1, 1 },
 102         [QOP_UNPACK_8A_F] = { "unpack_8a_f", 1, 1 },
 103         [QOP_UNPACK_8B_F] = { "unpack_8b_f", 1, 1 },
 104         [QOP_UNPACK_8C_F] = { "unpack_8c_f", 1, 1 },
 105         [QOP_UNPACK_8D_F] = { "unpack_8d_f", 1, 1 },
 106         [QOP_UNPACK_16A_F] = { "unpack_16a_f", 1, 1 },
 107         [QOP_UNPACK_16B_F] = { "unpack_16b_f", 1, 1 },
 108         [QOP_UNPACK_8A_I] = { "unpack_8a_i", 1, 1 },
 109         [QOP_UNPACK_8B_I] = { "unpack_8b_i", 1, 1 },
 110         [QOP_UNPACK_8C_I] = { "unpack_8c_i", 1, 1 },
 111         [QOP_UNPACK_8D_I] = { "unpack_8d_i", 1, 1 },
 112         [QOP_UNPACK_16A_I] = { "unpack_16a_i", 1, 1 },
 113         [QOP_UNPACK_16B_I] = { "unpack_16b_i", 1, 1 },
 114 };
 115
 116 static const char *
 117 qir_get_op_name(enum qop qop)
 118 {
 119         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 120                 return qir_op_info[qop].name;
 121         else
 122                 return "???";
 123 }
 124
 125 int
 126 qir_get_op_nsrc(enum qop qop)
 127 {
 128         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 129                 return qir_op_info[qop].nsrc;
 130         else
 131                 abort();
 132 }
 133
 134 /**
 135  * Returns whether the instruction has any side effects that must be
 136  * preserved.
 137  */
 138 bool
 139 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 140 {
 141         /* We can dead-code eliminate varyings, because we only tell the VS
 142          * about the live ones at the end.  But we have to preserve the
 143          * point/line coordinates reads, because they're generated by
 144          * fixed-function hardware.
 145          */
 146         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 147                 if (inst->src[i].file == QFILE_VARY &&
 148                     c->input_semantics[inst->src[i].index].semantic == 0xff) {
 149                         return true;
 150                 }
 151         }
 152
 153         if (inst->dst.file == QFILE_VPM)
 154                 return true;
 155
 156         return qir_op_info[inst->op].has_side_effects;
 157 }
 158
 159 bool
 160 qir_is_multi_instruction(struct qinst *inst)
 161 {
 162         return qir_op_info[inst->op].multi_instruction;
 163 }
 164
 165 bool
 166 qir_depends_on_flags(struct qinst *inst)
 167 {
 168         switch (inst->op) {
 169         case QOP_SEL_X_0_NS:
 170         case QOP_SEL_X_0_NC:
 171         case QOP_SEL_X_0_ZS:
 172         case QOP_SEL_X_0_ZC:
 173         case QOP_SEL_X_Y_NS:
 174         case QOP_SEL_X_Y_NC:
 175         case QOP_SEL_X_Y_ZS:
 176         case QOP_SEL_X_Y_ZC:
 177                 return true;
 178         default:
 179                 return false;
 180         }
 181 }
 182
 183 bool
 184 qir_writes_r4(struct qinst *inst)
 185 {
 186         switch (inst->op) {
 187         case QOP_TEX_RESULT:
 188         case QOP_TLB_COLOR_READ:
 189         case QOP_RCP:
 190         case QOP_RSQ:
 191         case QOP_EXP2:
 192         case QOP_LOG2:
 193                 return true;
 194         default:
 195                 return false;
 196         }
 197 }
 198
 199 bool
 200 qir_reads_r4(struct qinst *inst)
 201 {
 202         switch (inst->op) {
 203         case QOP_R4_UNPACK_A:
 204         case QOP_R4_UNPACK_B:
 205         case QOP_R4_UNPACK_C:
 206         case QOP_R4_UNPACK_D:
 207                 return true;
 208         default:
 209                 return false;
 210         }
 211 }
 212
 213 static void
 214 qir_print_reg(struct vc4_compile *c, struct qreg reg)
 215 {
 216         static const char *files[] = {
 217                 [QFILE_TEMP] = "t",
 218                 [QFILE_VARY] = "v",
 219                 [QFILE_UNIF] = "u",
 220         };
 221
 222         if (reg.file == QFILE_NULL) {
 223                 fprintf(stderr, "null");
 224         } else if (reg.file == QFILE_SMALL_IMM) {
 225                 if ((int)reg.index >= -16 && (int)reg.index <= 15)
 226                         fprintf(stderr, "%d", reg.index);
 227                 else
 228                         fprintf(stderr, "%f", uif(reg.index));
 229         } else if (reg.file == QFILE_VPM) {
 230                 fprintf(stderr, "vpm");
 231         } else {
 232                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 233         }
 234
 235         if (reg.file == QFILE_UNIF &&
 236             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 237                 fprintf(stderr, " (0x%08x / %f)",
 238                         c->uniform_data[reg.index],
 239                         uif(c->uniform_data[reg.index]));
 240         }
 241 }
 242
 243 void
 244 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 245 {
 246         fprintf(stderr, "%s ", qir_get_op_name(inst->op));
 247
 248         qir_print_reg(c, inst->dst);
 249         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 250                 fprintf(stderr, ", ");
 251                 qir_print_reg(c, inst->src[i]);
 252         }
 253 }
 254
 255 void
 256 qir_dump(struct vc4_compile *c)
 257 {
 258         struct simple_node *node;
 259
 260         foreach(node, &c->instructions) {
 261                 struct qinst *inst = (struct qinst *)node;
 262                 qir_dump_inst(c, inst);
 263                 fprintf(stderr, "\n");
 264         }
 265 }
 266
 267 struct qreg
 268 qir_get_temp(struct vc4_compile *c)
 269 {
 270         struct qreg reg;
 271
 272         reg.file = QFILE_TEMP;
 273         reg.index = c->num_temps++;
 274
 275         return reg;
 276 }
 277
 278 struct qinst *
 279 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 280 {
 281         struct qinst *inst = CALLOC_STRUCT(qinst);
 282
 283         inst->op = op;
 284         inst->dst = dst;
 285         inst->src = calloc(2, sizeof(inst->src[0]));
 286         inst->src[0] = src0;
 287         inst->src[1] = src1;
 288
 289         return inst;
 290 }
 291
 292 struct qinst *
 293 qir_inst4(enum qop op, struct qreg dst,
 294           struct qreg a,
 295           struct qreg b,
 296           struct qreg c,
 297           struct qreg d)
 298 {
 299         struct qinst *inst = CALLOC_STRUCT(qinst);
 300
 301         inst->op = op;
 302         inst->dst = dst;
 303         inst->src = calloc(4, sizeof(*inst->src));
 304         inst->src[0] = a;
 305         inst->src[1] = b;
 306         inst->src[2] = c;
 307         inst->src[3] = d;
 308
 309         return inst;
 310 }
 311
 312 void
 313 qir_emit(struct vc4_compile *c, struct qinst *inst)
 314 {
 315         insert_at_tail(&c->instructions, &inst->link);
 316 }
 317
 318 bool
 319 qir_reg_equals(struct qreg a, struct qreg b)
 320 {
 321         return a.file == b.file && a.index == b.index;
 322 }
 323
 324 struct vc4_compile *
 325 qir_compile_init(void)
 326 {
 327         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 328
 329         make_empty_list(&c->instructions);
 330
 331         c->output_position_index = -1;
 332         c->output_clipvertex_index = -1;
 333         c->output_color_index = -1;
 334         c->output_point_size_index = -1;
 335
 336         return c;
 337 }
 338
 339 void
 340 qir_remove_instruction(struct qinst *qinst)
 341 {
 342         remove_from_list(&qinst->link);
 343         free(qinst->src);
 344         free(qinst);
 345 }
 346
 347 struct qreg
 348 qir_follow_movs(struct qinst **defs, struct qreg reg)
 349 {
 350         while (reg.file == QFILE_TEMP && defs[reg.index]->op == QOP_MOV)
 351                 reg = defs[reg.index]->src[0];
 352
 353         return reg;
 354 }
 355
 356 void
 357 qir_compile_destroy(struct vc4_compile *c)
 358 {
 359         while (!is_empty_list(&c->instructions)) {
 360                 struct qinst *qinst =
 361                         (struct qinst *)first_elem(&c->instructions);
 362                 qir_remove_instruction(qinst);
 363         }
 364
 365         ralloc_free(c);
 366 }
 367
 368 const char *
 369 qir_get_stage_name(enum qstage stage)
 370 {
 371         static const char *names[] = {
 372                 [QSTAGE_FRAG] = "FS",
 373                 [QSTAGE_VERT] = "VS",
 374                 [QSTAGE_COORD] = "CS",
 375         };
 376
 377         return names[stage];
 378 }
 379
 380 #define OPTPASS(func)                                                   \
 381         do {                                                            \
 382                 bool stage_progress = func(c);                          \
 383                 if (stage_progress) {                                   \
 384                         progress = true;                                \
 385                         if (print_opt_debug) {                          \
 386                                 fprintf(stderr,                         \
 387                                         "QIR opt pass %2d: %s progress\n", \
 388                                         pass, #func);                   \
 389                         }                                               \
 390                 }                                                       \
 391         } while (0)
 392
 393 void
 394 qir_optimize(struct vc4_compile *c)
 395 {
 396         bool print_opt_debug = false;
 397         int pass = 1;
 398
 399         while (true) {
 400                 bool progress = false;
 401
 402                 OPTPASS(qir_opt_algebraic);
 403                 OPTPASS(qir_opt_cse);
 404                 OPTPASS(qir_opt_copy_propagation);
 405                 OPTPASS(qir_opt_dead_code);
 406                 OPTPASS(qir_opt_small_immediates);
 407                 OPTPASS(qir_opt_vpm_writes);
 408
 409                 if (!progress)
 410                         break;
 411
 412                 pass++;
 413         }
 414 }