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 };
  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_SF] = { "sf", 0, 1 },
  62         [QOP_SEL_X_0_NS] = { "fsel_x_0_ns", 1, 1 },
  63         [QOP_SEL_X_0_NC] = { "fsel_x_0_nc", 1, 1 },
  64         [QOP_SEL_X_0_ZS] = { "fsel_x_0_zs", 1, 1 },
  65         [QOP_SEL_X_0_ZC] = { "fsel_x_0_zc", 1, 1 },
  66         [QOP_SEL_X_Y_NS] = { "fsel_x_y_ns", 1, 2 },
  67         [QOP_SEL_X_Y_NC] = { "fsel_x_y_nc", 1, 2 },
  68         [QOP_SEL_X_Y_ZS] = { "fsel_x_y_zs", 1, 2 },
  69         [QOP_SEL_X_Y_ZC] = { "fsel_x_y_zc", 1, 2 },
  70
  71         [QOP_RCP] = { "rcp", 1, 1 },
  72         [QOP_RSQ] = { "rsq", 1, 1 },
  73         [QOP_EXP2] = { "exp2", 1, 2 },
  74         [QOP_LOG2] = { "log2", 1, 2 },
  75         [QOP_PACK_COLORS] = { "pack_colors", 1, 4 },
  76         [QOP_PACK_SCALED] = { "pack_scaled", 1, 2 },
  77         [QOP_VPM_WRITE] = { "vpm_write", 0, 1, 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, true },
  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_RESULT] = { "tex_result", 1, 0, true },
  97         [QOP_R4_UNPACK_A] = { "r4_unpack_a", 1, 1 },
  98         [QOP_R4_UNPACK_B] = { "r4_unpack_b", 1, 1 },
  99         [QOP_R4_UNPACK_C] = { "r4_unpack_c", 1, 1 },
 100         [QOP_R4_UNPACK_D] = { "r4_unpack_d", 1, 1 },
 101         [QOP_UNPACK_8A] = { "unpack_8a", 1, 1 },
 102         [QOP_UNPACK_8B] = { "unpack_8b", 1, 1 },
 103         [QOP_UNPACK_8C] = { "unpack_8c", 1, 1 },
 104         [QOP_UNPACK_8D] = { "unpack_8d", 1, 1 },
 105 };
 106
 107 static const char *
 108 qir_get_op_name(enum qop qop)
 109 {
 110         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 111                 return qir_op_info[qop].name;
 112         else
 113                 return "???";
 114 }
 115
 116 int
 117 qir_get_op_nsrc(enum qop qop)
 118 {
 119         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 120                 return qir_op_info[qop].nsrc;
 121         else
 122                 abort();
 123 }
 124
 125 bool
 126 qir_has_side_effects(struct qinst *inst)
 127 {
 128         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 129                 if (inst->src[i].file == QFILE_VARY)
 130                         return true;
 131         }
 132
 133         return qir_op_info[inst->op].has_side_effects;
 134 }
 135
 136 bool
 137 qir_depends_on_flags(struct qinst *inst)
 138 {
 139         switch (inst->op) {
 140         case QOP_SEL_X_0_NS:
 141         case QOP_SEL_X_0_NC:
 142         case QOP_SEL_X_0_ZS:
 143         case QOP_SEL_X_0_ZC:
 144         case QOP_SEL_X_Y_NS:
 145         case QOP_SEL_X_Y_NC:
 146         case QOP_SEL_X_Y_ZS:
 147         case QOP_SEL_X_Y_ZC:
 148                 return true;
 149         default:
 150                 return false;
 151         }
 152 }
 153
 154 bool
 155 qir_writes_r4(struct qinst *inst)
 156 {
 157         switch (inst->op) {
 158         case QOP_TEX_RESULT:
 159         case QOP_TLB_COLOR_READ:
 160         case QOP_RCP:
 161         case QOP_RSQ:
 162         case QOP_EXP2:
 163         case QOP_LOG2:
 164                 return true;
 165         default:
 166                 return false;
 167         }
 168 }
 169
 170 bool
 171 qir_reads_r4(struct qinst *inst)
 172 {
 173         switch (inst->op) {
 174         case QOP_R4_UNPACK_A:
 175         case QOP_R4_UNPACK_B:
 176         case QOP_R4_UNPACK_C:
 177         case QOP_R4_UNPACK_D:
 178                 return true;
 179         default:
 180                 return false;
 181         }
 182 }
 183
 184 static void
 185 qir_print_reg(struct vc4_compile *c, struct qreg reg)
 186 {
 187         const char *files[] = {
 188                 [QFILE_TEMP] = "t",
 189                 [QFILE_VARY] = "v",
 190                 [QFILE_UNIF] = "u",
 191         };
 192
 193         if (reg.file == QFILE_NULL)
 194                 fprintf(stderr, "null");
 195         else
 196                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 197
 198         if (reg.file == QFILE_UNIF &&
 199             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 200                 fprintf(stderr, " (0x%08x / %f)",
 201                         c->uniform_data[reg.index],
 202                         uif(c->uniform_data[reg.index]));
 203         }
 204 }
 205
 206 void
 207 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 208 {
 209         fprintf(stderr, "%s ", qir_get_op_name(inst->op));
 210
 211         qir_print_reg(c, inst->dst);
 212         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 213                 fprintf(stderr, ", ");
 214                 qir_print_reg(c, inst->src[i]);
 215         }
 216 }
 217
 218 void
 219 qir_dump(struct vc4_compile *c)
 220 {
 221         struct simple_node *node;
 222
 223         foreach(node, &c->instructions) {
 224                 struct qinst *inst = (struct qinst *)node;
 225                 qir_dump_inst(c, inst);
 226                 fprintf(stderr, "\n");
 227         }
 228 }
 229
 230 struct qreg
 231 qir_get_temp(struct vc4_compile *c)
 232 {
 233         struct qreg reg;
 234
 235         reg.file = QFILE_TEMP;
 236         reg.index = c->num_temps++;
 237
 238         return reg;
 239 }
 240
 241 struct qinst *
 242 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 243 {
 244         struct qinst *inst = CALLOC_STRUCT(qinst);
 245
 246         inst->op = op;
 247         inst->dst = dst;
 248         inst->src = calloc(2, sizeof(inst->src[0]));
 249         inst->src[0] = src0;
 250         inst->src[1] = src1;
 251
 252         return inst;
 253 }
 254
 255 struct qinst *
 256 qir_inst4(enum qop op, struct qreg dst,
 257           struct qreg a,
 258           struct qreg b,
 259           struct qreg c,
 260           struct qreg d)
 261 {
 262         struct qinst *inst = CALLOC_STRUCT(qinst);
 263
 264         inst->op = op;
 265         inst->dst = dst;
 266         inst->src = calloc(4, sizeof(*inst->src));
 267         inst->src[0] = a;
 268         inst->src[1] = b;
 269         inst->src[2] = c;
 270         inst->src[3] = d;
 271
 272         return inst;
 273 }
 274
 275 void
 276 qir_emit(struct vc4_compile *c, struct qinst *inst)
 277 {
 278         insert_at_tail(&c->instructions, &inst->link);
 279 }
 280
 281 bool
 282 qir_reg_equals(struct qreg a, struct qreg b)
 283 {
 284         return a.file == b.file && a.index == b.index;
 285 }
 286
 287 struct vc4_compile *
 288 qir_compile_init(void)
 289 {
 290         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 291
 292         make_empty_list(&c->instructions);
 293
 294         c->output_position_index = -1;
 295         c->output_color_index = -1;
 296
 297         return c;
 298 }
 299
 300 void
 301 qir_remove_instruction(struct qinst *qinst)
 302 {
 303         remove_from_list(&qinst->link);
 304         free(qinst->src);
 305         free(qinst);
 306 }
 307
 308 void
 309 qir_compile_destroy(struct vc4_compile *c)
 310 {
 311         while (!is_empty_list(&c->instructions)) {
 312                 struct qinst *qinst =
 313                         (struct qinst *)first_elem(&c->instructions);
 314                 qir_remove_instruction(qinst);
 315         }
 316
 317         ralloc_free(c);
 318 }
 319
 320 const char *
 321 qir_get_stage_name(enum qstage stage)
 322 {
 323         static const char *names[] = {
 324                 [QSTAGE_FRAG] = "FS",
 325                 [QSTAGE_VERT] = "VS",
 326                 [QSTAGE_COORD] = "CS",
 327         };
 328
 329         return names[stage];
 330 }
 331
 332 #define OPTPASS(func)                                                   \
 333         do {                                                            \
 334                 bool stage_progress = func(c);                          \
 335                 if (stage_progress) {                                   \
 336                         progress = true;                                \
 337                         if (print_opt_debug) {                          \
 338                                 fprintf(stderr,                         \
 339                                         "QIR opt pass %2d: %s progress\n", \
 340                                         pass, #func);                   \
 341                         }                                               \
 342                 }                                                       \
 343         } while (0)
 344
 345 void
 346 qir_optimize(struct vc4_compile *c)
 347 {
 348         bool print_opt_debug = false;
 349         int pass = 1;
 350
 351         while (true) {
 352                 bool progress = false;
 353
 354                 OPTPASS(qir_opt_algebraic);
 355                 OPTPASS(qir_opt_cse);
 356                 OPTPASS(qir_opt_copy_propagation);
 357                 OPTPASS(qir_opt_dead_code);
 358
 359                 if (!progress)
 360                         break;
 361
 362                 pass++;
 363         }
 364 }