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_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] = { "unpack_8a", 1, 1 },
 103         [QOP_UNPACK_8B] = { "unpack_8b", 1, 1 },
 104         [QOP_UNPACK_8C] = { "unpack_8c", 1, 1 },
 105         [QOP_UNPACK_8D] = { "unpack_8d", 1, 1 },
 106 };
 107
 108 static const char *
 109 qir_get_op_name(enum qop qop)
 110 {
 111         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 112                 return qir_op_info[qop].name;
 113         else
 114                 return "???";
 115 }
 116
 117 int
 118 qir_get_op_nsrc(enum qop qop)
 119 {
 120         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 121                 return qir_op_info[qop].nsrc;
 122         else
 123                 abort();
 124 }
 125
 126 /**
 127  * Returns whether the instruction has any side effects that must be
 128  * preserved.
 129  */
 130 bool
 131 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 132 {
 133         /* We can dead-code eliminate varyings, because we only tell the VS
 134          * about the live ones at the end.  But we have to preserve the
 135          * point/line coordinates reads, because they're generated by
 136          * fixed-function hardware.
 137          */
 138         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 139                 if (inst->src[i].file == QFILE_VARY &&
 140                     c->input_semantics[inst->src[i].index].semantic == 0xff) {
 141                         return true;
 142                 }
 143         }
 144
 145         return qir_op_info[inst->op].has_side_effects;
 146 }
 147
 148 bool
 149 qir_depends_on_flags(struct qinst *inst)
 150 {
 151         switch (inst->op) {
 152         case QOP_SEL_X_0_NS:
 153         case QOP_SEL_X_0_NC:
 154         case QOP_SEL_X_0_ZS:
 155         case QOP_SEL_X_0_ZC:
 156         case QOP_SEL_X_Y_NS:
 157         case QOP_SEL_X_Y_NC:
 158         case QOP_SEL_X_Y_ZS:
 159         case QOP_SEL_X_Y_ZC:
 160                 return true;
 161         default:
 162                 return false;
 163         }
 164 }
 165
 166 bool
 167 qir_writes_r4(struct qinst *inst)
 168 {
 169         switch (inst->op) {
 170         case QOP_TEX_RESULT:
 171         case QOP_TLB_COLOR_READ:
 172         case QOP_RCP:
 173         case QOP_RSQ:
 174         case QOP_EXP2:
 175         case QOP_LOG2:
 176                 return true;
 177         default:
 178                 return false;
 179         }
 180 }
 181
 182 bool
 183 qir_reads_r4(struct qinst *inst)
 184 {
 185         switch (inst->op) {
 186         case QOP_R4_UNPACK_A:
 187         case QOP_R4_UNPACK_B:
 188         case QOP_R4_UNPACK_C:
 189         case QOP_R4_UNPACK_D:
 190                 return true;
 191         default:
 192                 return false;
 193         }
 194 }
 195
 196 static void
 197 qir_print_reg(struct vc4_compile *c, struct qreg reg)
 198 {
 199         const char *files[] = {
 200                 [QFILE_TEMP] = "t",
 201                 [QFILE_VARY] = "v",
 202                 [QFILE_UNIF] = "u",
 203         };
 204
 205         if (reg.file == QFILE_NULL)
 206                 fprintf(stderr, "null");
 207         else
 208                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 209
 210         if (reg.file == QFILE_UNIF &&
 211             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT) {
 212                 fprintf(stderr, " (0x%08x / %f)",
 213                         c->uniform_data[reg.index],
 214                         uif(c->uniform_data[reg.index]));
 215         }
 216 }
 217
 218 void
 219 qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
 220 {
 221         fprintf(stderr, "%s ", qir_get_op_name(inst->op));
 222
 223         qir_print_reg(c, inst->dst);
 224         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 225                 fprintf(stderr, ", ");
 226                 qir_print_reg(c, inst->src[i]);
 227         }
 228 }
 229
 230 void
 231 qir_dump(struct vc4_compile *c)
 232 {
 233         struct simple_node *node;
 234
 235         foreach(node, &c->instructions) {
 236                 struct qinst *inst = (struct qinst *)node;
 237                 qir_dump_inst(c, inst);
 238                 fprintf(stderr, "\n");
 239         }
 240 }
 241
 242 struct qreg
 243 qir_get_temp(struct vc4_compile *c)
 244 {
 245         struct qreg reg;
 246
 247         reg.file = QFILE_TEMP;
 248         reg.index = c->num_temps++;
 249
 250         return reg;
 251 }
 252
 253 struct qinst *
 254 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 255 {
 256         struct qinst *inst = CALLOC_STRUCT(qinst);
 257
 258         inst->op = op;
 259         inst->dst = dst;
 260         inst->src = calloc(2, sizeof(inst->src[0]));
 261         inst->src[0] = src0;
 262         inst->src[1] = src1;
 263
 264         return inst;
 265 }
 266
 267 struct qinst *
 268 qir_inst4(enum qop op, struct qreg dst,
 269           struct qreg a,
 270           struct qreg b,
 271           struct qreg c,
 272           struct qreg d)
 273 {
 274         struct qinst *inst = CALLOC_STRUCT(qinst);
 275
 276         inst->op = op;
 277         inst->dst = dst;
 278         inst->src = calloc(4, sizeof(*inst->src));
 279         inst->src[0] = a;
 280         inst->src[1] = b;
 281         inst->src[2] = c;
 282         inst->src[3] = d;
 283
 284         return inst;
 285 }
 286
 287 void
 288 qir_emit(struct vc4_compile *c, struct qinst *inst)
 289 {
 290         insert_at_tail(&c->instructions, &inst->link);
 291 }
 292
 293 bool
 294 qir_reg_equals(struct qreg a, struct qreg b)
 295 {
 296         return a.file == b.file && a.index == b.index;
 297 }
 298
 299 struct vc4_compile *
 300 qir_compile_init(void)
 301 {
 302         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 303
 304         make_empty_list(&c->instructions);
 305
 306         c->output_position_index = -1;
 307         c->output_clipvertex_index = -1;
 308         c->output_color_index = -1;
 309         c->output_point_size_index = -1;
 310
 311         return c;
 312 }
 313
 314 void
 315 qir_remove_instruction(struct qinst *qinst)
 316 {
 317         remove_from_list(&qinst->link);
 318         free(qinst->src);
 319         free(qinst);
 320 }
 321
 322 void
 323 qir_compile_destroy(struct vc4_compile *c)
 324 {
 325         while (!is_empty_list(&c->instructions)) {
 326                 struct qinst *qinst =
 327                         (struct qinst *)first_elem(&c->instructions);
 328                 qir_remove_instruction(qinst);
 329         }
 330
 331         ralloc_free(c);
 332 }
 333
 334 const char *
 335 qir_get_stage_name(enum qstage stage)
 336 {
 337         static const char *names[] = {
 338                 [QSTAGE_FRAG] = "FS",
 339                 [QSTAGE_VERT] = "VS",
 340                 [QSTAGE_COORD] = "CS",
 341         };
 342
 343         return names[stage];
 344 }
 345
 346 #define OPTPASS(func)                                                   \
 347         do {                                                            \
 348                 bool stage_progress = func(c);                          \
 349                 if (stage_progress) {                                   \
 350                         progress = true;                                \
 351                         if (print_opt_debug) {                          \
 352                                 fprintf(stderr,                         \
 353                                         "QIR opt pass %2d: %s progress\n", \
 354                                         pass, #func);                   \
 355                         }                                               \
 356                 }                                                       \
 357         } while (0)
 358
 359 void
 360 qir_optimize(struct vc4_compile *c)
 361 {
 362         bool print_opt_debug = false;
 363         int pass = 1;
 364
 365         while (true) {
 366                 bool progress = false;
 367
 368                 OPTPASS(qir_opt_algebraic);
 369                 OPTPASS(qir_opt_cse);
 370                 OPTPASS(qir_opt_copy_propagation);
 371                 OPTPASS(qir_opt_dead_code);
 372
 373                 if (!progress)
 374                         break;
 375
 376                 pass++;
 377         }
 378 }