src/gallium/drivers/lima/ir/pp/ppir.h

   1 /*
   2  * Copyright (c) 2017 Lima Project
   3  * Copyright (c) 2013 Connor Abbott
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a copy
   6  * of this software and associated documentation files (the "Software"), to deal
   7  * in the Software without restriction, including without limitation the rights
   8  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
   9  * copies of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be included in
  13  * all copies or substantial portions of the 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 THE
  18  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  21  * THE SOFTWARE.
  22  *
  23  */
  24
  25 #ifndef LIMA_IR_PP_PPIR_H
  26 #define LIMA_IR_PP_PPIR_H
  27
  28 #include "util/u_math.h"
  29 #include "util/list.h"
  30 #include "util/set.h"
  31
  32 #include "ir/lima_ir.h"
  33
  34 typedef enum {
  35    ppir_op_mov,
  36    ppir_op_abs,
  37    ppir_op_neg,
  38    ppir_op_sat,
  39    ppir_op_add,
  40
  41    ppir_op_ddx,
  42    ppir_op_ddy,
  43
  44    ppir_op_mul,
  45    ppir_op_rcp,
  46
  47    ppir_op_sin_lut,
  48    ppir_op_cos_lut,
  49
  50    ppir_op_sum3,
  51    ppir_op_sum4,
  52
  53    ppir_op_normalize2,
  54    ppir_op_normalize3,
  55    ppir_op_normalize4,
  56
  57    ppir_op_sel_cond,
  58    ppir_op_select,
  59
  60    ppir_op_sin,
  61    ppir_op_cos,
  62    ppir_op_tan,
  63    ppir_op_asin,
  64    ppir_op_acos,
  65
  66    ppir_op_atan,
  67    ppir_op_atan2,
  68    ppir_op_atan_pt1,
  69    ppir_op_atan2_pt1,
  70    ppir_op_atan_pt2,
  71
  72    ppir_op_exp,
  73    ppir_op_log,
  74    ppir_op_exp2,
  75    ppir_op_log2,
  76    ppir_op_sqrt,
  77    ppir_op_rsqrt,
  78
  79    ppir_op_sign,
  80    ppir_op_floor,
  81    ppir_op_ceil,
  82    ppir_op_fract,
  83    ppir_op_mod,
  84    ppir_op_min,
  85    ppir_op_max,
  86    ppir_op_trunc,
  87
  88    ppir_op_and,
  89    ppir_op_or,
  90    ppir_op_xor,
  91
  92    ppir_op_lt,
  93    ppir_op_gt,
  94    ppir_op_le,
  95    ppir_op_ge,
  96    ppir_op_eq,
  97    ppir_op_ne,
  98    ppir_op_not,
  99
 100    ppir_op_load_uniform,
 101    ppir_op_load_varying,
 102    ppir_op_load_coords,
 103    ppir_op_load_coords_reg,
 104    ppir_op_load_fragcoord,
 105    ppir_op_load_pointcoord,
 106    ppir_op_load_frontface,
 107    ppir_op_load_texture,
 108    ppir_op_load_temp,
 109
 110    ppir_op_store_temp,
 111    ppir_op_store_color,
 112
 113    ppir_op_const,
 114
 115    ppir_op_discard,
 116    ppir_op_branch,
 117
 118    ppir_op_undef,
 119    ppir_op_dummy,
 120
 121    ppir_op_num,
 122 } ppir_op;
 123
 124 typedef enum {
 125    ppir_node_type_alu,
 126    ppir_node_type_const,
 127    ppir_node_type_load,
 128    ppir_node_type_store,
 129    ppir_node_type_load_texture,
 130    ppir_node_type_discard,
 131    ppir_node_type_branch,
 132 } ppir_node_type;
 133
 134 typedef struct {
 135    char *name;
 136    ppir_node_type type;
 137    int *slots;
 138 } ppir_op_info;
 139
 140 extern const ppir_op_info ppir_op_infos[];
 141
 142 typedef enum {
 143    ppir_dep_src,
 144    ppir_dep_write_after_read,
 145    ppir_dep_sequence,
 146 } ppir_dep_type;
 147
 148 typedef struct {
 149    void *pred, *succ;
 150    ppir_dep_type type;
 151    struct list_head pred_link;
 152    struct list_head succ_link;
 153 } ppir_dep;
 154
 155 typedef struct ppir_node {
 156    struct list_head list;
 157    ppir_op op;
 158    ppir_node_type type;
 159    int index;
 160    char name[16];
 161    bool printed;
 162    struct ppir_instr *instr;
 163    int instr_pos;
 164    struct ppir_block *block;
 165
 166    /* for scheduler */
 167    struct list_head succ_list;
 168    struct list_head pred_list;
 169 } ppir_node;
 170
 171 typedef enum {
 172    ppir_pipeline_reg_const0,
 173    ppir_pipeline_reg_const1,
 174    ppir_pipeline_reg_sampler,
 175    ppir_pipeline_reg_uniform,
 176    ppir_pipeline_reg_vmul,
 177    ppir_pipeline_reg_fmul,
 178    ppir_pipeline_reg_discard, /* varying load */
 179 } ppir_pipeline;
 180
 181 typedef struct ppir_reg {
 182    struct list_head list;
 183    int index;
 184    int regalloc_index;
 185    int num_components;
 186
 187    /* whether this reg has to start from the x component
 188     * of a full physical reg, this is true for reg used
 189     * in load/store instr which has no swizzle field */
 190    bool is_head;
 191    bool spilled;
 192    bool undef;
 193 } ppir_reg;
 194
 195 typedef enum {
 196    ppir_target_ssa,
 197    ppir_target_pipeline,
 198    ppir_target_register,
 199 } ppir_target;
 200
 201 typedef struct ppir_src {
 202    ppir_target type;
 203    ppir_node *node;
 204
 205    union {
 206       ppir_reg *ssa;
 207       ppir_reg *reg;
 208       ppir_pipeline pipeline;
 209    };
 210
 211    uint8_t swizzle[4];
 212    bool absolute, negate;
 213 } ppir_src;
 214
 215 typedef enum {
 216    ppir_outmod_none,
 217    ppir_outmod_clamp_fraction,
 218    ppir_outmod_clamp_positive,
 219    ppir_outmod_round,
 220 } ppir_outmod;
 221
 222 typedef struct ppir_dest {
 223    ppir_target type;
 224
 225    union {
 226       ppir_reg ssa;
 227       ppir_reg *reg;
 228       ppir_pipeline pipeline;
 229    };
 230
 231    ppir_outmod modifier;
 232    unsigned write_mask : 4;
 233 } ppir_dest;
 234
 235 typedef struct {
 236    ppir_node node;
 237    ppir_dest dest;
 238    ppir_src src[3];
 239    int num_src;
 240    int shift : 3; /* Only used for ppir_op_mul */
 241 } ppir_alu_node;
 242
 243 typedef struct ppir_const {
 244    union fi value[4];
 245    int num;
 246 } ppir_const;
 247
 248 typedef struct {
 249    ppir_node node;
 250    ppir_const constant;
 251    ppir_dest dest;
 252 } ppir_const_node;
 253
 254 typedef struct {
 255    ppir_node node;
 256    int index;
 257    int num_components;
 258    ppir_dest dest;
 259    ppir_src src;
 260    int num_src;
 261 } ppir_load_node;
 262
 263 typedef struct {
 264    ppir_node node;
 265    int index;
 266    int num_components;
 267    ppir_src src;
 268 } ppir_store_node;
 269
 270 typedef struct {
 271    ppir_node node;
 272    ppir_dest dest;
 273    ppir_src src[2]; /* src[0] temporarily stores src_coords,
 274                        not to be used after lowering */
 275    int num_src;
 276    int sampler;
 277    int sampler_dim;
 278    bool lod_bias_en;
 279    bool explicit_lod;
 280 } ppir_load_texture_node;
 281
 282 typedef struct {
 283    ppir_node node;
 284 } ppir_discard_node;
 285
 286 enum ppir_instr_slot {
 287    PPIR_INSTR_SLOT_VARYING,
 288    PPIR_INSTR_SLOT_TEXLD,
 289    PPIR_INSTR_SLOT_UNIFORM,
 290    PPIR_INSTR_SLOT_ALU_VEC_MUL,
 291    PPIR_INSTR_SLOT_ALU_SCL_MUL,
 292    PPIR_INSTR_SLOT_ALU_VEC_ADD,
 293    PPIR_INSTR_SLOT_ALU_SCL_ADD,
 294    PPIR_INSTR_SLOT_ALU_COMBINE,
 295    PPIR_INSTR_SLOT_STORE_TEMP,
 296    PPIR_INSTR_SLOT_BRANCH,
 297    PPIR_INSTR_SLOT_NUM,
 298    PPIR_INSTR_SLOT_END,
 299    PPIR_INSTR_SLOT_ALU_START = PPIR_INSTR_SLOT_ALU_VEC_MUL,
 300    PPIR_INSTR_SLOT_ALU_END = PPIR_INSTR_SLOT_ALU_COMBINE,
 301 };
 302
 303 struct ppir_liveness {
 304    ppir_reg *reg;
 305    unsigned mask : 4;
 306 };
 307
 308 typedef struct ppir_instr {
 309    struct list_head list;
 310    int index;
 311    bool printed;
 312    int seq; /* command sequence after schedule */
 313
 314    ppir_node *slots[PPIR_INSTR_SLOT_NUM];
 315    ppir_const constant[2];
 316    bool is_end;
 317
 318    /* for scheduler */
 319    struct list_head succ_list;
 320    struct list_head pred_list;
 321    float reg_pressure;
 322    int est; /* earliest start time */
 323    int parent_index;
 324    bool scheduled;
 325    int offset;
 326    int encode_size;
 327
 328    /* for liveness analysis */
 329    struct ppir_liveness *live_in;
 330    struct ppir_liveness *live_out;
 331    /* live_internal is to mark registers only live within an
 332     * instruction, without propagation */
 333    struct ppir_liveness *live_internal;
 334    struct set *live_in_set;
 335    struct set *live_out_set;
 336    struct set *live_internal_set;
 337 } ppir_instr;
 338
 339 typedef struct ppir_block {
 340    struct list_head list;
 341    struct list_head node_list;
 342    struct list_head instr_list;
 343
 344    struct ppir_block *successors[2];
 345
 346    struct ppir_compiler *comp;
 347
 348    /* for scheduler */
 349    int sched_instr_index;
 350    int sched_instr_base;
 351    int index;
 352
 353    /* for liveness analysis */
 354    struct ppir_liveness *live_in;
 355    struct ppir_liveness *live_out;
 356    struct set *live_in_set;
 357    struct set *live_out_set;
 358 } ppir_block;
 359
 360 typedef struct {
 361    ppir_node node;
 362    ppir_src src[2];
 363    int num_src;
 364    bool cond_gt;
 365    bool cond_eq;
 366    bool cond_lt;
 367    bool negate;
 368    ppir_block *target;
 369 } ppir_branch_node;
 370
 371 struct ra_regs;
 372 struct lima_fs_shader_state;
 373
 374 typedef struct ppir_compiler {
 375    struct list_head block_list;
 376    struct hash_table_u64 *blocks;
 377    int cur_index;
 378    int cur_instr_index;
 379
 380    struct list_head reg_list;
 381
 382    /* array for searching ssa/reg node */
 383    ppir_node **var_nodes;
 384    unsigned reg_base;
 385
 386    struct ra_regs *ra;
 387    struct lima_fs_shader_state *prog;
 388
 389    /* for scheduler */
 390    int sched_instr_base;
 391
 392    /* for regalloc spilling debug */
 393    int force_spilling;
 394
 395    /* shaderdb */
 396    int num_loops;
 397    int num_spills;
 398    int num_fills;
 399
 400    ppir_block *discard_block;
 401    ppir_block *current_block;
 402    ppir_block *loop_break_block;
 403    ppir_block *loop_cont_block;
 404 } ppir_compiler;
 405
 406 void *ppir_node_create(ppir_block *block, ppir_op op, int index, unsigned mask);
 407 void ppir_node_add_dep(ppir_node *succ, ppir_node *pred, ppir_dep_type type);
 408 void ppir_node_remove_dep(ppir_dep *dep);
 409 void ppir_node_delete(ppir_node *node);
 410 void ppir_node_print_prog(ppir_compiler *comp);
 411 void ppir_node_replace_child(ppir_node *parent, ppir_node *old_child, ppir_node *new_child);
 412 void ppir_node_replace_all_succ(ppir_node *dst, ppir_node *src);
 413 void ppir_node_replace_pred(ppir_dep *dep, ppir_node *new_pred);
 414 void ppir_delete_if_orphan(ppir_block *block, ppir_node *node);
 415 ppir_dep *ppir_dep_for_pred(ppir_node *node, ppir_node *pred);
 416 ppir_node *ppir_node_clone(ppir_block *block, ppir_node *node);
 417 /* Assumes that node successors are in the same block */
 418 ppir_node *ppir_node_insert_mov(ppir_node *node);
 419 ppir_node *ppir_node_insert_mov_all_blocks(ppir_node *node);
 420
 421 static inline bool ppir_node_is_root(ppir_node *node)
 422 {
 423    return list_is_empty(&node->succ_list);
 424 }
 425
 426 static inline bool ppir_node_is_leaf(ppir_node *node)
 427 {
 428    return list_is_empty(&node->pred_list);
 429 }
 430
 431 static inline bool ppir_node_has_single_succ(ppir_node *node)
 432 {
 433    return list_is_singular(&node->succ_list);
 434 }
 435
 436 bool ppir_node_has_single_src_succ(ppir_node *node);
 437
 438 static inline ppir_node *ppir_node_first_succ(ppir_node *node)
 439 {
 440    return list_first_entry(&node->succ_list, ppir_dep, succ_link)->succ;
 441 }
 442
 443 static inline bool ppir_node_has_single_pred(ppir_node *node)
 444 {
 445    return list_is_singular(&node->pred_list);
 446 }
 447
 448 static inline ppir_node *ppir_node_first_pred(ppir_node *node)
 449 {
 450    return list_first_entry(&node->pred_list, ppir_dep, pred_link)->pred;
 451 }
 452
 453 #define ppir_node_foreach_succ(node, dep) \
 454    list_for_each_entry(ppir_dep, dep, &node->succ_list, succ_link)
 455 #define ppir_node_foreach_succ_safe(node, dep) \
 456    list_for_each_entry_safe(ppir_dep, dep, &node->succ_list, succ_link)
 457 #define ppir_node_foreach_pred(node, dep) \
 458    list_for_each_entry(ppir_dep, dep, &node->pred_list, pred_link)
 459 #define ppir_node_foreach_pred_safe(node, dep) \
 460    list_for_each_entry_safe(ppir_dep, dep, &node->pred_list, pred_link)
 461
 462 #define ppir_node_to_alu(node) ((ppir_alu_node *)(node))
 463 #define ppir_node_to_const(node) ((ppir_const_node *)(node))
 464 #define ppir_node_to_load(node) ((ppir_load_node *)(node))
 465 #define ppir_node_to_store(node) ((ppir_store_node *)(node))
 466 #define ppir_node_to_load_texture(node) ((ppir_load_texture_node *)(node))
 467 #define ppir_node_to_discard(node) ((ppir_discard_node *)(node))
 468 #define ppir_node_to_branch(node) ((ppir_branch_node *)(node))
 469
 470 static inline ppir_dest *ppir_node_get_dest(ppir_node *node)
 471 {
 472    switch (node->type) {
 473    case ppir_node_type_alu:
 474       return &ppir_node_to_alu(node)->dest;
 475    case ppir_node_type_load:
 476       return &ppir_node_to_load(node)->dest;
 477    case ppir_node_type_const:
 478       return &ppir_node_to_const(node)->dest;
 479    case ppir_node_type_load_texture:
 480       return &ppir_node_to_load_texture(node)->dest;
 481    default:
 482       return NULL;
 483    }
 484 }
 485
 486 static inline int ppir_node_get_src_num(ppir_node *node)
 487 {
 488    switch (node->type) {
 489    case ppir_node_type_alu:
 490       return ppir_node_to_alu(node)->num_src;
 491    case ppir_node_type_branch:
 492       return ppir_node_to_branch(node)->num_src;
 493    case ppir_node_type_load:
 494       return ppir_node_to_load(node)->num_src;
 495    case ppir_node_type_load_texture:
 496       return ppir_node_to_load_texture(node)->num_src;
 497    case ppir_node_type_store:
 498       return 1;
 499    default:
 500       return 0;
 501    }
 502
 503    return 0;
 504 }
 505
 506 static inline ppir_src *ppir_node_get_src(ppir_node *node, int idx)
 507 {
 508    if (idx < 0 || idx >= ppir_node_get_src_num(node))
 509       return NULL;
 510
 511    switch (node->type) {
 512    case ppir_node_type_alu:
 513       return &ppir_node_to_alu(node)->src[idx];
 514    case ppir_node_type_branch:
 515       return &ppir_node_to_branch(node)->src[idx];
 516    case ppir_node_type_load_texture:
 517       return &ppir_node_to_load_texture(node)->src[idx];
 518    case ppir_node_type_load:
 519       return &ppir_node_to_load(node)->src;
 520    case ppir_node_type_store:
 521       return &ppir_node_to_store(node)->src;
 522    default:
 523       break;
 524    }
 525
 526    return NULL;
 527 }
 528
 529 static inline ppir_reg *ppir_src_get_reg(ppir_src *src)
 530 {
 531    switch (src->type) {
 532    case ppir_target_ssa:
 533       return src->ssa;
 534    case ppir_target_register:
 535       return src->reg;
 536    default:
 537       return NULL;
 538    }
 539 }
 540
 541 static inline ppir_reg *ppir_dest_get_reg(ppir_dest *dest)
 542 {
 543    switch (dest->type) {
 544    case ppir_target_ssa:
 545       return &dest->ssa;
 546    case ppir_target_register:
 547       return dest->reg;
 548    default:
 549       return NULL;
 550    }
 551 }
 552
 553 static inline ppir_src *ppir_node_get_src_for_pred(ppir_node *node, ppir_node *pred)
 554 {
 555    for (int i = 0; i < ppir_node_get_src_num(node); i++) {
 556       ppir_src *src = ppir_node_get_src(node, i);
 557       if (src && src->node == pred)
 558          return src;
 559    }
 560
 561    return NULL;
 562 }
 563
 564 static inline void ppir_node_target_assign(ppir_src *src, ppir_node *node)
 565 {
 566    ppir_dest *dest = ppir_node_get_dest(node);
 567    src->type = dest->type;
 568    switch (src->type) {
 569    case ppir_target_ssa:
 570       src->ssa = &dest->ssa;
 571       src->node = node;
 572       break;
 573    case ppir_target_register:
 574       src->reg = dest->reg;
 575       /* Registers can be assigned from multiple nodes, so don't keep
 576        * pointer to the node here
 577        */
 578       src->node = NULL;
 579       break;
 580    case ppir_target_pipeline:
 581       src->pipeline = dest->pipeline;
 582       src->node = node;
 583       break;
 584    }
 585 }
 586
 587 static inline bool ppir_node_target_equal(ppir_src *src, ppir_dest *dest)
 588 {
 589    if (src->type != dest->type ||
 590        (src->type == ppir_target_ssa && src->ssa != &dest->ssa) ||
 591        (src->type == ppir_target_register && src->reg != dest->reg) ||
 592        (src->type == ppir_target_pipeline && src->pipeline != dest->pipeline))
 593       return false;
 594
 595    return true;
 596 }
 597
 598 static inline int ppir_target_get_src_reg_index(ppir_src *src)
 599 {
 600    switch (src->type) {
 601    case ppir_target_ssa:
 602       if (src->ssa)
 603          return src->ssa->index;
 604       break;
 605    case ppir_target_register:
 606       if (src->reg)
 607          return src->reg->index;
 608       break;
 609    case ppir_target_pipeline:
 610       if (src->pipeline == ppir_pipeline_reg_discard)
 611          return 15 * 4;
 612       return (src->pipeline + 12) * 4;
 613    }
 614
 615    return -1;
 616 }
 617
 618 static inline int ppir_target_get_dest_reg_index(ppir_dest *dest)
 619 {
 620    switch (dest->type) {
 621    case ppir_target_ssa:
 622       return dest->ssa.index;
 623    case ppir_target_register:
 624       return dest->reg->index;
 625    case ppir_target_pipeline:
 626       if (dest->pipeline == ppir_pipeline_reg_discard)
 627          return 15 * 4;
 628       return (dest->pipeline + 12) * 4;
 629    }
 630
 631    return -1;
 632 }
 633
 634 static inline int ppir_src_get_mask(ppir_src *src)
 635 {
 636    ppir_reg *reg = ppir_src_get_reg(src);
 637    int mask = 0;
 638
 639    for (int i = 0; i < reg->num_components; i++)
 640       mask |= (1 << src->swizzle[i]);
 641
 642    return mask;
 643 }
 644
 645 static inline bool ppir_target_is_scaler(ppir_dest *dest)
 646 {
 647    switch (dest->type) {
 648    case ppir_target_ssa:
 649       return dest->ssa.num_components == 1;
 650    case ppir_target_register:
 651       /* only one bit in mask is set */
 652       if ((dest->write_mask & 0x3) == 0x3 ||
 653           (dest->write_mask & 0x5) == 0x5 ||
 654           (dest->write_mask & 0x9) == 0x9 ||
 655           (dest->write_mask & 0x6) == 0x6 ||
 656           (dest->write_mask & 0xa) == 0xa ||
 657           (dest->write_mask & 0xc) == 0xc)
 658          return false;
 659       else
 660          return true;
 661    case ppir_target_pipeline:
 662       if (dest->pipeline == ppir_pipeline_reg_fmul)
 663          return true;
 664       else
 665          return false;
 666    default:
 667       return false;
 668    }
 669 }
 670
 671 ppir_instr *ppir_instr_create(ppir_block *block);
 672 bool ppir_instr_insert_node(ppir_instr *instr, ppir_node *node);
 673 void ppir_instr_add_dep(ppir_instr *succ, ppir_instr *pred);
 674 void ppir_instr_print_list(ppir_compiler *comp);
 675 void ppir_instr_print_dep(ppir_compiler *comp);
 676 void ppir_instr_insert_mul_node(ppir_node *add, ppir_node *mul);
 677
 678 #define ppir_instr_foreach_succ(instr, dep) \
 679    list_for_each_entry(ppir_dep, dep, &instr->succ_list, succ_link)
 680 #define ppir_instr_foreach_succ_safe(instr, dep) \
 681    list_for_each_entry_safe(ppir_dep, dep, &instr->succ_list, succ_link)
 682 #define ppir_instr_foreach_pred(instr, dep) \
 683    list_for_each_entry(ppir_dep, dep, &instr->pred_list, pred_link)
 684 #define ppir_instr_foreach_pred_safe(instr, dep) \
 685    list_for_each_entry_safe(ppir_dep, dep, &instr->pred_list, pred_link)
 686
 687 static inline bool ppir_instr_is_root(ppir_instr *instr)
 688 {
 689    return list_is_empty(&instr->succ_list);
 690 }
 691
 692 static inline bool ppir_instr_is_leaf(ppir_instr *instr)
 693 {
 694    return list_is_empty(&instr->pred_list);
 695 }
 696
 697 bool ppir_lower_prog(ppir_compiler *comp);
 698 bool ppir_node_to_instr(ppir_compiler *comp);
 699 bool ppir_schedule_prog(ppir_compiler *comp);
 700 bool ppir_regalloc_prog(ppir_compiler *comp);
 701 bool ppir_codegen_prog(ppir_compiler *comp);
 702 void ppir_liveness_analysis(ppir_compiler *comp);
 703
 704 #endif