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
  31 #include "ir/lima_ir.h"
  32
  33 typedef enum {
  34    ppir_op_mov,
  35    ppir_op_abs,
  36    ppir_op_neg,
  37    ppir_op_sat,
  38    ppir_op_add,
  39
  40    ppir_op_ddx,
  41    ppir_op_ddy,
  42
  43    ppir_op_mul,
  44    ppir_op_rcp,
  45
  46    ppir_op_sin_lut,
  47    ppir_op_cos_lut,
  48
  49    ppir_op_sum3,
  50    ppir_op_sum4,
  51
  52    ppir_op_normalize2,
  53    ppir_op_normalize3,
  54    ppir_op_normalize4,
  55
  56    ppir_op_sel_cond,
  57    ppir_op_select,
  58
  59    ppir_op_sin,
  60    ppir_op_cos,
  61    ppir_op_tan,
  62    ppir_op_asin,
  63    ppir_op_acos,
  64
  65    ppir_op_atan,
  66    ppir_op_atan2,
  67    ppir_op_atan_pt1,
  68    ppir_op_atan2_pt1,
  69    ppir_op_atan_pt2,
  70
  71    ppir_op_exp,
  72    ppir_op_log,
  73    ppir_op_exp2,
  74    ppir_op_log2,
  75    ppir_op_sqrt,
  76    ppir_op_rsqrt,
  77
  78    ppir_op_sign,
  79    ppir_op_floor,
  80    ppir_op_ceil,
  81    ppir_op_fract,
  82    ppir_op_mod,
  83    ppir_op_min,
  84    ppir_op_max,
  85    ppir_op_trunc,
  86
  87    ppir_op_and,
  88    ppir_op_or,
  89    ppir_op_xor,
  90
  91    ppir_op_lt,
  92    ppir_op_gt,
  93    ppir_op_le,
  94    ppir_op_ge,
  95    ppir_op_eq,
  96    ppir_op_ne,
  97    ppir_op_not,
  98
  99    ppir_op_load_uniform,
 100    ppir_op_load_varying,
 101    ppir_op_load_coords,
 102    ppir_op_load_coords_reg,
 103    ppir_op_load_fragcoord,
 104    ppir_op_load_pointcoord,
 105    ppir_op_load_frontface,
 106    ppir_op_load_texture,
 107    ppir_op_load_temp,
 108
 109    ppir_op_store_temp,
 110    ppir_op_store_color,
 111
 112    ppir_op_const,
 113
 114    ppir_op_discard,
 115    ppir_op_branch,
 116
 117    ppir_op_undef,
 118
 119    ppir_op_num,
 120 } ppir_op;
 121
 122 typedef enum {
 123    ppir_node_type_alu,
 124    ppir_node_type_const,
 125    ppir_node_type_load,
 126    ppir_node_type_store,
 127    ppir_node_type_load_texture,
 128    ppir_node_type_discard,
 129    ppir_node_type_branch,
 130 } ppir_node_type;
 131
 132 typedef struct {
 133    char *name;
 134    ppir_node_type type;
 135    int *slots;
 136 } ppir_op_info;
 137
 138 extern const ppir_op_info ppir_op_infos[];
 139
 140 typedef enum {
 141    ppir_dep_src,
 142    ppir_dep_write_after_read,
 143    ppir_dep_sequence,
 144 } ppir_dep_type;
 145
 146 typedef struct {
 147    void *pred, *succ;
 148    ppir_dep_type type;
 149    struct list_head pred_link;
 150    struct list_head succ_link;
 151 } ppir_dep;
 152
 153 typedef struct ppir_node {
 154    struct list_head list;
 155    ppir_op op;
 156    ppir_node_type type;
 157    int index;
 158    char name[16];
 159    bool printed;
 160    struct ppir_instr *instr;
 161    int instr_pos;
 162    struct ppir_block *block;
 163
 164    /* for scheduler */
 165    struct list_head succ_list;
 166    struct list_head pred_list;
 167 } ppir_node;
 168
 169 typedef enum {
 170    ppir_pipeline_reg_const0,
 171    ppir_pipeline_reg_const1,
 172    ppir_pipeline_reg_sampler,
 173    ppir_pipeline_reg_uniform,
 174    ppir_pipeline_reg_vmul,
 175    ppir_pipeline_reg_fmul,
 176    ppir_pipeline_reg_discard, /* varying load */
 177 } ppir_pipeline;
 178
 179 typedef struct ppir_reg {
 180    struct list_head list;
 181    int index;
 182    int regalloc_index;
 183    int num_components;
 184    /* whether this reg has to start from the x component
 185     * of a full physical reg, this is true for reg used
 186     * in load/store instr which has no swizzle field
 187     */
 188    bool is_head;
 189    /* instr live range */
 190    int live_in, live_out;
 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 typedef struct ppir_instr {
 304    struct list_head list;
 305    int index;
 306    bool printed;
 307    int seq; /* command sequence after schedule */
 308
 309    ppir_node *slots[PPIR_INSTR_SLOT_NUM];
 310    ppir_const constant[2];
 311    bool is_end;
 312
 313    /* for scheduler */
 314    struct list_head succ_list;
 315    struct list_head pred_list;
 316    float reg_pressure;
 317    int est; /* earliest start time */
 318    int parent_index;
 319    bool scheduled;
 320    int offset;
 321    int encode_size;
 322 } ppir_instr;
 323
 324 typedef struct ppir_block {
 325    struct list_head list;
 326    struct list_head node_list;
 327    struct list_head instr_list;
 328
 329    struct ppir_block *successors[2];
 330
 331    struct ppir_compiler *comp;
 332
 333    /* for scheduler */
 334    int sched_instr_index;
 335    int sched_instr_base;
 336    int index;
 337
 338    /*  for liveness analysis */
 339    BITSET_WORD *def;
 340    BITSET_WORD *use;
 341    BITSET_WORD *live_in;
 342    BITSET_WORD *live_out;
 343 } ppir_block;
 344
 345 typedef struct {
 346    ppir_node node;
 347    ppir_src src[2];
 348    int num_src;
 349    bool cond_gt;
 350    bool cond_eq;
 351    bool cond_lt;
 352    bool negate;
 353    ppir_block *target;
 354 } ppir_branch_node;
 355
 356 struct ra_regs;
 357 struct lima_fs_shader_state;
 358
 359 typedef struct ppir_compiler {
 360    struct list_head block_list;
 361    struct hash_table_u64 *blocks;
 362    int cur_index;
 363    int cur_instr_index;
 364
 365    struct list_head reg_list;
 366
 367    /* array for searching ssa/reg node */
 368    ppir_node **var_nodes;
 369    unsigned reg_base;
 370
 371    struct ra_regs *ra;
 372    struct lima_fs_shader_state *prog;
 373
 374    /* for scheduler */
 375    int sched_instr_base;
 376
 377    /* for regalloc spilling debug */
 378    int force_spilling;
 379
 380    /* shaderdb */
 381    int num_loops;
 382    int num_spills;
 383    int num_fills;
 384
 385    ppir_block *discard_block;
 386    ppir_block *current_block;
 387    ppir_block *loop_break_block;
 388    ppir_block *loop_cont_block;
 389 } ppir_compiler;
 390
 391 void *ppir_node_create(ppir_block *block, ppir_op op, int index, unsigned mask);
 392 void ppir_node_add_dep(ppir_node *succ, ppir_node *pred, ppir_dep_type type);
 393 void ppir_node_remove_dep(ppir_dep *dep);
 394 void ppir_node_delete(ppir_node *node);
 395 void ppir_node_print_prog(ppir_compiler *comp);
 396 void ppir_node_replace_child(ppir_node *parent, ppir_node *old_child, ppir_node *new_child);
 397 void ppir_node_replace_all_succ(ppir_node *dst, ppir_node *src);
 398 void ppir_node_replace_pred(ppir_dep *dep, ppir_node *new_pred);
 399 ppir_dep *ppir_dep_for_pred(ppir_node *node, ppir_node *pred);
 400 ppir_node *ppir_node_clone(ppir_block *block, ppir_node *node);
 401 /* Assumes that node successors are in the same block */
 402 ppir_node *ppir_node_insert_mov(ppir_node *node);
 403
 404 static inline bool ppir_node_is_root(ppir_node *node)
 405 {
 406    return list_is_empty(&node->succ_list);
 407 }
 408
 409 static inline bool ppir_node_is_leaf(ppir_node *node)
 410 {
 411    return list_is_empty(&node->pred_list);
 412 }
 413
 414 static inline bool ppir_node_has_single_succ(ppir_node *node)
 415 {
 416    return list_is_singular(&node->succ_list);
 417 }
 418
 419 bool ppir_node_has_single_src_succ(ppir_node *node);
 420
 421 static inline ppir_node *ppir_node_first_succ(ppir_node *node)
 422 {
 423    return list_first_entry(&node->succ_list, ppir_dep, succ_link)->succ;
 424 }
 425
 426 static inline bool ppir_node_has_single_pred(ppir_node *node)
 427 {
 428    return list_is_singular(&node->pred_list);
 429 }
 430
 431 static inline ppir_node *ppir_node_first_pred(ppir_node *node)
 432 {
 433    return list_first_entry(&node->pred_list, ppir_dep, pred_link)->pred;
 434 }
 435
 436 #define ppir_node_foreach_succ(node, dep) \
 437    list_for_each_entry(ppir_dep, dep, &node->succ_list, succ_link)
 438 #define ppir_node_foreach_succ_safe(node, dep) \
 439    list_for_each_entry_safe(ppir_dep, dep, &node->succ_list, succ_link)
 440 #define ppir_node_foreach_pred(node, dep) \
 441    list_for_each_entry(ppir_dep, dep, &node->pred_list, pred_link)
 442 #define ppir_node_foreach_pred_safe(node, dep) \
 443    list_for_each_entry_safe(ppir_dep, dep, &node->pred_list, pred_link)
 444
 445 #define ppir_node_to_alu(node) ((ppir_alu_node *)(node))
 446 #define ppir_node_to_const(node) ((ppir_const_node *)(node))
 447 #define ppir_node_to_load(node) ((ppir_load_node *)(node))
 448 #define ppir_node_to_store(node) ((ppir_store_node *)(node))
 449 #define ppir_node_to_load_texture(node) ((ppir_load_texture_node *)(node))
 450 #define ppir_node_to_discard(node) ((ppir_discard_node *)(node))
 451 #define ppir_node_to_branch(node) ((ppir_branch_node *)(node))
 452
 453 static inline ppir_dest *ppir_node_get_dest(ppir_node *node)
 454 {
 455    switch (node->type) {
 456    case ppir_node_type_alu:
 457       return &ppir_node_to_alu(node)->dest;
 458    case ppir_node_type_load:
 459       return &ppir_node_to_load(node)->dest;
 460    case ppir_node_type_const:
 461       return &ppir_node_to_const(node)->dest;
 462    case ppir_node_type_load_texture:
 463       return &ppir_node_to_load_texture(node)->dest;
 464    default:
 465       return NULL;
 466    }
 467 }
 468
 469 static inline int ppir_node_get_src_num(ppir_node *node)
 470 {
 471    switch (node->type) {
 472    case ppir_node_type_alu:
 473       return ppir_node_to_alu(node)->num_src;
 474    case ppir_node_type_branch:
 475       return ppir_node_to_branch(node)->num_src;
 476    case ppir_node_type_load:
 477       return ppir_node_to_load(node)->num_src;
 478    case ppir_node_type_load_texture:
 479       return ppir_node_to_load_texture(node)->num_src;
 480    case ppir_node_type_store:
 481       return 1;
 482    default:
 483       return 0;
 484    }
 485
 486    return 0;
 487 }
 488
 489 static inline ppir_src *ppir_node_get_src(ppir_node *node, int idx)
 490 {
 491    if (idx < 0 || idx >= ppir_node_get_src_num(node))
 492       return NULL;
 493
 494    switch (node->type) {
 495    case ppir_node_type_alu:
 496       return &ppir_node_to_alu(node)->src[idx];
 497    case ppir_node_type_branch:
 498       return &ppir_node_to_branch(node)->src[idx];
 499    case ppir_node_type_load_texture:
 500       return &ppir_node_to_load_texture(node)->src[idx];
 501    case ppir_node_type_load:
 502       return &ppir_node_to_load(node)->src;
 503    case ppir_node_type_store:
 504       return &ppir_node_to_store(node)->src;
 505    default:
 506       break;
 507    }
 508
 509    return NULL;
 510 }
 511
 512 static inline ppir_reg *ppir_src_get_reg(ppir_src *src)
 513 {
 514    switch (src->type) {
 515    case ppir_target_ssa:
 516       return src->ssa;
 517    case ppir_target_register:
 518       return src->reg;
 519    default:
 520       return NULL;
 521    }
 522 }
 523
 524 static inline ppir_reg *ppir_dest_get_reg(ppir_dest *dest)
 525 {
 526    switch (dest->type) {
 527    case ppir_target_ssa:
 528       return &dest->ssa;
 529    case ppir_target_register:
 530       return dest->reg;
 531    default:
 532       return NULL;
 533    }
 534 }
 535
 536 static inline ppir_src *ppir_node_get_src_for_pred(ppir_node *node, ppir_node *pred)
 537 {
 538    for (int i = 0; i < ppir_node_get_src_num(node); i++) {
 539       ppir_src *src = ppir_node_get_src(node, i);
 540       if (src && src->node == pred)
 541          return src;
 542    }
 543
 544    return NULL;
 545 }
 546
 547 static inline void ppir_node_target_assign(ppir_src *src, ppir_node *node)
 548 {
 549    ppir_dest *dest = ppir_node_get_dest(node);
 550    src->type = dest->type;
 551    switch (src->type) {
 552    case ppir_target_ssa:
 553       src->ssa = &dest->ssa;
 554       src->node = node;
 555       break;
 556    case ppir_target_register:
 557       src->reg = dest->reg;
 558       /* Registers can be assigned from multiple nodes, so don't keep
 559        * pointer to the node here
 560        */
 561       src->node = NULL;
 562       break;
 563    case ppir_target_pipeline:
 564       src->pipeline = dest->pipeline;
 565       src->node = node;
 566       break;
 567    }
 568 }
 569
 570 static inline bool ppir_node_target_equal(ppir_src *src, ppir_dest *dest)
 571 {
 572    if (src->type != dest->type ||
 573        (src->type == ppir_target_ssa && src->ssa != &dest->ssa) ||
 574        (src->type == ppir_target_register && src->reg != dest->reg) ||
 575        (src->type == ppir_target_pipeline && src->pipeline != dest->pipeline))
 576       return false;
 577
 578    return true;
 579 }
 580
 581 static inline int ppir_target_get_src_reg_index(ppir_src *src)
 582 {
 583    switch (src->type) {
 584    case ppir_target_ssa:
 585       if (src->ssa)
 586          return src->ssa->index;
 587       break;
 588    case ppir_target_register:
 589       if (src->reg)
 590          return src->reg->index;
 591       break;
 592    case ppir_target_pipeline:
 593       if (src->pipeline == ppir_pipeline_reg_discard)
 594          return 15 * 4;
 595       return (src->pipeline + 12) * 4;
 596    }
 597
 598    return -1;
 599 }
 600
 601 static inline int ppir_target_get_dest_reg_index(ppir_dest *dest)
 602 {
 603    switch (dest->type) {
 604    case ppir_target_ssa:
 605       return dest->ssa.index;
 606    case ppir_target_register:
 607       return dest->reg->index;
 608    case ppir_target_pipeline:
 609       if (dest->pipeline == ppir_pipeline_reg_discard)
 610          return 15 * 4;
 611       return (dest->pipeline + 12) * 4;
 612    }
 613
 614    return -1;
 615 }
 616
 617 static inline bool ppir_target_is_scaler(ppir_dest *dest)
 618 {
 619    switch (dest->type) {
 620    case ppir_target_ssa:
 621       return dest->ssa.num_components == 1;
 622    case ppir_target_register:
 623       /* only one bit in mask is set */
 624       if ((dest->write_mask & 0x3) == 0x3 ||
 625           (dest->write_mask & 0x5) == 0x5 ||
 626           (dest->write_mask & 0x9) == 0x9 ||
 627           (dest->write_mask & 0x6) == 0x6 ||
 628           (dest->write_mask & 0xa) == 0xa ||
 629           (dest->write_mask & 0xc) == 0xc)
 630          return false;
 631       else
 632          return true;
 633    case ppir_target_pipeline:
 634       if (dest->pipeline == ppir_pipeline_reg_fmul)
 635          return true;
 636       else
 637          return false;
 638    default:
 639       return false;
 640    }
 641 }
 642
 643 ppir_instr *ppir_instr_create(ppir_block *block);
 644 bool ppir_instr_insert_node(ppir_instr *instr, ppir_node *node);
 645 void ppir_instr_add_dep(ppir_instr *succ, ppir_instr *pred);
 646 void ppir_instr_print_list(ppir_compiler *comp);
 647 void ppir_instr_print_dep(ppir_compiler *comp);
 648 void ppir_instr_insert_mul_node(ppir_node *add, ppir_node *mul);
 649
 650 #define ppir_instr_foreach_succ(instr, dep) \
 651    list_for_each_entry(ppir_dep, dep, &instr->succ_list, succ_link)
 652 #define ppir_instr_foreach_succ_safe(instr, dep) \
 653    list_for_each_entry_safe(ppir_dep, dep, &instr->succ_list, succ_link)
 654 #define ppir_instr_foreach_pred(instr, dep) \
 655    list_for_each_entry(ppir_dep, dep, &instr->pred_list, pred_link)
 656 #define ppir_instr_foreach_pred_safe(instr, dep) \
 657    list_for_each_entry_safe(ppir_dep, dep, &instr->pred_list, pred_link)
 658
 659 static inline bool ppir_instr_is_root(ppir_instr *instr)
 660 {
 661    return list_is_empty(&instr->succ_list);
 662 }
 663
 664 static inline bool ppir_instr_is_leaf(ppir_instr *instr)
 665 {
 666    return list_is_empty(&instr->pred_list);
 667 }
 668
 669 bool ppir_lower_prog(ppir_compiler *comp);
 670 bool ppir_node_to_instr(ppir_compiler *comp);
 671 bool ppir_schedule_prog(ppir_compiler *comp);
 672 bool ppir_regalloc_prog(ppir_compiler *comp);
 673 bool ppir_codegen_prog(ppir_compiler *comp);
 674 void ppir_liveness_analysis(ppir_compiler *comp);
 675
 676 #endif