src/gallium/drivers/nv50/nv50_pc.h

   1 /*
   2  * Copyright 2010 Christoph Bumiller
   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 shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
  18  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
  19  * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  20  * SOFTWARE.
  21  */
  22
  23 #ifndef __NV50_COMPILER_H__
  24 #define __NV50_COMPILER_H__
  25
  26 #include "nv50_debug.h"
  27
  28 #include "pipe/p_defines.h"
  29 #include "util/u_inlines.h"
  30 #include "util/u_memory.h"
  31
  32 #define NV_OP_PHI       0
  33 #define NV_OP_EXTRACT   1
  34 #define NV_OP_COMBINE   2
  35 #define NV_OP_LDA       3
  36 #define NV_OP_STA       4
  37 #define NV_OP_MOV       5
  38 #define NV_OP_ADD       6
  39 #define NV_OP_SUB       7
  40 #define NV_OP_NEG       8
  41 #define NV_OP_MUL       9
  42 #define NV_OP_MAD       10
  43 #define NV_OP_CVT       11
  44 #define NV_OP_SAT       12
  45 #define NV_OP_NOT       13
  46 #define NV_OP_AND       14
  47 #define NV_OP_OR        15
  48 #define NV_OP_XOR       16
  49 #define NV_OP_SHL       17
  50 #define NV_OP_SHR       18
  51 #define NV_OP_RCP       19
  52 #define NV_OP_UNDEF     20
  53 #define NV_OP_RSQ       21
  54 #define NV_OP_LG2       22
  55 #define NV_OP_SIN       23
  56 #define NV_OP_COS       24
  57 #define NV_OP_EX2       25
  58 #define NV_OP_PRESIN    26
  59 #define NV_OP_PREEX2    27
  60 #define NV_OP_MIN       28
  61 #define NV_OP_MAX       29
  62 #define NV_OP_SET       30
  63 #define NV_OP_SAD       31
  64 #define NV_OP_KIL       32
  65 #define NV_OP_BRA       33
  66 #define NV_OP_CALL      34
  67 #define NV_OP_RET       35
  68 #define NV_OP_BREAK     36
  69 #define NV_OP_BREAKADDR 37
  70 #define NV_OP_JOINAT    38
  71 #define NV_OP_TEX       39
  72 #define NV_OP_TXB       40
  73 #define NV_OP_TXL       41
  74 #define NV_OP_TXF       42
  75 #define NV_OP_TXQ       43
  76 #define NV_OP_DFDX      44
  77 #define NV_OP_DFDY      45
  78 #define NV_OP_QUADOP    46
  79 #define NV_OP_LINTERP   47
  80 #define NV_OP_PINTERP   48
  81 #define NV_OP_ABS       49
  82 #define NV_OP_CEIL      50
  83 #define NV_OP_FLOOR     51
  84 #define NV_OP_TRUNC     52
  85 #define NV_OP_NOP       53
  86 #define NV_OP_SELECT    54
  87 #define NV_OP_EXPORT    55
  88 #define NV_OP_JOIN      56
  89 #define NV_OP_COUNT     57
  90
  91 #define NV_FILE_GPR      0
  92 #define NV_FILE_OUT      1
  93 #define NV_FILE_ADDR     2
  94 #define NV_FILE_FLAGS    3
  95 #define NV_FILE_IMM      16
  96 #define NV_FILE_MEM_S    32
  97 #define NV_FILE_MEM_P    33
  98 #define NV_FILE_MEM_V    34
  99 #define NV_FILE_MEM_L    48
 100 #define NV_FILE_MEM_G(i) (64 + i)
 101 #define NV_FILE_MEM_C(i) (80 + i)
 102
 103 #define NV_MOD_NEG 1
 104 #define NV_MOD_ABS 2
 105 #define NV_MOD_NOT 4
 106 #define NV_MOD_SAT 8
 107
 108 #define NV_TYPE_U8  0x00
 109 #define NV_TYPE_S8  0x01
 110 #define NV_TYPE_U16 0x02
 111 #define NV_TYPE_S16 0x03
 112 #define NV_TYPE_U32 0x04
 113 #define NV_TYPE_S32 0x05
 114 #define NV_TYPE_P32 0x07
 115 #define NV_TYPE_F32 0x09
 116 #define NV_TYPE_F64 0x0b
 117 #define NV_TYPE_VEC(x, n) (NV_TYPE_##x | (n << 4))
 118 #define NV_TYPE_LO  0x00
 119 #define NV_TYPE_HI  0x80
 120 #define NV_TYPE_ANY 0xff
 121
 122 #define NV_TYPE_ISINT(t) ((t) <= 5)
 123 #define NV_TYPE_ISFLT(t) ((t) & 0x08)
 124
 125 /* $cX registers contain 4 bits: OCSZ (Z is bit 0) */
 126 #define NV_CC_FL 0x0
 127 #define NV_CC_LT 0x1
 128 #define NV_CC_EQ 0x2
 129 #define NV_CC_LE 0x3
 130 #define NV_CC_GT 0x4
 131 #define NV_CC_NE 0x5
 132 #define NV_CC_GE 0x6
 133 #define NV_CC_U  0x8
 134 #define NV_CC_TR 0xf
 135 #define NV_CC_O  0x10
 136 #define NV_CC_C  0x11
 137 #define NV_CC_A  0x12
 138 #define NV_CC_S  0x13
 139
 140 #define NV_PC_MAX_INSTRUCTIONS 2048
 141 #define NV_PC_MAX_VALUES (NV_PC_MAX_INSTRUCTIONS * 4)
 142
 143 #define NV_PC_MAX_BASIC_BLOCKS 1024
 144
 145 static INLINE boolean
 146 nv_is_vector_op(uint opcode)
 147 {
 148    return (opcode >= NV_OP_TEX) && (opcode <= NV_OP_TXQ);
 149 }
 150
 151 static INLINE uint
 152 nv_type_order(ubyte type)
 153 {
 154    switch (type & 0xf) {
 155    case NV_TYPE_U8:
 156    case NV_TYPE_S8:
 157       return 0;
 158    case NV_TYPE_U16:
 159    case NV_TYPE_S16:
 160       return 1;
 161    case NV_TYPE_U32:
 162    case NV_TYPE_F32:
 163    case NV_TYPE_S32:
 164    case NV_TYPE_P32:
 165       return 2;
 166    case NV_TYPE_F64:
 167       return 3;
 168    }
 169    assert(0);
 170    return 0;
 171 }
 172
 173 static INLINE uint
 174 nv_type_sizeof(ubyte type)
 175 {
 176    if (type & 0xf0)
 177       return (1 << nv_type_order(type)) * (type >> 4);
 178    return 1 << nv_type_order(type);
 179 }
 180
 181 static INLINE uint
 182 nv_type_sizeof_base(ubyte type)
 183 {
 184    return 1 << nv_type_order(type);
 185 }
 186
 187 struct nv_reg {
 188    int id;
 189    ubyte file;
 190    ubyte type; /* type of generating instruction's result */
 191    ubyte as_type; /* default type for new references to this value */
 192    union {
 193       float f32;
 194       double f64;
 195       int32_t s32;
 196       uint32_t u32;
 197    } imm;
 198 };
 199
 200 struct nv_range {
 201    struct nv_range *next;
 202    int bgn;
 203    int end;
 204 };
 205
 206 struct nv_value {
 207    struct nv_reg reg;
 208    struct nv_instruction *insn;
 209    struct nv_value *join;
 210    int n;
 211    struct nv_range *livei;
 212    int refc;
 213
 214    struct nv_value *next;
 215    struct nv_value *prev;
 216 };
 217
 218 struct nv_ref {
 219    struct nv_value *value;
 220    ubyte mod;
 221    ubyte typecast;
 222    ubyte flags; /* not used yet */
 223 };
 224
 225 #define NV_REF_FLAG_REGALLOC_PRIV (1 << 0)
 226
 227 struct nv_basic_block;
 228
 229 struct nv_instruction {
 230    struct nv_instruction *next;
 231    struct nv_instruction *prev;
 232    uint opcode;
 233    int serial;
 234    struct nv_value *def[4];
 235    struct nv_value *flags_def;
 236    struct nv_ref *src[5];
 237    struct nv_ref *flags_src;
 238    struct nv_basic_block *bb;
 239    struct nv_basic_block *target; /* target block of control flow insn */
 240    ubyte cc;
 241    unsigned set_cond      : 4;
 242    unsigned fixed         : 1; /* don't optimize away */
 243    unsigned is_terminator : 1;
 244    unsigned is_join       : 1;
 245    unsigned is_long       : 1; /* for emission */
 246    /* */
 247    unsigned saturate : 1;
 248    unsigned centroid : 1;
 249    unsigned flat     : 1;
 250    unsigned lanes    : 4;
 251    unsigned tex_live : 1;
 252    /* */
 253    ubyte tex_t; /* TIC binding */
 254    ubyte tex_s; /* TSC binding */
 255    unsigned tex_argc : 3;
 256    unsigned tex_cube : 1;
 257    unsigned tex_mask : 4;
 258    /* */
 259    ubyte quadop;
 260 };
 261
 262 static INLINE int
 263 nvi_vector_size(struct nv_instruction *nvi)
 264 {
 265    int i;
 266    assert(nvi);
 267    for (i = 0; i < 4 && nvi->def[i]; ++i);
 268    return i;
 269 }
 270
 271 #define CFG_EDGE_FORWARD     0
 272 #define CFG_EDGE_BACK        1
 273 #define CFG_EDGE_LOOP_ENTER  2
 274 #define CFG_EDGE_LOOP_LEAVE  4
 275 #define CFG_EDGE_FAKE        8
 276
 277 /* 'WALL' edge means where reachability check doesn't follow */
 278 /* 'LOOP' edge means just having to do with loops */
 279 #define IS_LOOP_EDGE(k) ((k) & 7)
 280 #define IS_WALL_EDGE(k) ((k) & 9)
 281
 282 struct nv_basic_block {
 283    struct nv_instruction *entry; /* first non-phi instruction */
 284    struct nv_instruction *exit;
 285    struct nv_instruction *phi; /* very first instruction */
 286    int num_instructions;
 287
 288    struct nv_basic_block *out[2]; /* no indirect branches -> 2 */
 289    struct nv_basic_block *in[8]; /* hope that suffices */
 290    uint num_in;
 291    ubyte out_kind[2];
 292    ubyte in_kind[8];
 293
 294    int id;
 295    int subroutine;
 296    uint priv; /* reset to 0 after you're done */
 297    uint pass_seq;
 298
 299    uint32_t bin_pos; /* position, size in emitted code */
 300    uint32_t bin_size;
 301
 302    uint32_t live_set[NV_PC_MAX_VALUES / 32];
 303 };
 304
 305 struct nv50_translation_info;
 306
 307 struct nv_pc {
 308    struct nv_basic_block **root;
 309    struct nv_basic_block *current_block;
 310    struct nv_basic_block *parent_block;
 311
 312    int loop_nesting_bound;
 313    uint pass_seq;
 314
 315    struct nv_value values[NV_PC_MAX_VALUES];
 316    struct nv_instruction instructions[NV_PC_MAX_INSTRUCTIONS];
 317    struct nv_ref **refs;
 318    struct nv_basic_block *bb_list[NV_PC_MAX_BASIC_BLOCKS];
 319    int num_values;
 320    int num_instructions;
 321    int num_refs;
 322    int num_blocks;
 323    int num_subroutines;
 324
 325    int max_reg[4];
 326
 327    uint32_t *immd_buf; /* populated on emit */
 328    unsigned immd_count;
 329
 330    uint32_t *emit;
 331    unsigned bin_size;
 332    unsigned bin_pos;
 333
 334    void *fixups;
 335    unsigned num_fixups;
 336
 337    /* optimization enables */
 338    boolean opt_reload_elim;
 339 };
 340
 341 void nvbb_insert_tail(struct nv_basic_block *, struct nv_instruction *);
 342 void nvi_insert_after(struct nv_instruction *, struct nv_instruction *);
 343
 344 static INLINE struct nv_instruction *
 345 nv_alloc_instruction(struct nv_pc *pc, uint opcode)
 346 {
 347    struct nv_instruction *insn;
 348
 349    insn = &pc->instructions[pc->num_instructions++];
 350    assert(pc->num_instructions < NV_PC_MAX_INSTRUCTIONS);
 351
 352    insn->cc = NV_CC_TR;
 353    insn->opcode = opcode;
 354
 355    return insn;
 356 }
 357
 358 static INLINE struct nv_instruction *
 359 new_instruction(struct nv_pc *pc, uint opcode)
 360 {
 361    struct nv_instruction *insn = nv_alloc_instruction(pc, opcode);
 362
 363    nvbb_insert_tail(pc->current_block, insn);
 364    return insn;
 365 }
 366
 367 static INLINE struct nv_instruction *
 368 new_instruction_at(struct nv_pc *pc, struct nv_instruction *at, uint opcode)
 369 {
 370    struct nv_instruction *insn = nv_alloc_instruction(pc, opcode);
 371
 372    nvi_insert_after(at, insn);
 373    return insn;
 374 }
 375
 376 static INLINE struct nv_value *
 377 new_value(struct nv_pc *pc, ubyte file, ubyte type)
 378 {
 379    struct nv_value *value = &pc->values[pc->num_values];
 380
 381    assert(pc->num_values < NV_PC_MAX_VALUES - 1);
 382
 383    value->n = pc->num_values++;
 384    value->join = value;
 385    value->reg.id = -1;
 386    value->reg.file = file;
 387    value->reg.type = value->reg.as_type = type;
 388    return value;
 389 }
 390
 391 static INLINE struct nv_value *
 392 new_value_like(struct nv_pc *pc, struct nv_value *like)
 393 {
 394    struct nv_value *val = new_value(pc, like->reg.file, like->reg.type);
 395    val->reg.as_type = like->reg.as_type;
 396    return val;
 397 }
 398
 399 static INLINE struct nv_ref *
 400 new_ref(struct nv_pc *pc, struct nv_value *val)
 401 {
 402    int i;
 403    struct nv_ref *ref;
 404
 405    if ((pc->num_refs % 64) == 0) {
 406       const unsigned old_size = pc->num_refs * sizeof(struct nv_ref *);
 407       const unsigned new_size = (pc->num_refs + 64) * sizeof(struct nv_ref *);
 408
 409       pc->refs = REALLOC(pc->refs, old_size, new_size);
 410
 411       ref = CALLOC(64, sizeof(struct nv_ref));
 412       for (i = 0; i < 64; ++i)
 413          pc->refs[pc->num_refs + i] = &ref[i];
 414    }
 415
 416    ref = pc->refs[pc->num_refs++];
 417    ref->value = val;
 418    ref->typecast = val->reg.as_type;
 419
 420    ++val->refc;
 421    return ref;
 422 }
 423
 424 static INLINE struct nv_basic_block *
 425 new_basic_block(struct nv_pc *pc)
 426 {
 427    struct nv_basic_block *bb;
 428
 429    if (pc->num_blocks >= NV_PC_MAX_BASIC_BLOCKS)
 430       return NULL;
 431
 432    bb = CALLOC_STRUCT(nv_basic_block);
 433
 434    bb->id = pc->num_blocks;
 435    pc->bb_list[pc->num_blocks++] = bb;
 436    return bb;
 437 }
 438
 439 static INLINE void
 440 nv_reference(struct nv_pc *pc, struct nv_ref **d, struct nv_value *s)
 441 {
 442    if (*d)
 443       --(*d)->value->refc;
 444
 445    if (s) {
 446       if (!*d)
 447          *d = new_ref(pc, s);
 448       else {
 449          (*d)->value = s;
 450          ++(s->refc);
 451       }
 452    } else {
 453       *d = NULL;
 454    }
 455 }
 456
 457 /* nv50_emit.c */
 458 void nv50_emit_instruction(struct nv_pc *, struct nv_instruction *);
 459 unsigned nv50_inst_min_size(struct nv_instruction *);
 460
 461 /* nv50_print.c */
 462 const char *nv_opcode_name(uint opcode);
 463 void nv_print_instruction(struct nv_instruction *);
 464
 465 /* nv50_pc.c */
 466
 467 void nv_print_function(struct nv_basic_block *root);
 468 void nv_print_program(struct nv_pc *);
 469
 470 boolean nv_op_commutative(uint opcode);
 471 int nv50_indirect_opnd(struct nv_instruction *);
 472 boolean nv50_nvi_can_use_imm(struct nv_instruction *, int s);
 473 boolean nv50_nvi_can_predicate(struct nv_instruction *);
 474 boolean nv50_nvi_can_load(struct nv_instruction *, int s, struct nv_value *);
 475 boolean nv50_op_can_write_flags(uint opcode);
 476 ubyte nv50_supported_src_mods(uint opcode, int s);
 477 int nv_nvi_refcount(struct nv_instruction *);
 478 void nv_nvi_delete(struct nv_instruction *);
 479 void nv_nvi_permute(struct nv_instruction *, struct nv_instruction *);
 480 void nvbb_attach_block(struct nv_basic_block *parent,
 481                        struct nv_basic_block *, ubyte edge_kind);
 482 boolean nvbb_dominated_by(struct nv_basic_block *, struct nv_basic_block *);
 483 boolean nvbb_reachable_by(struct nv_basic_block *, struct nv_basic_block *,
 484                           struct nv_basic_block *);
 485 struct nv_basic_block *nvbb_dom_frontier(struct nv_basic_block *);
 486 int nvcg_replace_value(struct nv_pc *pc, struct nv_value *old_val,
 487                        struct nv_value *new_val);
 488 struct nv_value *nvcg_find_immediate(struct nv_ref *);
 489 struct nv_value *nvcg_find_constant(struct nv_ref *);
 490
 491 typedef void (*nv_pc_pass_func)(void *priv, struct nv_basic_block *b);
 492
 493 void nv_pc_pass_in_order(struct nv_basic_block *, nv_pc_pass_func, void *);
 494
 495 int nv_pc_exec_pass0(struct nv_pc *pc);
 496 int nv_pc_exec_pass1(struct nv_pc *pc);
 497 int nv_pc_exec_pass2(struct nv_pc *pc);
 498
 499 int nv50_tgsi_to_nc(struct nv_pc *, struct nv50_translation_info *);
 500
 501 #endif // NV50_COMPILER_H