1 //===-- SparcInternals.h - Header file for Sparc backend ---------*- C++ -*--=//
3 // This file defines stuff that is to be private to the Sparc backend, but is
4 // shared among different portions of the backend.
6 //===----------------------------------------------------------------------===//
8 #ifndef SPARC_INTERNALS_H
9 #define SPARC_INTERNALS_H
11 #include "SparcRegInfo.h"
12 #include "llvm/Target/MachineSchedInfo.h"
13 #include "llvm/Type.h"
15 #include <sys/types.h>
19 // OpCodeMask definitions for the Sparc V9
21 const OpCodeMask Immed = 0x00002000; // immed or reg operand?
22 const OpCodeMask Annul = 0x20000000; // annul delay instr?
23 const OpCodeMask PredictTaken = 0x00080000; // predict branch taken?
26 enum SparcInstrSchedClass {
27 SPARC_NONE, /* Instructions with no scheduling restrictions */
28 SPARC_IEUN, /* Integer class that can use IEU0 or IEU1 */
29 SPARC_IEU0, /* Integer class IEU0 */
30 SPARC_IEU1, /* Integer class IEU1 */
31 SPARC_FPM, /* FP Multiply or Divide instructions */
32 SPARC_FPA, /* All other FP instructions */
33 SPARC_CTI, /* Control-transfer instructions */
34 SPARC_LD, /* Load instructions */
35 SPARC_ST, /* Store instructions */
36 SPARC_SINGLE, /* Instructions that must issue by themselves */
38 SPARC_INV, /* This should stay at the end for the next value */
39 SPARC_NUM_SCHED_CLASSES = SPARC_INV
43 //---------------------------------------------------------------------------
44 // enum SparcMachineOpCode.
45 // const MachineInstrDescriptor SparcMachineInstrDesc[]
48 // Description of UltraSparc machine instructions.
50 //---------------------------------------------------------------------------
52 enum SparcMachineOpCode {
56 // Synthetic SPARC assembly opcodes for setting a register to a constant
60 // Set high-order bits of register and clear low-order bits
63 // Add or add with carry.
64 // Immed bit specifies if second operand is immediate(1) or register(0)
70 // Subtract or subtract with carry.
71 // Immed bit specifies if second operand is immediate(1) or register(0)
77 // Integer multiply, signed divide, unsigned divide.
78 // Note that the deprecated 32-bit multiply and multiply-step are not used.
83 // Floating point add, subtract, compare
93 // NOTE: FCMPE{S,D,Q}: FP Compare With Exception are currently unused!
95 // Floating point multiply or divide.
108 // Logical operations
130 // Floating point move, negate, and abs instructions
141 // Convert from floating point to floating point formats
149 // Convert from floating point to integer formats
157 // Convert from integer to floating point formats
165 // Branch on integer comparison with zero.
166 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
167 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
175 // Branch on integer condition code.
176 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
177 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
195 // Branch on floating point condition code.
196 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
197 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
215 // Conditional move on integer comparison with zero.
223 // Conditional move on integer condition code.
241 // Conditional move on floating point condition code.
242 // Note that the enum name is not the same as the assembly mnemonic below
243 // because that would duplicate some entries with those above.
244 // Therefore, we use MOVF here instead of MOV.
262 // Conditional move of floating point register on each of the above:
263 // i. on integer comparison with zero.
264 // ii. on integer condition code
265 // iii. on floating point condition code
266 // Note that the same set is repeated for S,D,Q register classes.
390 // Load integer instructions
399 // Load floating-point instructions
401 LDD, // use of this for integers is deprecated for Sparc V9
404 // Store integer instructions
410 // Store floating-point instructions
414 // Call, Return, and "Jump and link"
415 // Immed bit specifies if second operand is immediate(1) or register(0)
418 RETURN, // last valid opcode
420 // Synthetic phi operation for near-SSA form of machine code
423 // End-of-array marker
425 NUM_REAL_OPCODES = RETURN+1, // number of valid opcodes
426 NUM_TOTAL_OPCODES = INVALID_OPCODE
429 const MachineInstrDescriptor SparcMachineInstrDesc[] = {
431 // Fields of each structure:
434 // resultPosition (0-based; -1 if no result),
436 // immedIsSignExtended,
437 // numDelaySlots (in cycles)
438 // latency (in cycles)
439 // instr sched class (defined above)
440 // instr class flags (defined in TargretMachine.h)
442 { "NOP", 0, -1, 0, false, 0, 1, SPARC_NONE, M_NOP_FLAG },
444 // Synthetic SPARC assembly opcodes for setting a register to a constant.
445 // Max immediate constant should be ignored for both these instructions.
446 { "SETSW", 2, 1, 0, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
447 { "SETUW", 2, 1, 0, false, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG | M_ARITH_FLAG },
449 // Set high-order bits of register and clear low-order bits
450 { "SETHI", 2, 1, (1 << 22) - 1, false, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG | M_ARITH_FLAG },
452 // Add or add with carry.
453 { "ADD", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
454 { "ADDcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
455 { "ADDC", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
456 { "ADDCcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
458 // Sub tract or subtract with carry.
459 { "SUB", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
460 { "SUBcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
461 { "SUBC", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
462 { "SUBCcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
464 // Integer multiply, signed divide, unsigned divide.
465 // Note that the deprecated 32-bit multiply and multiply-step are not used.
466 { "MULX", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
467 { "SDIVX", 3, 2, (1 << 12) - 1, true, 0, 6, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
468 { "UDIVX", 3, 2, (1 << 12) - 1, true, 0, 6, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
470 // Floating point add, subtract, compare.
471 // Note that destination of FCMP* instructions is operand 0, not operand 2.
472 { "FADDS", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
473 { "FADDD", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
474 { "FADDQ", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
475 { "FSUBS", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
476 { "FSUBD", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
477 { "FSUBQ", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
478 { "FCMPS", 3, 0, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
479 { "FCMPD", 3, 0, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
480 { "FCMPQ", 3, 0, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
481 // NOTE: FCMPE{S,D,Q}: FP Compare With Exception are currently unused!
483 // Floating point multiply or divide.
484 { "FMULS", 3, 2, 0, false, 0, 3, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
485 { "FMULD", 3, 2, 0, false, 0, 3, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
486 { "FMULQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
487 { "FSMULD", 3, 2, 0, false, 0, 3, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
488 { "FDMULQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
489 { "FDIVS", 3, 2, 0, false, 0, 12, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
490 { "FDIVD", 3, 2, 0, false, 0, 22, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
491 { "FDIVQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
492 { "FSQRTS", 3, 2, 0, false, 0, 12, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
493 { "FSQRTD", 3, 2, 0, false, 0, 22, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
494 { "FSQRTQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
496 // Logical operations
497 { "AND", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
498 { "ANDcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
499 { "ANDN", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
500 { "ANDNcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
501 { "OR", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
502 { "ORcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
503 { "ORN", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
504 { "ORNcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
505 { "XOR", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
506 { "XORcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
507 { "XNOR", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
508 { "XNORcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
511 { "SLL", 3, 2, (1 << 5) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
512 { "SRL", 3, 2, (1 << 5) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
513 { "SRA", 3, 2, (1 << 5) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_ARITH_FLAG },
514 { "SLLX", 3, 2, (1 << 6) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
515 { "SRLX", 3, 2, (1 << 6) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
516 { "SRAX", 3, 2, (1 << 6) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_ARITH_FLAG },
518 // Floating point move, negate, and abs instructions
519 { "FMOVS", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
520 { "FMOVD", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
521 //{ "FMOVQ", 2, 1, 0, false, 0, ?, SPARC_FPA, M_FLOAT_FLAG },
522 { "FNEGS", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
523 { "FNEGD", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
524 //{ "FNEGQ", 2, 1, 0, false, 0, ?, SPARC_FPA, M_FLOAT_FLAG },
525 { "FABSS", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
526 { "FABSD", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
527 //{ "FABSQ", 2, 1, 0, false, 0, ?, SPARC_FPA, M_FLOAT_FLAG },
529 // Convert from floating point to floating point formats
530 { "FSTOD", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
531 { "FSTOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
532 { "FDTOS", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
533 { "FDTOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
534 { "FQTOS", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
535 { "FQTOD", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
537 // Convert from floating point to integer formats.
538 // Note that this accesses both integer and floating point registers.
539 { "FSTOX", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
540 { "FDTOX", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
541 { "FQTOX", 2, 1, 0, false, 0, 2, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
542 { "FSTOI", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
543 { "FDTOI", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
544 { "FQTOI", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
546 // Convert from integer to floating point formats
547 // Note that this accesses both integer and floating point registers.
548 { "FXTOS", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
549 { "FXTOD", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
550 { "FXTOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
551 { "FITOS", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
552 { "FITOD", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
553 { "FITOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
555 // Branch on integer comparison with zero.
556 // Latency includes the delay slot.
557 { "BRZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
558 { "BRLEZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
559 { "BRLZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
560 { "BRNZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
561 { "BRGZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
562 { "BRGEZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
564 // Branch on condition code.
565 // The first argument specifies the ICC register: %icc or %xcc
566 // Latency includes the delay slot.
567 { "BA", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
568 { "BN", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
569 { "BNE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
570 { "BE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
571 { "BG", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
572 { "BLE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
573 { "BGE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
574 { "BL", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
575 { "BGU", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
576 { "BLEU", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
577 { "BCC", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
578 { "BCS", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
579 { "BPOS", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
580 { "BNEG", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
581 { "BVC", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
582 { "BVS", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
584 // Branch on floating point condition code.
585 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
586 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
587 // The first argument is the FCCn register (0 <= n <= 3).
588 // Latency includes the delay slot.
589 { "FBA", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
590 { "FBN", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
591 { "FBU", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
592 { "FBG", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
593 { "FBUG", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
594 { "FBL", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
595 { "FBUL", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
596 { "FBLG", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
597 { "FBNE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
598 { "FBE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
599 { "FBUE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
600 { "FBGE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
601 { "FBUGE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
602 { "FBLE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
603 { "FBULE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
604 { "FBO", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
606 // Conditional move on integer comparison with zero.
607 { "MOVRZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
608 { "MOVRLEZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
609 { "MOVRLZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
610 { "MOVRNZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
611 { "MOVRGZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
612 { "MOVRGEZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
614 // Conditional move on integer condition code.
615 // The first argument specifies the ICC register: %icc or %xcc
616 { "MOVA", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
617 { "MOVN", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
618 { "MOVNE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
619 { "MOVE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
620 { "MOVG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
621 { "MOVLE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
622 { "MOVGE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
623 { "MOVL", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
624 { "MOVGU", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
625 { "MOVLEU", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
626 { "MOVCC", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
627 { "MOVCS", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
628 { "MOVPOS", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
629 { "MOVNEG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
630 { "MOVVC", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
631 { "MOVVS", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
633 // Conditional move (of integer register) on floating point condition code.
634 // The first argument is the FCCn register (0 <= n <= 3).
635 // Note that the enum name above is not the same as the assembly mnemonic
636 // because some of the assembly mnemonics are the same as the move on
637 // integer CC (e.g., MOVG), and we cannot have the same enum entry twice.
638 { "MOVA", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
639 { "MOVN", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
640 { "MOVU", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
641 { "MOVG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
642 { "MOVUG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
643 { "MOVL", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
644 { "MOVUL", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
645 { "MOVLG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
646 { "MOVNE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
647 { "MOVE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
648 { "MOVUE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
649 { "MOVGE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
650 { "MOVUGE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
651 { "MOVLE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
652 { "MOVULE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
653 { "MOVO", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
655 // Conditional move of floating point register on each of the above:
656 // i. on integer comparison with zero.
657 // ii. on integer condition code
658 // iii. on floating point condition code
659 // Note that the same set is repeated for S,D,Q register classes.
660 { "FMOVRSZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
661 { "FMOVRSLEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
662 { "FMOVRSLZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
663 { "FMOVRSNZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
664 { "FMOVRSGZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
665 { "FMOVRSGEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
667 { "FMOVSA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
668 { "FMOVSN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
669 { "FMOVSNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
670 { "FMOVSE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
671 { "FMOVSG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
672 { "FMOVSLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
673 { "FMOVSGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
674 { "FMOVSL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
675 { "FMOVSGU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
676 { "FMOVSLEU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
677 { "FMOVSCC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
678 { "FMOVSCS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
679 { "FMOVSPOS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
680 { "FMOVSNEG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
681 { "FMOVSVC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
682 { "FMOVSVS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
684 { "FMOVSA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
685 { "FMOVSN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
686 { "FMOVSU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
687 { "FMOVSG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
688 { "FMOVSUG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
689 { "FMOVSL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
690 { "FMOVSUL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
691 { "FMOVSLG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
692 { "FMOVSNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
693 { "FMOVSE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
694 { "FMOVSUE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
695 { "FMOVSGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
696 { "FMOVSUGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
697 { "FMOVSLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
698 { "FMOVSULE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
699 { "FMOVSO", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
701 { "FMOVRDZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
702 { "FMOVRDLEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
703 { "FMOVRDLZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
704 { "FMOVRDNZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
705 { "FMOVRDGZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
706 { "FMOVRDGEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
708 { "FMOVDA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
709 { "FMOVDN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
710 { "FMOVDNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
711 { "FMOVDE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
712 { "FMOVDG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
713 { "FMOVDLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
714 { "FMOVDGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
715 { "FMOVDL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
716 { "FMOVDGU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
717 { "FMOVDLEU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
718 { "FMOVDCC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
719 { "FMOVDCS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
720 { "FMOVDPOS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
721 { "FMOVDNEG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
722 { "FMOVDVC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
723 { "FMOVDVS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
725 { "FMOVDA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
726 { "FMOVDN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
727 { "FMOVDU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
728 { "FMOVDG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
729 { "FMOVDUG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
730 { "FMOVDL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
731 { "FMOVDUL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
732 { "FMOVDLG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
733 { "FMOVDNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
734 { "FMOVDE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
735 { "FMOVDUE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
736 { "FMOVDGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
737 { "FMOVDUGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
738 { "FMOVDLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
739 { "FMOVDULE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
740 { "FMOVDO", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
742 { "FMOVRQZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
743 { "FMOVRQLEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
744 { "FMOVRQLZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
745 { "FMOVRQNZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
746 { "FMOVRQGZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
747 { "FMOVRQGEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
749 { "FMOVQA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
750 { "FMOVQN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
751 { "FMOVQNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
752 { "FMOVQE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
753 { "FMOVQG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
754 { "FMOVQLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
755 { "FMOVQGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
756 { "FMOVQL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
757 { "FMOVQGU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
758 { "FMOVQLEU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
759 { "FMOVQCC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
760 { "FMOVQCS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
761 { "FMOVQPOS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
762 { "FMOVQNEG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
763 { "FMOVQVC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
764 { "FMOVQVS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
766 { "FMOVQA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
767 { "FMOVQN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
768 { "FMOVQU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
769 { "FMOVQG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
770 { "FMOVQUG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
771 { "FMOVQL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
772 { "FMOVQUL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
773 { "FMOVQLG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
774 { "FMOVQNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
775 { "FMOVQE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
776 { "FMOVQUE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
777 { "FMOVQGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
778 { "FMOVQUGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
779 { "FMOVQLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
780 { "FMOVQULE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
781 { "FMOVQO", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
783 // Load integer instructions
784 // Latency includes 1 cycle for address generation (Sparc IIi)
785 // Signed loads of less than 64 bits need an extra cycle for sign-extension.
787 // Not reflected here: After a 3-cycle loads, all subsequent consecutive
788 // loads also require 3 cycles to avoid contention for the load return
789 // stage. Latency returns to 2 cycles after the first cycle with no load.
790 { "LDSB", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
791 { "LDSH", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
792 { "LDSW", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
793 { "LDUB", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
794 { "LDUH", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
795 { "LDUW", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
796 { "LDX", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
798 // Load floating-point instructions
799 // Latency includes 1 cycle for address generation (Sparc IIi)
800 { "LD", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_FLOAT_FLAG | M_LOAD_FLAG },
801 { "LDD", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_FLOAT_FLAG | M_LOAD_FLAG },
802 { "LDQ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_FLOAT_FLAG | M_LOAD_FLAG },
804 // Store integer instructions
805 // Latency includes 1 cycle for address generation (Sparc IIi)
806 { "STB", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
807 { "STH", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
808 { "STW", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
809 { "STX", 3, -1, (1 << 12) - 1, true, 0, 3, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
811 // Store floating-point instructions (Sparc IIi)
812 { "ST", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_FLOAT_FLAG | M_STORE_FLAG},
813 { "STD", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_FLOAT_FLAG | M_STORE_FLAG},
815 // Call, Return and "Jump and link".
816 // Latency includes the delay slot.
817 { "CALL", 1, -1, (1 << 29) - 1, true, 1, 2, SPARC_CTI, M_BRANCH_FLAG | M_CALL_FLAG},
818 { "JMPL", 3, -1, (1 << 12) - 1, true, 1, 2, SPARC_CTI, M_BRANCH_FLAG | M_CALL_FLAG},
819 { "RETURN", 2, -1, 0, false, 1, 2, SPARC_CTI, M_BRANCH_FLAG | M_RET_FLAG },
821 // Synthetic phi operation for near-SSA form of machine code
822 // Number of operands is variable, indicated by -1. Result is the first op.
824 { "PHI", -1, 0, 0, false, 0, 0, SPARC_INV, M_DUMMY_PHI_FLAG },
830 //---------------------------------------------------------------------------
831 // class UltraSparcInstrInfo
834 // Information about individual instructions.
835 // Most information is stored in the SparcMachineInstrDesc array above.
836 // Other information is computed on demand, and most such functions
837 // default to member functions in base class MachineInstrInfo.
838 //---------------------------------------------------------------------------
840 class UltraSparcInstrInfo : public MachineInstrInfo {
842 /*ctor*/ UltraSparcInstrInfo();
844 virtual bool hasResultInterlock (MachineOpCode opCode)
846 // All UltraSPARC instructions have interlocks (note that delay slots
847 // are not considered here).
848 // However, instructions that use the result of an FCMP produce a
849 // 9-cycle stall if they are issued less than 3 cycles after the FCMP.
850 // Force the compiler to insert a software interlock (i.e., gap of
851 // 2 other groups, including NOPs if necessary).
852 return (opCode == FCMPS || opCode == FCMPD || opCode == FCMPQ);
858 //---------------------------------------------------------------------------
859 // class UltraSparcRegInfo
862 // This class provides info about sparc register classes.
863 //---------------------------------------------------------------------------
869 class UltraSparcRegInfo : public MachineRegInfo
881 // WARNING: If the above enum order must be changed, also modify
882 // getRegisterClassOfValue method below since it assumes this particular
883 // order for efficiency.
886 // reverse pointer to get info about the ultra sparc machine
887 const UltraSparc *const UltraSparcInfo;
889 // Int arguments can be passed in 6 int regs - %o0 to %o5 (cannot be changed)
890 unsigned const NumOfIntArgRegs;
892 // Float arguments can be passed in this many regs - can be canged if needed
893 // %f0 - %f5 are used (can hold 6 floats or 3 doubles)
894 unsigned const NumOfFloatArgRegs;
896 void setCallArgColor(LiveRange *const LR, const unsigned RegNo) const;
902 UltraSparcRegInfo(const UltraSparc *const USI )
908 MachineRegClassArr.push_back( new SparcIntRegClass(IntRegClassID) );
909 MachineRegClassArr.push_back( new SparcFloatRegClass(FloatRegClassID) );
910 MachineRegClassArr.push_back( new SparcIntCCRegClass(IntCCRegClassID) );
911 MachineRegClassArr.push_back( new SparcFloatCCRegClass(FloatCCRegClassID));
913 assert( SparcFloatRegOrder::StartOfNonVolatileRegs == 6 &&
914 "6 Float regs are used for float arg passing");
918 ~UltraSparcRegInfo(void) { } // empty destructor
921 inline const UltraSparc & getUltraSparcInfo() const {
922 return *UltraSparcInfo;
925 // returns the register that is hardwired to zero
926 virtual inline int getZeroRegNum() const {
927 return (int) SparcIntRegOrder::g0;
931 inline unsigned getRegClassIDOfValue (const Value *const Val,
932 bool isCCReg = false) const {
934 Type::PrimitiveID ty = (Val->getType())->getPrimitiveID();
938 if( ty && ty <= Type::LongTyID || (ty == Type::PointerTyID) )
939 res = IntRegClassID; // sparc int reg (ty=0: void)
940 else if( ty <= Type::DoubleTyID)
941 res = FloatRegClassID; // sparc float reg class
943 cout << "TypeID: " << ty << endl;
944 assert(0 && "Cannot resolve register class for type");
949 return res + 2; // corresponidng condition code regiser
956 void colorArgs(const Method *const Meth, LiveRangeInfo& LRI) const;
958 static void printReg(const LiveRange *const LR) ;
960 void colorCallArgs(vector<const Instruction *> & CallInstrList,
962 AddedInstrMapType& AddedInstrMap ) const;
964 // this method provides a unique number for each register
965 inline int getUnifiedRegNum(int RegClassID, int reg) const {
967 if( RegClassID == IntRegClassID && reg < 32 )
969 else if ( RegClassID == FloatRegClassID && reg < 64)
970 return reg + 32; // we have 32 int regs
971 else if( RegClassID == FloatCCRegClassID && reg < 4)
972 return reg + 32 + 64; // 32 int, 64 float
973 else if( RegClassID == IntCCRegClassID )
974 return 4+ 32 + 64; // only int cc reg
976 assert(0 && "Invalid register class or reg number");
980 // given the unified register number, this gives the name
981 inline const string getUnifiedRegName(int reg) const {
984 return SparcIntRegOrder::getRegName(reg);
985 else if ( reg < (64 + 32) )
986 return SparcFloatRegOrder::getRegName( reg - 32);
987 else if( reg < (64+32+4) )
988 return SparcFloatCCRegOrder::getRegName( reg -32 - 64);
989 else if ( reg == 64+32+4)
990 return "xcc"; // only integer cc reg
992 assert(0 && "Invalid register number");
1001 /*---------------------------------------------------------------------------
1002 Scheduling guidelines for SPARC IIi:
1004 I-Cache alignment rules (pg 326)
1005 -- Align a branch target instruction so that it's entire group is within
1006 the same cache line (may be 1-4 instructions).
1007 ** Don't let a branch that is predicted taken be the last instruction
1008 on an I-cache line: delay slot will need an entire line to be fetched
1009 -- Make a FP instruction or a branch be the 4th instruction in a group.
1010 For branches, there are tradeoffs in reordering to make this happen
1012 ** Don't put a branch in a group that crosses a 32-byte boundary!
1013 An artificial branch is inserted after every 32 bytes, and having
1014 another branch will force the group to be broken into 2 groups.
1017 -- Don't let a loop span two memory pages, if possible
1019 Branch prediction performance:
1020 -- Don't make the branch in a delay slot the target of a branch
1021 -- Try not to have 2 predicted branches within a group of 4 instructions
1022 (because each such group has a single branch target field).
1023 -- Try to align branches in slots 0, 2, 4 or 6 of a cache line (to avoid
1024 the wrong prediction bits being used in some cases).
1026 D-Cache timing constraints:
1027 -- Signed int loads of less than 64 bits have 3 cycle latency, not 2
1028 -- All other loads that hit in D-Cache have 2 cycle latency
1029 -- All loads are returned IN ORDER, so a D-Cache miss will delay a later hit
1030 -- Mis-aligned loads or stores cause a trap. In particular, replace
1031 mis-aligned FP double precision l/s with 2 single-precision l/s.
1032 -- Simulations of integer codes show increase in avg. group size of
1033 33% when code (including esp. non-faulting loads) is moved across
1034 one branch, and 50% across 2 branches.
1036 E-Cache timing constraints:
1037 -- Scheduling for E-cache (D-Cache misses) is effective (due to load buffering)
1039 Store buffer timing constraints:
1040 -- Stores can be executed in same cycle as instruction producing the value
1041 -- Stores are buffered and have lower priority for E-cache until
1042 highwater mark is reached in the store buffer (5 stores)
1044 Pipeline constraints:
1045 -- Shifts can only use IEU0.
1046 -- CC setting instructions can only use IEU1.
1047 -- Several other instructions must only use IEU1:
1048 EDGE(?), ARRAY(?), CALL, JMPL, BPr, PST, and FCMP.
1049 -- Two instructions cannot store to the same register file in a single cycle
1050 (single write port per file).
1052 Issue and grouping constraints:
1053 -- FP and branch instructions must use slot 4.
1054 -- Shift instructions cannot be grouped with other IEU0-specific instructions.
1055 -- CC setting instructions cannot be grouped with other IEU1-specific instrs.
1056 -- Several instructions must be issued in a single-instruction group:
1057 MOVcc or MOVr, MULs/x and DIVs/x, SAVE/RESTORE, many others
1058 -- A CALL or JMPL breaks a group, ie, is not combined with subsequent instrs.
1062 Branch delay slot scheduling rules:
1063 -- A CTI couple (two back-to-back CTI instructions in the dynamic stream)
1064 has a 9-instruction penalty: the entire pipeline is flushed when the
1065 second instruction reaches stage 9 (W-Writeback).
1066 -- Avoid putting multicycle instructions, and instructions that may cause
1067 load misses, in the delay slot of an annulling branch.
1068 -- Avoid putting WR, SAVE..., RESTORE and RETURN instructions in the
1069 delay slot of an annulling branch.
1071 *--------------------------------------------------------------------------- */
1073 //---------------------------------------------------------------------------
1074 // List of CPUResources for UltraSPARC IIi.
1075 //---------------------------------------------------------------------------
1077 const CPUResource AllIssueSlots( "All Instr Slots", 4);
1078 const CPUResource IntIssueSlots( "Int Instr Slots", 3);
1079 const CPUResource First3IssueSlots("Instr Slots 0-3", 3);
1080 const CPUResource LSIssueSlots( "Load-Store Instr Slot", 1);
1081 const CPUResource CTIIssueSlots( "Ctrl Transfer Instr Slot", 1);
1082 const CPUResource FPAIssueSlots( "Int Instr Slot 1", 1);
1083 const CPUResource FPMIssueSlots( "Int Instr Slot 1", 1);
1085 // IEUN instructions can use either Alu and should use IAluN.
1086 // IEU0 instructions must use Alu 1 and should use both IAluN and IAlu0.
1087 // IEU1 instructions must use Alu 2 and should use both IAluN and IAlu1.
1088 const CPUResource IAluN("Int ALU 1or2", 2);
1089 const CPUResource IAlu0("Int ALU 1", 1);
1090 const CPUResource IAlu1("Int ALU 2", 1);
1092 const CPUResource LSAluC1("Load/Store Unit Addr Cycle", 1);
1093 const CPUResource LSAluC2("Load/Store Unit Issue Cycle", 1);
1094 const CPUResource LdReturn("Load Return Unit", 1);
1096 const CPUResource FPMAluC1("FP Mul/Div Alu Cycle 1", 1);
1097 const CPUResource FPMAluC2("FP Mul/Div Alu Cycle 2", 1);
1098 const CPUResource FPMAluC3("FP Mul/Div Alu Cycle 3", 1);
1100 const CPUResource FPAAluC1("FP Other Alu Cycle 1", 1);
1101 const CPUResource FPAAluC2("FP Other Alu Cycle 2", 1);
1102 const CPUResource FPAAluC3("FP Other Alu Cycle 3", 1);
1104 const CPUResource IRegReadPorts("Int Reg ReadPorts", INT_MAX); // CHECK
1105 const CPUResource IRegWritePorts("Int Reg WritePorts", 2); // CHECK
1106 const CPUResource FPRegReadPorts("FP Reg Read Ports", INT_MAX); // CHECK
1107 const CPUResource FPRegWritePorts("FP Reg Write Ports", 1); // CHECK
1109 const CPUResource CTIDelayCycle( "CTI delay cycle", 1);
1110 const CPUResource FCMPDelayCycle("FCMP delay cycle", 1);
1113 //---------------------------------------------------------------------------
1114 // const InstrClassRUsage SparcRUsageDesc[]
1117 // Resource usage information for instruction in each scheduling class.
1118 // The InstrRUsage Objects for individual classes are specified first.
1119 // Note that fetch and decode are decoupled from the execution pipelines
1120 // via an instr buffer, so they are not included in the cycles below.
1121 //---------------------------------------------------------------------------
1123 const InstrClassRUsage NoneClassRUsage = {
1127 /* maxIssueNum */ 4,
1128 /* isSingleIssue */ false,
1129 /* breaksGroup */ false,
1133 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1147 const InstrClassRUsage IEUNClassRUsage = {
1151 /* maxIssueNum */ 3,
1152 /* isSingleIssue */ false,
1153 /* breaksGroup */ false,
1157 /* feasibleSlots[] */ { 0, 1, 2 },
1161 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1162 { IntIssueSlots.rid, 0, 1 },
1163 /*Cycle E */ { IAluN.rid, 1, 1 },
1168 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1172 const InstrClassRUsage IEU0ClassRUsage = {
1176 /* maxIssueNum */ 1,
1177 /* isSingleIssue */ false,
1178 /* breaksGroup */ false,
1182 /* feasibleSlots[] */ { 0, 1, 2 },
1186 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1187 { IntIssueSlots.rid, 0, 1 },
1188 /*Cycle E */ { IAluN.rid, 1, 1 },
1189 { IAlu0.rid, 1, 1 },
1194 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1198 const InstrClassRUsage IEU1ClassRUsage = {
1202 /* maxIssueNum */ 1,
1203 /* isSingleIssue */ false,
1204 /* breaksGroup */ false,
1208 /* feasibleSlots[] */ { 0, 1, 2 },
1212 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1213 { IntIssueSlots.rid, 0, 1 },
1214 /*Cycle E */ { IAluN.rid, 1, 1 },
1215 { IAlu1.rid, 1, 1 },
1220 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1224 const InstrClassRUsage FPMClassRUsage = {
1228 /* maxIssueNum */ 1,
1229 /* isSingleIssue */ false,
1230 /* breaksGroup */ false,
1234 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1238 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1239 { FPMIssueSlots.rid, 0, 1 },
1240 /*Cycle E */ { FPRegReadPorts.rid, 1, 1 },
1241 /*Cycle C */ { FPMAluC1.rid, 2, 1 },
1242 /*Cycle N1*/ { FPMAluC2.rid, 3, 1 },
1243 /*Cycle N1*/ { FPMAluC3.rid, 4, 1 },
1245 /*Cycle W */ { FPRegWritePorts.rid, 6, 1 }
1249 const InstrClassRUsage FPAClassRUsage = {
1253 /* maxIssueNum */ 1,
1254 /* isSingleIssue */ false,
1255 /* breaksGroup */ false,
1259 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1263 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1264 { FPAIssueSlots.rid, 0, 1 },
1265 /*Cycle E */ { FPRegReadPorts.rid, 1, 1 },
1266 /*Cycle C */ { FPAAluC1.rid, 2, 1 },
1267 /*Cycle N1*/ { FPAAluC2.rid, 3, 1 },
1268 /*Cycle N1*/ { FPAAluC3.rid, 4, 1 },
1270 /*Cycle W */ { FPRegWritePorts.rid, 6, 1 }
1274 const InstrClassRUsage LDClassRUsage = {
1278 /* maxIssueNum */ 1,
1279 /* isSingleIssue */ false,
1280 /* breaksGroup */ false,
1284 /* feasibleSlots[] */ { 0, 1, 2, },
1288 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1289 { First3IssueSlots.rid, 0, 1 },
1290 { LSIssueSlots.rid, 0, 1 },
1291 /*Cycle E */ { LSAluC1.rid, 1, 1 },
1292 /*Cycle C */ { LSAluC2.rid, 2, 1 },
1293 { LdReturn.rid, 2, 1 },
1297 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1301 const InstrClassRUsage STClassRUsage = {
1305 /* maxIssueNum */ 1,
1306 /* isSingleIssue */ false,
1307 /* breaksGroup */ false,
1311 /* feasibleSlots[] */ { 0, 1, 2 },
1315 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1316 { First3IssueSlots.rid, 0, 1 },
1317 { LSIssueSlots.rid, 0, 1 },
1318 /*Cycle E */ { LSAluC1.rid, 1, 1 },
1319 /*Cycle C */ { LSAluC2.rid, 2, 1 }
1327 const InstrClassRUsage CTIClassRUsage = {
1331 /* maxIssueNum */ 1,
1332 /* isSingleIssue */ false,
1333 /* breaksGroup */ false,
1337 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1341 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1342 { CTIIssueSlots.rid, 0, 1 },
1343 /*Cycle E */ { IAlu0.rid, 1, 1 },
1344 /*Cycles E-C */ { CTIDelayCycle.rid, 1, 2 }
1353 const InstrClassRUsage SingleClassRUsage = {
1357 /* maxIssueNum */ 1,
1358 /* isSingleIssue */ true,
1359 /* breaksGroup */ false,
1363 /* feasibleSlots[] */ { 0 },
1367 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1368 { AllIssueSlots.rid, 0, 1 },
1369 { AllIssueSlots.rid, 0, 1 },
1370 { AllIssueSlots.rid, 0, 1 },
1371 /*Cycle E */ { IAlu0.rid, 1, 1 }
1381 const InstrClassRUsage SparcRUsageDesc[] = {
1395 //---------------------------------------------------------------------------
1396 // const InstrIssueDelta SparcInstrIssueDeltas[]
1399 // Changes to issue restrictions information in InstrClassRUsage for
1400 // instructions that differ from other instructions in their class.
1401 //---------------------------------------------------------------------------
1403 const InstrIssueDelta SparcInstrIssueDeltas[] = {
1405 // opCode, isSingleIssue, breaksGroup, numBubbles
1407 // Special cases for single-issue only
1408 // Other single issue cases are below.
1409 //{ LDDA, true, true, 0 },
1410 //{ STDA, true, true, 0 },
1411 //{ LDDF, true, true, 0 },
1412 //{ LDDFA, true, true, 0 },
1413 { ADDC, true, true, 0 },
1414 { ADDCcc, true, true, 0 },
1415 { SUBC, true, true, 0 },
1416 { SUBCcc, true, true, 0 },
1417 //{ SAVE, true, true, 0 },
1418 //{ RESTORE, true, true, 0 },
1419 //{ LDSTUB, true, true, 0 },
1420 //{ SWAP, true, true, 0 },
1421 //{ SWAPA, true, true, 0 },
1422 //{ CAS, true, true, 0 },
1423 //{ CASA, true, true, 0 },
1424 //{ CASX, true, true, 0 },
1425 //{ CASXA, true, true, 0 },
1426 //{ LDFSR, true, true, 0 },
1427 //{ LDFSRA, true, true, 0 },
1428 //{ LDXFSR, true, true, 0 },
1429 //{ LDXFSRA, true, true, 0 },
1430 //{ STFSR, true, true, 0 },
1431 //{ STFSRA, true, true, 0 },
1432 //{ STXFSR, true, true, 0 },
1433 //{ STXFSRA, true, true, 0 },
1434 //{ SAVED, true, true, 0 },
1435 //{ RESTORED, true, true, 0 },
1436 //{ FLUSH, true, true, 9 },
1437 //{ FLUSHW, true, true, 9 },
1438 //{ ALIGNADDR, true, true, 0 },
1439 { RETURN, true, true, 0 },
1440 //{ DONE, true, true, 0 },
1441 //{ RETRY, true, true, 0 },
1442 //{ WR, true, true, 0 },
1443 //{ WRPR, true, true, 4 },
1444 //{ RD, true, true, 0 },
1445 //{ RDPR, true, true, 0 },
1446 //{ TCC, true, true, 0 },
1447 //{ SHUTDOWN, true, true, 0 },
1449 // Special cases for breaking group *before*
1450 // CURRENTLY NOT SUPPORTED!
1451 { CALL, false, false, 0 },
1452 { JMPL, false, false, 0 },
1454 // Special cases for breaking the group *after*
1455 { MULX, true, true, (4+34)/2 },
1456 { FDIVS, false, true, 0 },
1457 { FDIVD, false, true, 0 },
1458 { FDIVQ, false, true, 0 },
1459 { FSQRTS, false, true, 0 },
1460 { FSQRTD, false, true, 0 },
1461 { FSQRTQ, false, true, 0 },
1462 //{ FCMP{LE,GT,NE,EQ}, false, true, 0 },
1464 // Instructions that introduce bubbles
1465 //{ MULScc, true, true, 2 },
1466 //{ SMULcc, true, true, (4+18)/2 },
1467 //{ UMULcc, true, true, (4+19)/2 },
1468 { SDIVX, true, true, 68 },
1469 { UDIVX, true, true, 68 },
1470 //{ SDIVcc, true, true, 36 },
1471 //{ UDIVcc, true, true, 37 },
1472 //{ WR, false, false, 4 },
1473 //{ WRPR, false, false, 4 },
1477 //---------------------------------------------------------------------------
1478 // const InstrRUsageDelta SparcInstrUsageDeltas[]
1481 // Changes to resource usage information in InstrClassRUsage for
1482 // instructions that differ from other instructions in their class.
1483 //---------------------------------------------------------------------------
1485 const InstrRUsageDelta SparcInstrUsageDeltas[] = {
1487 // MachineOpCode, Resource, Start cycle, Num cycles
1490 // JMPL counts as a load/store instruction for issue!
1492 { JMPL, LSIssueSlots.rid, 0, 1 },
1495 // Many instructions cannot issue for the next 2 cycles after an FCMP
1496 // We model that with a fake resource FCMPDelayCycle.
1498 { FCMPS, FCMPDelayCycle.rid, 1, 3 },
1499 { FCMPD, FCMPDelayCycle.rid, 1, 3 },
1500 { FCMPQ, FCMPDelayCycle.rid, 1, 3 },
1502 { MULX, FCMPDelayCycle.rid, 1, 1 },
1503 { SDIVX, FCMPDelayCycle.rid, 1, 1 },
1504 { UDIVX, FCMPDelayCycle.rid, 1, 1 },
1505 //{ SMULcc, FCMPDelayCycle.rid, 1, 1 },
1506 //{ UMULcc, FCMPDelayCycle.rid, 1, 1 },
1507 //{ SDIVcc, FCMPDelayCycle.rid, 1, 1 },
1508 //{ UDIVcc, FCMPDelayCycle.rid, 1, 1 },
1509 { STD, FCMPDelayCycle.rid, 1, 1 },
1510 { FMOVRSZ, FCMPDelayCycle.rid, 1, 1 },
1511 { FMOVRSLEZ,FCMPDelayCycle.rid, 1, 1 },
1512 { FMOVRSLZ, FCMPDelayCycle.rid, 1, 1 },
1513 { FMOVRSNZ, FCMPDelayCycle.rid, 1, 1 },
1514 { FMOVRSGZ, FCMPDelayCycle.rid, 1, 1 },
1515 { FMOVRSGEZ,FCMPDelayCycle.rid, 1, 1 },
1518 // Some instructions are stalled in the GROUP stage if a CTI is in
1521 { LDD, CTIDelayCycle.rid, 1, 1 },
1522 //{ LDDA, CTIDelayCycle.rid, 1, 1 },
1523 //{ LDDSTUB, CTIDelayCycle.rid, 1, 1 },
1524 //{ LDDSTUBA, CTIDelayCycle.rid, 1, 1 },
1525 //{ SWAP, CTIDelayCycle.rid, 1, 1 },
1526 //{ SWAPA, CTIDelayCycle.rid, 1, 1 },
1527 //{ CAS, CTIDelayCycle.rid, 1, 1 },
1528 //{ CASA, CTIDelayCycle.rid, 1, 1 },
1529 //{ CASX, CTIDelayCycle.rid, 1, 1 },
1530 //{ CASXA, CTIDelayCycle.rid, 1, 1 },
1533 // Signed int loads of less than dword size return data in cycle N1 (not C)
1534 // and put all loads in consecutive cycles into delayed load return mode.
1536 { LDSB, LdReturn.rid, 2, -1 },
1537 { LDSB, LdReturn.rid, 3, 1 },
1539 { LDSH, LdReturn.rid, 2, -1 },
1540 { LDSH, LdReturn.rid, 3, 1 },
1542 { LDSW, LdReturn.rid, 2, -1 },
1543 { LDSW, LdReturn.rid, 3, 1 },
1546 #undef EXPLICIT_BUBBLES_NEEDED
1547 #ifdef EXPLICIT_BUBBLES_NEEDED
1549 // MULScc inserts one bubble.
1550 // This means it breaks the current group (captured in UltraSparcSchedInfo)
1551 // *and occupies all issue slots for the next cycle
1553 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1554 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1555 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1556 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1559 // SMULcc inserts between 4 and 18 bubbles, depending on #leading 0s in rs1.
1560 // We just model this with a simple average.
1562 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1563 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1564 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1565 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1567 // SMULcc inserts between 4 and 19 bubbles, depending on #leading 0s in rs1.
1568 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1569 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1570 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1571 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1574 // MULX inserts between 4 and 34 bubbles, depending on #leading 0s in rs1.
1576 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1577 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1578 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1579 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1582 // SDIVcc inserts 36 bubbles.
1584 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1585 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1586 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1587 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1589 // UDIVcc inserts 37 bubbles.
1590 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1591 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1592 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1593 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1596 // SDIVX inserts 68 bubbles.
1598 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1599 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1600 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1601 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1604 // UDIVX inserts 68 bubbles.
1606 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1607 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1608 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1609 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1612 // WR inserts 4 bubbles.
1614 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1615 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1616 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1617 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1620 // WRPR inserts 4 bubbles.
1622 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1623 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1624 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1625 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1628 // DONE inserts 9 bubbles.
1630 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1631 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1632 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1633 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1636 // RETRY inserts 9 bubbles.
1638 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1639 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1640 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1641 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1643 #endif EXPLICIT_BUBBLES_NEEDED
1648 // Additional delays to be captured in code:
1649 // 1. RDPR from several state registers (page 349)
1650 // 2. RD from *any* register (page 349)
1651 // 3. Writes to TICK, PSTATE, TL registers and FLUSH{W} instr (page 349)
1652 // 4. Integer store can be in same group as instr producing value to store.
1653 // 5. BICC and BPICC can be in the same group as instr producing CC (pg 350)
1654 // 6. FMOVr cannot be in the same or next group as an IEU instr (pg 351).
1655 // 7. The second instr. of a CTI group inserts 9 bubbles (pg 351)
1656 // 8. WR{PR}, SVAE, SAVED, RESTORE, RESTORED, RETURN, RETRY, and DONE that
1657 // follow an annulling branch cannot be issued in the same group or in
1658 // the 3 groups following the branch.
1659 // 9. A predicted annulled load does not stall dependent instructions.
1660 // Other annulled delay slot instructions *do* stall dependents, so
1661 // nothing special needs to be done for them during scheduling.
1662 //10. Do not put a load use that may be annulled in the same group as the
1663 // branch. The group will stall until the load returns.
1664 //11. Single-prec. FP loads lock 2 registers, for dependency checking.
1667 // Additional delays we cannot or will not capture:
1668 // 1. If DCTI is last word of cache line, it is delayed until next line can be
1669 // fetched. Also, other DCTI alignment-related delays (pg 352)
1670 // 2. Load-after-store is delayed by 7 extra cycles if load hits in D-Cache.
1671 // Also, several other store-load and load-store conflicts (pg 358)
1672 // 3. MEMBAR, LD{X}FSR, LDD{A} and a bunch of other load stalls (pg 358)
1673 // 4. There can be at most 8 outstanding buffered store instructions
1674 // (including some others like MEMBAR, LDSTUB, CAS{AX}, and FLUSH)
1678 //---------------------------------------------------------------------------
1679 // class UltraSparcSchedInfo
1682 // Interface to instruction scheduling information for UltraSPARC.
1683 // The parameter values above are based on UltraSPARC IIi.
1684 //---------------------------------------------------------------------------
1687 class UltraSparcSchedInfo: public MachineSchedInfo {
1689 /*ctor*/ UltraSparcSchedInfo (const MachineInstrInfo* mii);
1690 /*dtor*/ virtual ~UltraSparcSchedInfo () {}
1692 virtual void initializeResources ();
1696 //---------------------------------------------------------------------------
1697 // class UltraSparcMachine
1700 // Primary interface to machine description for the UltraSPARC.
1701 // Primarily just initializes machine-dependent parameters in
1702 // class TargetMachine, and creates machine-dependent subclasses
1703 // for classes such as InstrInfo, SchedInfo and RegInfo.
1704 //---------------------------------------------------------------------------
1706 class UltraSparc : public TargetMachine {
1708 UltraSparcInstrInfo instrInfo;
1709 UltraSparcSchedInfo schedInfo;
1710 UltraSparcRegInfo regInfo;
1713 virtual ~UltraSparc() {}
1715 virtual const MachineInstrInfo& getInstrInfo() const { return instrInfo; }
1717 virtual const MachineSchedInfo& getSchedInfo() const { return schedInfo; }
1719 virtual const MachineRegInfo& getRegInfo() const { return regInfo; }
1721 // compileMethod - For the sparc, we do instruction selection, followed by
1722 // delay slot scheduling, then register allocation.
1724 virtual bool compileMethod(Method *M);