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
12 #include "SparcRegClassInfo.h"
13 #include "llvm/Target/TargetMachine.h"
14 #include "llvm/Target/MachineInstrInfo.h"
16 #include "llvm/Target/MachineSchedInfo.h"
17 #include "llvm/Type.h"
19 #include <sys/types.h>
23 // OpCodeMask definitions for the Sparc V9
25 const OpCodeMask Immed = 0x00002000; // immed or reg operand?
26 const OpCodeMask Annul = 0x20000000; // annul delay instr?
27 const OpCodeMask PredictTaken = 0x00080000; // predict branch taken?
30 enum SparcInstrSchedClass {
31 SPARC_NONE, /* Instructions with no scheduling restrictions */
32 SPARC_IEUN, /* Integer class that can use IEU0 or IEU1 */
33 SPARC_IEU0, /* Integer class IEU0 */
34 SPARC_IEU1, /* Integer class IEU1 */
35 SPARC_FPM, /* FP Multiply or Divide instructions */
36 SPARC_FPA, /* All other FP instructions */
37 SPARC_CTI, /* Control-transfer instructions */
38 SPARC_LD, /* Load instructions */
39 SPARC_ST, /* Store instructions */
40 SPARC_SINGLE, /* Instructions that must issue by themselves */
42 SPARC_INV, /* This should stay at the end for the next value */
43 SPARC_NUM_SCHED_CLASSES = SPARC_INV
47 //---------------------------------------------------------------------------
48 // enum SparcMachineOpCode.
49 // const MachineInstrDescriptor SparcMachineInstrDesc[]
52 // Description of UltraSparc machine instructions.
54 //---------------------------------------------------------------------------
56 enum SparcMachineOpCode {
60 // Synthetic SPARC assembly opcodes for setting a register to a constant
64 // Set high-order bits of register and clear low-order bits
67 // Add or add with carry.
68 // Immed bit specifies if second operand is immediate(1) or register(0)
74 // Subtract or subtract with carry.
75 // Immed bit specifies if second operand is immediate(1) or register(0)
81 // Integer multiply, signed divide, unsigned divide.
82 // Note that the deprecated 32-bit multiply and multiply-step are not used.
87 // Floating point add, subtract, compare
97 // NOTE: FCMPE{S,D,Q}: FP Compare With Exception are currently unused!
99 // Floating point multiply or divide.
112 // Logical operations
134 // Floating point move, negate, and abs instructions
145 // Convert from floating point to floating point formats
153 // Convert from floating point to integer formats
161 // Convert from integer to floating point formats
169 // Branch on integer comparison with zero.
170 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
171 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
179 // Branch on integer condition code.
180 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
181 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
199 // Branch on floating point condition code.
200 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
201 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
219 // Conditional move on integer comparison with zero.
227 // Conditional move on integer condition code.
245 // Conditional move on floating point condition code.
246 // Note that the enum name is not the same as the assembly mnemonic below
247 // because that would duplicate some entries with those above.
248 // Therefore, we use MOVF here instead of MOV.
266 // Conditional move of floating point register on each of the above:
267 // i. on integer comparison with zero.
268 // ii. on integer condition code
269 // iii. on floating point condition code
270 // Note that the same set is repeated for S,D,Q register classes.
394 // Load integer instructions
403 // Load floating-point instructions
405 LDD, // use of this for integers is deprecated for Sparc V9
408 // Store integer instructions
414 // Store floating-point instructions
418 // Call, Return, and "Jump and link"
419 // Immed bit specifies if second operand is immediate(1) or register(0)
422 RETURN, // last valid opcode
424 // Synthetic phi operation for near-SSA form of machine code
427 // End-of-array marker
429 NUM_REAL_OPCODES = RETURN+1, // number of valid opcodes
430 NUM_TOTAL_OPCODES = INVALID_OPCODE
433 const MachineInstrDescriptor SparcMachineInstrDesc[] = {
435 // Fields of each structure:
438 // resultPosition (0-based; -1 if no result),
440 // immedIsSignExtended,
441 // numDelaySlots (in cycles)
442 // latency (in cycles)
443 // instr sched class (defined above)
444 // instr class flags (defined in MachineInstrInfo.h)
446 { "NOP", 0, -1, 0, false, 0, 1, SPARC_NONE, M_NOP_FLAG },
448 // Synthetic SPARC assembly opcodes for setting a register to a constant.
449 // Max immediate constant should be ignored for both these instructions.
450 { "SETSW", 2, 1, 0, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
451 { "SETUW", 2, 1, 0, false, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG | M_ARITH_FLAG },
453 // Set high-order bits of register and clear low-order bits
454 { "SETHI", 2, 1, (1 << 22) - 1, false, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG | M_ARITH_FLAG },
456 // Add or add with carry.
457 { "ADD", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
458 { "ADDcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
459 { "ADDC", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
460 { "ADDCcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
462 // Sub tract or subtract with carry.
463 { "SUB", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
464 { "SUBcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
465 { "SUBC", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
466 { "SUBCcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_ARITH_FLAG },
468 // Integer multiply, signed divide, unsigned divide.
469 // Note that the deprecated 32-bit multiply and multiply-step are not used.
470 { "MULX", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
471 { "SDIVX", 3, 2, (1 << 12) - 1, true, 0, 6, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
472 { "UDIVX", 3, 2, (1 << 12) - 1, true, 0, 6, SPARC_IEUN, M_INT_FLAG | M_ARITH_FLAG },
474 // Floating point add, subtract, compare.
475 // Note that destination of FCMP* instructions is operand 0, not operand 2.
476 { "FADDS", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
477 { "FADDD", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
478 { "FADDQ", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
479 { "FSUBS", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
480 { "FSUBD", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
481 { "FSUBQ", 3, 2, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
482 { "FCMPS", 3, 0, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
483 { "FCMPD", 3, 0, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
484 { "FCMPQ", 3, 0, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
485 // NOTE: FCMPE{S,D,Q}: FP Compare With Exception are currently unused!
487 // Floating point multiply or divide.
488 { "FMULS", 3, 2, 0, false, 0, 3, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
489 { "FMULD", 3, 2, 0, false, 0, 3, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
490 { "FMULQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
491 { "FSMULD", 3, 2, 0, false, 0, 3, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
492 { "FDMULQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
493 { "FDIVS", 3, 2, 0, false, 0, 12, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
494 { "FDIVD", 3, 2, 0, false, 0, 22, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
495 { "FDIVQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
496 { "FSQRTS", 3, 2, 0, false, 0, 12, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
497 { "FSQRTD", 3, 2, 0, false, 0, 22, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
498 { "FSQRTQ", 3, 2, 0, false, 0, 0, SPARC_FPM, M_FLOAT_FLAG | M_ARITH_FLAG },
500 // Logical operations
501 { "AND", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
502 { "ANDcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
503 { "ANDN", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
504 { "ANDNcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
505 { "OR", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
506 { "ORcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
507 { "ORN", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
508 { "ORNcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
509 { "XOR", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
510 { "XORcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
511 { "XNOR", 3, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEUN, M_INT_FLAG | M_LOGICAL_FLAG},
512 { "XNORcc", 4, 2, (1 << 12) - 1, true, 0, 1, SPARC_IEU1, M_INT_FLAG | M_LOGICAL_FLAG},
515 { "SLL", 3, 2, (1 << 5) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
516 { "SRL", 3, 2, (1 << 5) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
517 { "SRA", 3, 2, (1 << 5) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_ARITH_FLAG },
518 { "SLLX", 3, 2, (1 << 6) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
519 { "SRLX", 3, 2, (1 << 6) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_LOGICAL_FLAG},
520 { "SRAX", 3, 2, (1 << 6) - 1, true, 0, 1, SPARC_IEU0, M_INT_FLAG | M_ARITH_FLAG },
522 // Floating point move, negate, and abs instructions
523 { "FMOVS", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
524 { "FMOVD", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
525 //{ "FMOVQ", 2, 1, 0, false, 0, ?, SPARC_FPA, M_FLOAT_FLAG },
526 { "FNEGS", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
527 { "FNEGD", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
528 //{ "FNEGQ", 2, 1, 0, false, 0, ?, SPARC_FPA, M_FLOAT_FLAG },
529 { "FABSS", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
530 { "FABSD", 2, 1, 0, false, 0, 1, SPARC_FPA, M_FLOAT_FLAG },
531 //{ "FABSQ", 2, 1, 0, false, 0, ?, SPARC_FPA, M_FLOAT_FLAG },
533 // Convert from floating point to floating point formats
534 { "FSTOD", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
535 { "FSTOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
536 { "FDTOS", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
537 { "FDTOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
538 { "FQTOS", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
539 { "FQTOD", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_ARITH_FLAG },
541 // Convert from floating point to integer formats.
542 // Note that this accesses both integer and floating point registers.
543 { "FSTOX", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
544 { "FDTOX", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
545 { "FQTOX", 2, 1, 0, false, 0, 2, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
546 { "FSTOI", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
547 { "FDTOI", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
548 { "FQTOI", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
550 // Convert from integer to floating point formats
551 // Note that this accesses both integer and floating point registers.
552 { "FXTOS", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
553 { "FXTOD", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
554 { "FXTOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
555 { "FITOS", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
556 { "FITOD", 2, 1, 0, false, 0, 3, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
557 { "FITOQ", 2, 1, 0, false, 0, 0, SPARC_FPA, M_FLOAT_FLAG | M_INT_FLAG | M_ARITH_FLAG },
559 // Branch on integer comparison with zero.
560 // Latency includes the delay slot.
561 { "BRZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
562 { "BRLEZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
563 { "BRLZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
564 { "BRNZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
565 { "BRGZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
566 { "BRGEZ", 2, -1, (1 << 15) - 1, true, 1, 2, SPARC_CTI, M_INT_FLAG | M_BRANCH_FLAG },
568 // Branch on condition code.
569 // The first argument specifies the ICC register: %icc or %xcc
570 // Latency includes the delay slot.
571 { "BA", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
572 { "BN", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
573 { "BNE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
574 { "BE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
575 { "BG", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
576 { "BLE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
577 { "BGE", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
578 { "BL", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
579 { "BGU", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
580 { "BLEU", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
581 { "BCC", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
582 { "BCS", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
583 { "BPOS", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
584 { "BNEG", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
585 { "BVC", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
586 { "BVS", 2, -1, (1 << 21) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
588 // Branch on floating point condition code.
589 // Annul bit specifies if intruction in delay slot is annulled(1) or not(0).
590 // PredictTaken bit hints if branch should be predicted taken(1) or not(0).
591 // The first argument is the FCCn register (0 <= n <= 3).
592 // Latency includes the delay slot.
593 { "FBA", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
594 { "FBN", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
595 { "FBU", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
596 { "FBG", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
597 { "FBUG", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
598 { "FBL", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
599 { "FBUL", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
600 { "FBLG", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
601 { "FBNE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
602 { "FBE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
603 { "FBUE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
604 { "FBGE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
605 { "FBUGE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
606 { "FBLE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
607 { "FBULE", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
608 { "FBO", 2, -1, (1 << 18) - 1, true, 1, 2, SPARC_CTI, M_CC_FLAG | M_BRANCH_FLAG },
610 // Conditional move on integer comparison with zero.
611 { "MOVRZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
612 { "MOVRLEZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
613 { "MOVRLZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
614 { "MOVRNZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
615 { "MOVRGZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
616 { "MOVRGEZ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_INT_FLAG },
618 // Conditional move on integer condition code.
619 // The first argument specifies the ICC register: %icc or %xcc
620 { "MOVA", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
621 { "MOVN", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
622 { "MOVNE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
623 { "MOVE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
624 { "MOVG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
625 { "MOVLE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
626 { "MOVGE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
627 { "MOVL", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
628 { "MOVGU", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
629 { "MOVLEU", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
630 { "MOVCC", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
631 { "MOVCS", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
632 { "MOVPOS", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
633 { "MOVNEG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
634 { "MOVVC", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
635 { "MOVVS", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
637 // Conditional move (of integer register) on floating point condition code.
638 // The first argument is the FCCn register (0 <= n <= 3).
639 // Note that the enum name above is not the same as the assembly mnemonic
640 // because some of the assembly mnemonics are the same as the move on
641 // integer CC (e.g., MOVG), and we cannot have the same enum entry twice.
642 { "MOVA", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
643 { "MOVN", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
644 { "MOVU", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
645 { "MOVG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
646 { "MOVUG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
647 { "MOVL", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
648 { "MOVUL", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
649 { "MOVLG", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
650 { "MOVNE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
651 { "MOVE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
652 { "MOVUE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
653 { "MOVGE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
654 { "MOVUGE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
655 { "MOVLE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
656 { "MOVULE", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
657 { "MOVO", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_INT_FLAG },
659 // Conditional move of floating point register on each of the above:
660 // i. on integer comparison with zero.
661 // ii. on integer condition code
662 // iii. on floating point condition code
663 // Note that the same set is repeated for S,D,Q register classes.
664 { "FMOVRSZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
665 { "FMOVRSLEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
666 { "FMOVRSLZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
667 { "FMOVRSNZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
668 { "FMOVRSGZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
669 { "FMOVRSGEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
671 { "FMOVSA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
672 { "FMOVSN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
673 { "FMOVSNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
674 { "FMOVSE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
675 { "FMOVSG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
676 { "FMOVSLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
677 { "FMOVSGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
678 { "FMOVSL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
679 { "FMOVSGU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
680 { "FMOVSLEU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
681 { "FMOVSCC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
682 { "FMOVSCS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
683 { "FMOVSPOS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
684 { "FMOVSNEG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
685 { "FMOVSVC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
686 { "FMOVSVS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
688 { "FMOVSA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
689 { "FMOVSN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
690 { "FMOVSU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
691 { "FMOVSG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
692 { "FMOVSUG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
693 { "FMOVSL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
694 { "FMOVSUL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
695 { "FMOVSLG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
696 { "FMOVSNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
697 { "FMOVSE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
698 { "FMOVSUE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
699 { "FMOVSGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
700 { "FMOVSUGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
701 { "FMOVSLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
702 { "FMOVSULE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
703 { "FMOVSO", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
705 { "FMOVRDZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
706 { "FMOVRDLEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
707 { "FMOVRDLZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
708 { "FMOVRDNZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
709 { "FMOVRDGZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
710 { "FMOVRDGEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
712 { "FMOVDA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
713 { "FMOVDN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
714 { "FMOVDNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
715 { "FMOVDE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
716 { "FMOVDG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
717 { "FMOVDLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
718 { "FMOVDGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
719 { "FMOVDL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
720 { "FMOVDGU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
721 { "FMOVDLEU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
722 { "FMOVDCC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
723 { "FMOVDCS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
724 { "FMOVDPOS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
725 { "FMOVDNEG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
726 { "FMOVDVC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
727 { "FMOVDVS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
729 { "FMOVDA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
730 { "FMOVDN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
731 { "FMOVDU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
732 { "FMOVDG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
733 { "FMOVDUG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
734 { "FMOVDL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
735 { "FMOVDUL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
736 { "FMOVDLG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
737 { "FMOVDNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
738 { "FMOVDE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
739 { "FMOVDUE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
740 { "FMOVDGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
741 { "FMOVDUGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
742 { "FMOVDLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
743 { "FMOVDULE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
744 { "FMOVDO", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
746 { "FMOVRQZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
747 { "FMOVRQLEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
748 { "FMOVRQLZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
749 { "FMOVRQNZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
750 { "FMOVRQGZ", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
751 { "FMOVRQGEZ",3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CONDL_FLAG | M_FLOAT_FLAG | M_INT_FLAG },
753 { "FMOVQA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
754 { "FMOVQN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
755 { "FMOVQNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
756 { "FMOVQE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
757 { "FMOVQG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
758 { "FMOVQLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
759 { "FMOVQGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
760 { "FMOVQL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
761 { "FMOVQGU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
762 { "FMOVQLEU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
763 { "FMOVQCC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
764 { "FMOVQCS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
765 { "FMOVQPOS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
766 { "FMOVQNEG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
767 { "FMOVQVC", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
768 { "FMOVQVS", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
770 { "FMOVQA", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
771 { "FMOVQN", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
772 { "FMOVQU", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
773 { "FMOVQG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
774 { "FMOVQUG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
775 { "FMOVQL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
776 { "FMOVQUL", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
777 { "FMOVQLG", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
778 { "FMOVQNE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
779 { "FMOVQE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
780 { "FMOVQUE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
781 { "FMOVQGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
782 { "FMOVQUGE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
783 { "FMOVQLE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
784 { "FMOVQULE", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
785 { "FMOVQO", 3, 2, 0, false, 0, 2, SPARC_SINGLE, M_CC_FLAG | M_FLOAT_FLAG },
787 // Load integer instructions
788 // Latency includes 1 cycle for address generation (Sparc IIi)
789 // Signed loads of less than 64 bits need an extra cycle for sign-extension.
791 // Not reflected here: After a 3-cycle loads, all subsequent consecutive
792 // loads also require 3 cycles to avoid contention for the load return
793 // stage. Latency returns to 2 cycles after the first cycle with no load.
794 { "LDSB", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
795 { "LDSH", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
796 { "LDSW", 3, 2, (1 << 12) - 1, true, 0, 3, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
797 { "LDUB", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
798 { "LDUH", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
799 { "LDUW", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
800 { "LDX", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_INT_FLAG | M_LOAD_FLAG },
802 // Load floating-point instructions
803 // Latency includes 1 cycle for address generation (Sparc IIi)
804 { "LD", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_FLOAT_FLAG | M_LOAD_FLAG },
805 { "LDD", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_FLOAT_FLAG | M_LOAD_FLAG },
806 { "LDQ", 3, 2, (1 << 12) - 1, true, 0, 2, SPARC_LD, M_FLOAT_FLAG | M_LOAD_FLAG },
808 // Store integer instructions
809 // Latency includes 1 cycle for address generation (Sparc IIi)
810 { "STB", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
811 { "STH", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
812 { "STW", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
813 { "STX", 3, -1, (1 << 12) - 1, true, 0, 3, SPARC_ST, M_INT_FLAG | M_STORE_FLAG },
815 // Store floating-point instructions (Sparc IIi)
816 { "ST", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_FLOAT_FLAG | M_STORE_FLAG},
817 { "STD", 3, -1, (1 << 12) - 1, true, 0, 2, SPARC_ST, M_FLOAT_FLAG | M_STORE_FLAG},
819 // Call, Return and "Jump and link".
820 // Latency includes the delay slot.
821 { "CALL", 1, -1, (1 << 29) - 1, true, 1, 2, SPARC_CTI, M_BRANCH_FLAG | M_CALL_FLAG},
822 { "JMPL", 3, -1, (1 << 12) - 1, true, 1, 2, SPARC_CTI, M_BRANCH_FLAG | M_CALL_FLAG},
823 { "RETURN", 2, -1, 0, false, 1, 2, SPARC_CTI, M_BRANCH_FLAG | M_RET_FLAG },
825 // Synthetic phi operation for near-SSA form of machine code
826 // Number of operands is variable, indicated by -1. Result is the first op.
828 { "PHI", -1, 0, 0, false, 0, 0, SPARC_INV, M_DUMMY_PHI_FLAG },
834 //---------------------------------------------------------------------------
835 // class UltraSparcInstrInfo
838 // Information about individual instructions.
839 // Most information is stored in the SparcMachineInstrDesc array above.
840 // Other information is computed on demand, and most such functions
841 // default to member functions in base class MachineInstrInfo.
842 //---------------------------------------------------------------------------
844 class UltraSparcInstrInfo : public MachineInstrInfo {
846 /*ctor*/ UltraSparcInstrInfo();
848 virtual bool hasResultInterlock (MachineOpCode opCode)
850 // All UltraSPARC instructions have interlocks (note that delay slots
851 // are not considered here).
852 // However, instructions that use the result of an FCMP produce a
853 // 9-cycle stall if they are issued less than 3 cycles after the FCMP.
854 // Force the compiler to insert a software interlock (i.e., gap of
855 // 2 other groups, including NOPs if necessary).
856 return (opCode == FCMPS || opCode == FCMPD || opCode == FCMPQ);
868 class UltraSparcRegInfo : public MachineRegInfo
880 // WARNING: If the above enum order must be changed, also modify
881 // getRegisterClassOfValue method below since it assumes this particular
882 // order for efficiency.
885 // reverse pointer to get info about the ultra sparc machine
886 const UltraSparc *const UltraSparcInfo;
888 // Int arguments can be passed in 6 int regs - %o0 to %o5 (cannot be changed)
889 unsigned const NumOfIntArgRegs;
891 // Float arguments can be passed in this many regs - can be canged if needed
892 // %f0 - %f5 are used (can hold 6 floats or 3 doubles)
893 unsigned const NumOfFloatArgRegs;
895 //void setCallArgColor(LiveRange *const LR, const unsigned RegNo) const;
897 void setCallOrRetArgCol(LiveRange *const LR, const unsigned RegNo,
898 const MachineInstr *MI,AddedInstrMapType &AIMap)const;
900 MachineInstr * getCopy2RegMI(const Value *SrcVal, const unsigned Reg,
901 unsigned RegClassID) const ;
906 UltraSparcRegInfo(const UltraSparc *const USI ) : UltraSparcInfo(USI),
910 MachineRegClassArr.push_back( new SparcIntRegClass(IntRegClassID) );
911 MachineRegClassArr.push_back( new SparcFloatRegClass(FloatRegClassID) );
912 MachineRegClassArr.push_back( new SparcIntCCRegClass(IntCCRegClassID) );
913 MachineRegClassArr.push_back( new SparcFloatCCRegClass(FloatCCRegClassID));
915 assert( SparcFloatRegOrder::StartOfNonVolatileRegs == 6 &&
916 "6 Float regs are used for float arg passing");
920 ~UltraSparcRegInfo(void) { } // empty destructor
923 inline const UltraSparc & getUltraSparcInfo() const {
924 return *UltraSparcInfo;
929 inline unsigned getRegClassIDOfValue (const Value *const Val,
930 bool isCCReg = false) const {
932 Type::PrimitiveID ty = (Val->getType())->getPrimitiveID();
936 if( ty && ty <= Type::LongTyID || (ty == Type::PointerTyID) )
937 res = IntRegClassID; // sparc int reg (ty=0: void)
938 else if( ty <= Type::DoubleTyID)
939 res = FloatRegClassID; // sparc float reg class
941 cout << "TypeID: " << ty << endl;
942 assert(0 && "Cannot resolve register class for type");
947 return res + 2; // corresponidng condition code regiser
954 // returns the register tha contains always zero
955 inline int getZeroRegNum() const { return SparcIntRegOrder::g0; }
957 // returns the reg used for pushing the address when a method is called.
958 // This can be used for other purposes between calls
959 unsigned getCallAddressReg() const { return SparcIntRegOrder::o7; }
962 // and when we return from a method. It should be made sure that this
963 // register contains the return value when a return instruction is reached.
964 unsigned getReturnAddressReg() const { return SparcIntRegOrder::i7; }
966 void colorArgs(const Method *const Meth, LiveRangeInfo& LRI) const;
968 static void printReg(const LiveRange *const LR) ;
970 void colorCallArgs(vector<const Instruction *> & CallInstrList,
972 AddedInstrMapType& AddedInstrMap ) const;
974 void colorRetArg(vector<const Instruction *> &
975 RetInstrList, LiveRangeInfo& LRI,
976 AddedInstrMapType &AddedInstrMap) const;
979 // this method provides a unique number for each register
980 inline int getUnifiedRegNum(int RegClassID, int reg) const {
982 if( RegClassID == IntRegClassID && reg < 32 )
984 else if ( RegClassID == FloatRegClassID && reg < 64)
985 return reg + 32; // we have 32 int regs
986 else if( RegClassID == FloatCCRegClassID && reg < 4)
987 return reg + 32 + 64; // 32 int, 64 float
988 else if( RegClassID == IntCCRegClassID )
989 return 4+ 32 + 64; // only int cc reg
991 assert(0 && "Invalid register class or reg number");
995 // given the unified register number, this gives the name
996 inline const string getUnifiedRegName(int reg) const {
998 return SparcIntRegOrder::getRegName(reg);
999 else if ( reg < (64 + 32) )
1000 return SparcFloatRegOrder::getRegName( reg - 32);
1001 else if( reg < (64+32+4) )
1002 return SparcFloatCCRegOrder::getRegName( reg -32 - 64);
1003 else if ( reg == 64+32+4)
1004 return "xcc"; // only integer cc reg
1006 assert(0 && "Invalid register number");
1014 /*---------------------------------------------------------------------------
1015 Scheduling guidelines for SPARC IIi:
1017 I-Cache alignment rules (pg 326)
1018 -- Align a branch target instruction so that it's entire group is within
1019 the same cache line (may be 1-4 instructions).
1020 ** Don't let a branch that is predicted taken be the last instruction
1021 on an I-cache line: delay slot will need an entire line to be fetched
1022 -- Make a FP instruction or a branch be the 4th instruction in a group.
1023 For branches, there are tradeoffs in reordering to make this happen
1025 ** Don't put a branch in a group that crosses a 32-byte boundary!
1026 An artificial branch is inserted after every 32 bytes, and having
1027 another branch will force the group to be broken into 2 groups.
1030 -- Don't let a loop span two memory pages, if possible
1032 Branch prediction performance:
1033 -- Don't make the branch in a delay slot the target of a branch
1034 -- Try not to have 2 predicted branches within a group of 4 instructions
1035 (because each such group has a single branch target field).
1036 -- Try to align branches in slots 0, 2, 4 or 6 of a cache line (to avoid
1037 the wrong prediction bits being used in some cases).
1039 D-Cache timing constraints:
1040 -- Signed int loads of less than 64 bits have 3 cycle latency, not 2
1041 -- All other loads that hit in D-Cache have 2 cycle latency
1042 -- All loads are returned IN ORDER, so a D-Cache miss will delay a later hit
1043 -- Mis-aligned loads or stores cause a trap. In particular, replace
1044 mis-aligned FP double precision l/s with 2 single-precision l/s.
1045 -- Simulations of integer codes show increase in avg. group size of
1046 33% when code (including esp. non-faulting loads) is moved across
1047 one branch, and 50% across 2 branches.
1049 E-Cache timing constraints:
1050 -- Scheduling for E-cache (D-Cache misses) is effective (due to load buffering)
1052 Store buffer timing constraints:
1053 -- Stores can be executed in same cycle as instruction producing the value
1054 -- Stores are buffered and have lower priority for E-cache until
1055 highwater mark is reached in the store buffer (5 stores)
1057 Pipeline constraints:
1058 -- Shifts can only use IEU0.
1059 -- CC setting instructions can only use IEU1.
1060 -- Several other instructions must only use IEU1:
1061 EDGE(?), ARRAY(?), CALL, JMPL, BPr, PST, and FCMP.
1062 -- Two instructions cannot store to the same register file in a single cycle
1063 (single write port per file).
1065 Issue and grouping constraints:
1066 -- FP and branch instructions must use slot 4.
1067 -- Shift instructions cannot be grouped with other IEU0-specific instructions.
1068 -- CC setting instructions cannot be grouped with other IEU1-specific instrs.
1069 -- Several instructions must be issued in a single-instruction group:
1070 MOVcc or MOVr, MULs/x and DIVs/x, SAVE/RESTORE, many others
1071 -- A CALL or JMPL breaks a group, ie, is not combined with subsequent instrs.
1075 Branch delay slot scheduling rules:
1076 -- A CTI couple (two back-to-back CTI instructions in the dynamic stream)
1077 has a 9-instruction penalty: the entire pipeline is flushed when the
1078 second instruction reaches stage 9 (W-Writeback).
1079 -- Avoid putting multicycle instructions, and instructions that may cause
1080 load misses, in the delay slot of an annulling branch.
1081 -- Avoid putting WR, SAVE..., RESTORE and RETURN instructions in the
1082 delay slot of an annulling branch.
1084 *--------------------------------------------------------------------------- */
1086 //---------------------------------------------------------------------------
1087 // List of CPUResources for UltraSPARC IIi.
1088 //---------------------------------------------------------------------------
1090 const CPUResource AllIssueSlots( "All Instr Slots", 4);
1091 const CPUResource IntIssueSlots( "Int Instr Slots", 3);
1092 const CPUResource First3IssueSlots("Instr Slots 0-3", 3);
1093 const CPUResource LSIssueSlots( "Load-Store Instr Slot", 1);
1094 const CPUResource CTIIssueSlots( "Ctrl Transfer Instr Slot", 1);
1095 const CPUResource FPAIssueSlots( "Int Instr Slot 1", 1);
1096 const CPUResource FPMIssueSlots( "Int Instr Slot 1", 1);
1098 // IEUN instructions can use either Alu and should use IAluN.
1099 // IEU0 instructions must use Alu 1 and should use both IAluN and IAlu0.
1100 // IEU1 instructions must use Alu 2 and should use both IAluN and IAlu1.
1101 const CPUResource IAluN("Int ALU 1or2", 2);
1102 const CPUResource IAlu0("Int ALU 1", 1);
1103 const CPUResource IAlu1("Int ALU 2", 1);
1105 const CPUResource LSAluC1("Load/Store Unit Addr Cycle", 1);
1106 const CPUResource LSAluC2("Load/Store Unit Issue Cycle", 1);
1107 const CPUResource LdReturn("Load Return Unit", 1);
1109 const CPUResource FPMAluC1("FP Mul/Div Alu Cycle 1", 1);
1110 const CPUResource FPMAluC2("FP Mul/Div Alu Cycle 2", 1);
1111 const CPUResource FPMAluC3("FP Mul/Div Alu Cycle 3", 1);
1113 const CPUResource FPAAluC1("FP Other Alu Cycle 1", 1);
1114 const CPUResource FPAAluC2("FP Other Alu Cycle 2", 1);
1115 const CPUResource FPAAluC3("FP Other Alu Cycle 3", 1);
1117 const CPUResource IRegReadPorts("Int Reg ReadPorts", INT_MAX); // CHECK
1118 const CPUResource IRegWritePorts("Int Reg WritePorts", 2); // CHECK
1119 const CPUResource FPRegReadPorts("FP Reg Read Ports", INT_MAX); // CHECK
1120 const CPUResource FPRegWritePorts("FP Reg Write Ports", 1); // CHECK
1122 const CPUResource CTIDelayCycle( "CTI delay cycle", 1);
1123 const CPUResource FCMPDelayCycle("FCMP delay cycle", 1);
1126 //---------------------------------------------------------------------------
1127 // const InstrClassRUsage SparcRUsageDesc[]
1130 // Resource usage information for instruction in each scheduling class.
1131 // The InstrRUsage Objects for individual classes are specified first.
1132 // Note that fetch and decode are decoupled from the execution pipelines
1133 // via an instr buffer, so they are not included in the cycles below.
1134 //---------------------------------------------------------------------------
1136 const InstrClassRUsage NoneClassRUsage = {
1140 /* maxIssueNum */ 4,
1141 /* isSingleIssue */ false,
1142 /* breaksGroup */ false,
1146 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1160 const InstrClassRUsage IEUNClassRUsage = {
1164 /* maxIssueNum */ 3,
1165 /* isSingleIssue */ false,
1166 /* breaksGroup */ false,
1170 /* feasibleSlots[] */ { 0, 1, 2 },
1174 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1175 { IntIssueSlots.rid, 0, 1 },
1176 /*Cycle E */ { IAluN.rid, 1, 1 },
1181 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1185 const InstrClassRUsage IEU0ClassRUsage = {
1189 /* maxIssueNum */ 1,
1190 /* isSingleIssue */ false,
1191 /* breaksGroup */ false,
1195 /* feasibleSlots[] */ { 0, 1, 2 },
1199 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1200 { IntIssueSlots.rid, 0, 1 },
1201 /*Cycle E */ { IAluN.rid, 1, 1 },
1202 { IAlu0.rid, 1, 1 },
1207 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1211 const InstrClassRUsage IEU1ClassRUsage = {
1215 /* maxIssueNum */ 1,
1216 /* isSingleIssue */ false,
1217 /* breaksGroup */ false,
1221 /* feasibleSlots[] */ { 0, 1, 2 },
1225 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1226 { IntIssueSlots.rid, 0, 1 },
1227 /*Cycle E */ { IAluN.rid, 1, 1 },
1228 { IAlu1.rid, 1, 1 },
1233 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1237 const InstrClassRUsage FPMClassRUsage = {
1241 /* maxIssueNum */ 1,
1242 /* isSingleIssue */ false,
1243 /* breaksGroup */ false,
1247 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1251 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1252 { FPMIssueSlots.rid, 0, 1 },
1253 /*Cycle E */ { FPRegReadPorts.rid, 1, 1 },
1254 /*Cycle C */ { FPMAluC1.rid, 2, 1 },
1255 /*Cycle N1*/ { FPMAluC2.rid, 3, 1 },
1256 /*Cycle N1*/ { FPMAluC3.rid, 4, 1 },
1258 /*Cycle W */ { FPRegWritePorts.rid, 6, 1 }
1262 const InstrClassRUsage FPAClassRUsage = {
1266 /* maxIssueNum */ 1,
1267 /* isSingleIssue */ false,
1268 /* breaksGroup */ false,
1272 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1276 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1277 { FPAIssueSlots.rid, 0, 1 },
1278 /*Cycle E */ { FPRegReadPorts.rid, 1, 1 },
1279 /*Cycle C */ { FPAAluC1.rid, 2, 1 },
1280 /*Cycle N1*/ { FPAAluC2.rid, 3, 1 },
1281 /*Cycle N1*/ { FPAAluC3.rid, 4, 1 },
1283 /*Cycle W */ { FPRegWritePorts.rid, 6, 1 }
1287 const InstrClassRUsage LDClassRUsage = {
1291 /* maxIssueNum */ 1,
1292 /* isSingleIssue */ false,
1293 /* breaksGroup */ false,
1297 /* feasibleSlots[] */ { 0, 1, 2, },
1301 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1302 { First3IssueSlots.rid, 0, 1 },
1303 { LSIssueSlots.rid, 0, 1 },
1304 /*Cycle E */ { LSAluC1.rid, 1, 1 },
1305 /*Cycle C */ { LSAluC2.rid, 2, 1 },
1306 { LdReturn.rid, 2, 1 },
1310 /*Cycle W */ { IRegWritePorts.rid, 6, 1 }
1314 const InstrClassRUsage STClassRUsage = {
1318 /* maxIssueNum */ 1,
1319 /* isSingleIssue */ false,
1320 /* breaksGroup */ false,
1324 /* feasibleSlots[] */ { 0, 1, 2 },
1328 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1329 { First3IssueSlots.rid, 0, 1 },
1330 { LSIssueSlots.rid, 0, 1 },
1331 /*Cycle E */ { LSAluC1.rid, 1, 1 },
1332 /*Cycle C */ { LSAluC2.rid, 2, 1 }
1340 const InstrClassRUsage CTIClassRUsage = {
1344 /* maxIssueNum */ 1,
1345 /* isSingleIssue */ false,
1346 /* breaksGroup */ false,
1350 /* feasibleSlots[] */ { 0, 1, 2, 3 },
1354 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1355 { CTIIssueSlots.rid, 0, 1 },
1356 /*Cycle E */ { IAlu0.rid, 1, 1 },
1357 /*Cycles E-C */ { CTIDelayCycle.rid, 1, 2 }
1366 const InstrClassRUsage SingleClassRUsage = {
1370 /* maxIssueNum */ 1,
1371 /* isSingleIssue */ true,
1372 /* breaksGroup */ false,
1376 /* feasibleSlots[] */ { 0 },
1380 /*Cycle G */ { AllIssueSlots.rid, 0, 1 },
1381 { AllIssueSlots.rid, 0, 1 },
1382 { AllIssueSlots.rid, 0, 1 },
1383 { AllIssueSlots.rid, 0, 1 },
1384 /*Cycle E */ { IAlu0.rid, 1, 1 }
1394 const InstrClassRUsage SparcRUsageDesc[] = {
1408 //---------------------------------------------------------------------------
1409 // const InstrIssueDelta SparcInstrIssueDeltas[]
1412 // Changes to issue restrictions information in InstrClassRUsage for
1413 // instructions that differ from other instructions in their class.
1414 //---------------------------------------------------------------------------
1416 const InstrIssueDelta SparcInstrIssueDeltas[] = {
1418 // opCode, isSingleIssue, breaksGroup, numBubbles
1420 // Special cases for single-issue only
1421 // Other single issue cases are below.
1422 //{ LDDA, true, true, 0 },
1423 //{ STDA, true, true, 0 },
1424 //{ LDDF, true, true, 0 },
1425 //{ LDDFA, true, true, 0 },
1426 { ADDC, true, true, 0 },
1427 { ADDCcc, true, true, 0 },
1428 { SUBC, true, true, 0 },
1429 { SUBCcc, true, true, 0 },
1430 //{ SAVE, true, true, 0 },
1431 //{ RESTORE, true, true, 0 },
1432 //{ LDSTUB, true, true, 0 },
1433 //{ SWAP, true, true, 0 },
1434 //{ SWAPA, true, true, 0 },
1435 //{ CAS, true, true, 0 },
1436 //{ CASA, true, true, 0 },
1437 //{ CASX, true, true, 0 },
1438 //{ CASXA, true, true, 0 },
1439 //{ LDFSR, true, true, 0 },
1440 //{ LDFSRA, true, true, 0 },
1441 //{ LDXFSR, true, true, 0 },
1442 //{ LDXFSRA, true, true, 0 },
1443 //{ STFSR, true, true, 0 },
1444 //{ STFSRA, true, true, 0 },
1445 //{ STXFSR, true, true, 0 },
1446 //{ STXFSRA, true, true, 0 },
1447 //{ SAVED, true, true, 0 },
1448 //{ RESTORED, true, true, 0 },
1449 //{ FLUSH, true, true, 9 },
1450 //{ FLUSHW, true, true, 9 },
1451 //{ ALIGNADDR, true, true, 0 },
1452 { RETURN, true, true, 0 },
1453 //{ DONE, true, true, 0 },
1454 //{ RETRY, true, true, 0 },
1455 //{ WR, true, true, 0 },
1456 //{ WRPR, true, true, 4 },
1457 //{ RD, true, true, 0 },
1458 //{ RDPR, true, true, 0 },
1459 //{ TCC, true, true, 0 },
1460 //{ SHUTDOWN, true, true, 0 },
1462 // Special cases for breaking group *before*
1463 // CURRENTLY NOT SUPPORTED!
1464 { CALL, false, false, 0 },
1465 { JMPL, false, false, 0 },
1467 // Special cases for breaking the group *after*
1468 { MULX, true, true, (4+34)/2 },
1469 { FDIVS, false, true, 0 },
1470 { FDIVD, false, true, 0 },
1471 { FDIVQ, false, true, 0 },
1472 { FSQRTS, false, true, 0 },
1473 { FSQRTD, false, true, 0 },
1474 { FSQRTQ, false, true, 0 },
1475 //{ FCMP{LE,GT,NE,EQ}, false, true, 0 },
1477 // Instructions that introduce bubbles
1478 //{ MULScc, true, true, 2 },
1479 //{ SMULcc, true, true, (4+18)/2 },
1480 //{ UMULcc, true, true, (4+19)/2 },
1481 { SDIVX, true, true, 68 },
1482 { UDIVX, true, true, 68 },
1483 //{ SDIVcc, true, true, 36 },
1484 //{ UDIVcc, true, true, 37 },
1485 //{ WR, false, false, 4 },
1486 //{ WRPR, false, false, 4 },
1490 //---------------------------------------------------------------------------
1491 // const InstrRUsageDelta SparcInstrUsageDeltas[]
1494 // Changes to resource usage information in InstrClassRUsage for
1495 // instructions that differ from other instructions in their class.
1496 //---------------------------------------------------------------------------
1498 const InstrRUsageDelta SparcInstrUsageDeltas[] = {
1500 // MachineOpCode, Resource, Start cycle, Num cycles
1503 // JMPL counts as a load/store instruction for issue!
1505 { JMPL, LSIssueSlots.rid, 0, 1 },
1508 // Many instructions cannot issue for the next 2 cycles after an FCMP
1509 // We model that with a fake resource FCMPDelayCycle.
1511 { FCMPS, FCMPDelayCycle.rid, 1, 3 },
1512 { FCMPD, FCMPDelayCycle.rid, 1, 3 },
1513 { FCMPQ, FCMPDelayCycle.rid, 1, 3 },
1515 { MULX, FCMPDelayCycle.rid, 1, 1 },
1516 { SDIVX, FCMPDelayCycle.rid, 1, 1 },
1517 { UDIVX, FCMPDelayCycle.rid, 1, 1 },
1518 //{ SMULcc, FCMPDelayCycle.rid, 1, 1 },
1519 //{ UMULcc, FCMPDelayCycle.rid, 1, 1 },
1520 //{ SDIVcc, FCMPDelayCycle.rid, 1, 1 },
1521 //{ UDIVcc, FCMPDelayCycle.rid, 1, 1 },
1522 { STD, FCMPDelayCycle.rid, 1, 1 },
1523 { FMOVRSZ, FCMPDelayCycle.rid, 1, 1 },
1524 { FMOVRSLEZ,FCMPDelayCycle.rid, 1, 1 },
1525 { FMOVRSLZ, FCMPDelayCycle.rid, 1, 1 },
1526 { FMOVRSNZ, FCMPDelayCycle.rid, 1, 1 },
1527 { FMOVRSGZ, FCMPDelayCycle.rid, 1, 1 },
1528 { FMOVRSGEZ,FCMPDelayCycle.rid, 1, 1 },
1531 // Some instructions are stalled in the GROUP stage if a CTI is in
1534 { LDD, CTIDelayCycle.rid, 1, 1 },
1535 //{ LDDA, CTIDelayCycle.rid, 1, 1 },
1536 //{ LDDSTUB, CTIDelayCycle.rid, 1, 1 },
1537 //{ LDDSTUBA, CTIDelayCycle.rid, 1, 1 },
1538 //{ SWAP, CTIDelayCycle.rid, 1, 1 },
1539 //{ SWAPA, CTIDelayCycle.rid, 1, 1 },
1540 //{ CAS, CTIDelayCycle.rid, 1, 1 },
1541 //{ CASA, CTIDelayCycle.rid, 1, 1 },
1542 //{ CASX, CTIDelayCycle.rid, 1, 1 },
1543 //{ CASXA, CTIDelayCycle.rid, 1, 1 },
1546 // Signed int loads of less than dword size return data in cycle N1 (not C)
1547 // and put all loads in consecutive cycles into delayed load return mode.
1549 { LDSB, LdReturn.rid, 2, -1 },
1550 { LDSB, LdReturn.rid, 3, 1 },
1552 { LDSH, LdReturn.rid, 2, -1 },
1553 { LDSH, LdReturn.rid, 3, 1 },
1555 { LDSW, LdReturn.rid, 2, -1 },
1556 { LDSW, LdReturn.rid, 3, 1 },
1559 #undef EXPLICIT_BUBBLES_NEEDED
1560 #ifdef EXPLICIT_BUBBLES_NEEDED
1562 // MULScc inserts one bubble.
1563 // This means it breaks the current group (captured in UltraSparcSchedInfo)
1564 // *and occupies all issue slots for the next cycle
1566 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1567 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1568 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1569 //{ MULScc, AllIssueSlots.rid, 2, 2-1 },
1572 // SMULcc inserts between 4 and 18 bubbles, depending on #leading 0s in rs1.
1573 // We just model this with a simple average.
1575 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1576 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1577 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1578 //{ SMULcc, AllIssueSlots.rid, 2, ((4+18)/2)-1 },
1580 // SMULcc inserts between 4 and 19 bubbles, depending on #leading 0s in rs1.
1581 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1582 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1583 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1584 //{ UMULcc, AllIssueSlots.rid, 2, ((4+19)/2)-1 },
1587 // MULX inserts between 4 and 34 bubbles, depending on #leading 0s in rs1.
1589 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1590 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1591 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1592 { MULX, AllIssueSlots.rid, 2, ((4+34)/2)-1 },
1595 // SDIVcc inserts 36 bubbles.
1597 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1598 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1599 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1600 //{ SDIVcc, AllIssueSlots.rid, 2, 36-1 },
1602 // UDIVcc inserts 37 bubbles.
1603 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1604 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1605 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1606 //{ UDIVcc, AllIssueSlots.rid, 2, 37-1 },
1609 // SDIVX inserts 68 bubbles.
1611 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1612 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1613 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1614 { SDIVX, AllIssueSlots.rid, 2, 68-1 },
1617 // UDIVX inserts 68 bubbles.
1619 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1620 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1621 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1622 { UDIVX, AllIssueSlots.rid, 2, 68-1 },
1625 // WR inserts 4 bubbles.
1627 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1628 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1629 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1630 //{ WR, AllIssueSlots.rid, 2, 68-1 },
1633 // WRPR inserts 4 bubbles.
1635 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1636 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1637 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1638 //{ WRPR, AllIssueSlots.rid, 2, 68-1 },
1641 // DONE inserts 9 bubbles.
1643 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1644 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1645 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1646 //{ DONE, AllIssueSlots.rid, 2, 9-1 },
1649 // RETRY inserts 9 bubbles.
1651 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1652 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1653 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1654 //{ RETRY, AllIssueSlots.rid, 2, 9-1 },
1656 #endif EXPLICIT_BUBBLES_NEEDED
1661 // Additional delays to be captured in code:
1662 // 1. RDPR from several state registers (page 349)
1663 // 2. RD from *any* register (page 349)
1664 // 3. Writes to TICK, PSTATE, TL registers and FLUSH{W} instr (page 349)
1665 // 4. Integer store can be in same group as instr producing value to store.
1666 // 5. BICC and BPICC can be in the same group as instr producing CC (pg 350)
1667 // 6. FMOVr cannot be in the same or next group as an IEU instr (pg 351).
1668 // 7. The second instr. of a CTI group inserts 9 bubbles (pg 351)
1669 // 8. WR{PR}, SVAE, SAVED, RESTORE, RESTORED, RETURN, RETRY, and DONE that
1670 // follow an annulling branch cannot be issued in the same group or in
1671 // the 3 groups following the branch.
1672 // 9. A predicted annulled load does not stall dependent instructions.
1673 // Other annulled delay slot instructions *do* stall dependents, so
1674 // nothing special needs to be done for them during scheduling.
1675 //10. Do not put a load use that may be annulled in the same group as the
1676 // branch. The group will stall until the load returns.
1677 //11. Single-prec. FP loads lock 2 registers, for dependency checking.
1680 // Additional delays we cannot or will not capture:
1681 // 1. If DCTI is last word of cache line, it is delayed until next line can be
1682 // fetched. Also, other DCTI alignment-related delays (pg 352)
1683 // 2. Load-after-store is delayed by 7 extra cycles if load hits in D-Cache.
1684 // Also, several other store-load and load-store conflicts (pg 358)
1685 // 3. MEMBAR, LD{X}FSR, LDD{A} and a bunch of other load stalls (pg 358)
1686 // 4. There can be at most 8 outstanding buffered store instructions
1687 // (including some others like MEMBAR, LDSTUB, CAS{AX}, and FLUSH)
1691 //---------------------------------------------------------------------------
1692 // class UltraSparcSchedInfo
1695 // Interface to instruction scheduling information for UltraSPARC.
1696 // The parameter values above are based on UltraSPARC IIi.
1697 //---------------------------------------------------------------------------
1700 class UltraSparcSchedInfo: public MachineSchedInfo {
1702 /*ctor*/ UltraSparcSchedInfo (const MachineInstrInfo* mii);
1703 /*dtor*/ virtual ~UltraSparcSchedInfo () {}
1705 virtual void initializeResources ();
1709 //---------------------------------------------------------------------------
1710 // class UltraSparcMachine
1713 // Primary interface to machine description for the UltraSPARC.
1714 // Primarily just initializes machine-dependent parameters in
1715 // class TargetMachine, and creates machine-dependent subclasses
1716 // for classes such as InstrInfo, SchedInfo and RegInfo.
1717 //---------------------------------------------------------------------------
1719 class UltraSparc : public TargetMachine {
1721 UltraSparcInstrInfo instrInfo;
1722 UltraSparcSchedInfo schedInfo;
1723 UltraSparcRegInfo regInfo;
1726 virtual ~UltraSparc() {}
1728 virtual const MachineInstrInfo &getInstrInfo() const { return instrInfo; }
1729 virtual const MachineSchedInfo &getSchedInfo() const { return schedInfo; }
1730 virtual const MachineRegInfo &getRegInfo() const { return regInfo; }
1732 // compileMethod - For the sparc, we do instruction selection, followed by
1733 // delay slot scheduling, then register allocation.
1735 virtual bool compileMethod(Method *M);
1738 // emitAssembly - Output assembly language code (a .s file) for the specified
1739 // module. The specified module must have been compiled before this may be
1742 virtual void emitAssembly(const Module *M, ostream &OutStr);