src/openpower/decoder/isa/caller.py

   1 # SPDX-License-Identifier: LGPLv3+
   2 # Copyright (C) 2020, 2021 Luke Kenneth Casson Leighton <lkcl@lkcl.net>
   3 # Copyright (C) 2020 Michael Nolan
   4 # Funded by NLnet http://nlnet.nl
   5 """core of the python-based POWER9 simulator
   6
   7 this is part of a cycle-accurate POWER9 simulator.  its primary purpose is
   8 not speed, it is for both learning and educational purposes, as well as
   9 a method of verifying the HDL.
  10
  11 related bugs:
  12
  13 * https://bugs.libre-soc.org/show_bug.cgi?id=424
  14 """
  15
  16 import re
  17 from nmigen.back.pysim import Settle
  18 from functools import wraps
  19 from copy import copy, deepcopy
  20 from openpower.decoder.orderedset import OrderedSet
  21 from openpower.decoder.selectable_int import (
  22     SelectableIntMapping,
  23     FieldSelectableInt,
  24     SelectableInt,
  25     selectconcat,
  26 )
  27 from openpower.decoder.power_enums import (spr_dict, spr_byname, XER_bits,
  28                                            insns, MicrOp,
  29                                            In1Sel, In2Sel, In3Sel,
  30                                            OutSel, CRInSel, CROutSel, LDSTMode,
  31                                            SVP64RMMode, SVP64PredMode,
  32                                            SVP64PredInt, SVP64PredCR,
  33                                            SVP64LDSTmode)
  34
  35 from openpower.decoder.power_enums import SVPtype
  36
  37 from openpower.decoder.helpers import (exts, gtu, ltu, undefined,
  38                                        ISACallerHelper, ISAFPHelpers)
  39 from openpower.consts import PIb, MSRb  # big-endian (PowerISA versions)
  40 from openpower.consts import (SVP64MODE,
  41                               SVP64CROffs,
  42                               )
  43 from openpower.decoder.power_svp64 import SVP64RM, decode_extra
  44
  45 from openpower.decoder.isa.radixmmu import RADIX
  46 from openpower.decoder.isa.mem import Mem, swap_order, MemException
  47 from openpower.decoder.isa.svshape import SVSHAPE
  48 from openpower.decoder.isa.svstate import SVP64State
  49
  50
  51 from openpower.util import log
  52
  53 from collections import namedtuple
  54 import math
  55 import sys
  56
  57 instruction_info = namedtuple('instruction_info',
  58                               'func read_regs uninit_regs write_regs ' +
  59                               'special_regs op_fields form asmregs')
  60
  61 special_sprs = {
  62     'LR': 8,
  63     'CTR': 9,
  64     'TAR': 815,
  65     'XER': 1,
  66     'VRSAVE': 256}
  67
  68
  69 REG_SORT_ORDER = {
  70     # TODO (lkcl): adjust other registers that should be in a particular order
  71     # probably CA, CA32, and CR
  72     "FRT": 0,
  73     "FRA": 0,
  74     "FRB": 0,
  75     "FRC": 0,
  76     "FRS": 0,
  77     "RT": 0,
  78     "RA": 0,
  79     "RB": 0,
  80     "RC": 0,
  81     "RS": 0,
  82     "BI": 0,
  83     "CR": 0,
  84     "LR": 0,
  85     "CTR": 0,
  86     "TAR": 0,
  87     "MSR": 0,
  88     "SVSTATE": 0,
  89     "SVSHAPE0": 0,
  90     "SVSHAPE1": 0,
  91     "SVSHAPE2": 0,
  92     "SVSHAPE3": 0,
  93
  94     "CA": 0,
  95     "CA32": 0,
  96
  97     "overflow": 7,  # should definitely be last
  98 }
  99
 100 fregs = ['FRA', 'FRB', 'FRC', 'FRS', 'FRT']
 101
 102
 103 def create_args(reglist, extra=None):
 104     retval = list(OrderedSet(reglist))
 105     retval.sort(key=lambda reg: REG_SORT_ORDER.get(reg, 0))
 106     if extra is not None:
 107         return [extra] + retval
 108     return retval
 109
 110
 111 class GPR(dict):
 112     def __init__(self, decoder, isacaller, svstate, regfile):
 113         dict.__init__(self)
 114         self.sd = decoder
 115         self.isacaller = isacaller
 116         self.svstate = svstate
 117         for i in range(len(regfile)):
 118             self[i] = SelectableInt(regfile[i], 64)
 119
 120     def __call__(self, ridx):
 121         if isinstance(ridx, SelectableInt):
 122             ridx = ridx.value
 123         return self[ridx]
 124
 125     def set_form(self, form):
 126         self.form = form
 127
 128     def __setitem__(self, rnum, value):
 129         # rnum = rnum.value # only SelectableInt allowed
 130         log("GPR setitem", rnum, value)
 131         if isinstance(rnum, SelectableInt):
 132             rnum = rnum.value
 133         dict.__setitem__(self, rnum, value)
 134
 135     def getz(self, rnum):
 136         # rnum = rnum.value # only SelectableInt allowed
 137         log("GPR getzero?", rnum)
 138         if rnum == 0:
 139             return SelectableInt(0, 64)
 140         return self[rnum]
 141
 142     def _get_regnum(self, attr):
 143         getform = self.sd.sigforms[self.form]
 144         rnum = getattr(getform, attr)
 145         return rnum
 146
 147     def ___getitem__(self, attr):
 148         """ XXX currently not used
 149         """
 150         rnum = self._get_regnum(attr)
 151         log("GPR getitem", attr, rnum)
 152         return self.regfile[rnum]
 153
 154     def dump(self, printout=True):
 155         res = []
 156         for i in range(len(self)):
 157             res.append(self[i].value)
 158         if printout:
 159             for i in range(0, len(res), 8):
 160                 s = []
 161                 for j in range(8):
 162                     s.append("%08x" % res[i+j])
 163                 s = ' '.join(s)
 164                 print("reg", "%2d" % i, s)
 165         return res
 166
 167
 168 class SPR(dict):
 169     def __init__(self, dec2, initial_sprs={}):
 170         self.sd = dec2
 171         dict.__init__(self)
 172         for key, v in initial_sprs.items():
 173             if isinstance(key, SelectableInt):
 174                 key = key.value
 175             key = special_sprs.get(key, key)
 176             if isinstance(key, int):
 177                 info = spr_dict[key]
 178             else:
 179                 info = spr_byname[key]
 180             if not isinstance(v, SelectableInt):
 181                 v = SelectableInt(v, info.length)
 182             self[key] = v
 183
 184     def __getitem__(self, key):
 185         log("get spr", key)
 186         log("dict", self.items())
 187         # if key in special_sprs get the special spr, otherwise return key
 188         if isinstance(key, SelectableInt):
 189             key = key.value
 190         if isinstance(key, int):
 191             key = spr_dict[key].SPR
 192         key = special_sprs.get(key, key)
 193         if key == 'HSRR0':  # HACK!
 194             key = 'SRR0'
 195         if key == 'HSRR1':  # HACK!
 196             key = 'SRR1'
 197         if key in self:
 198             res = dict.__getitem__(self, key)
 199         else:
 200             if isinstance(key, int):
 201                 info = spr_dict[key]
 202             else:
 203                 info = spr_byname[key]
 204             dict.__setitem__(self, key, SelectableInt(0, info.length))
 205             res = dict.__getitem__(self, key)
 206         log("spr returning", key, res)
 207         return res
 208
 209     def __setitem__(self, key, value):
 210         if isinstance(key, SelectableInt):
 211             key = key.value
 212         if isinstance(key, int):
 213             key = spr_dict[key].SPR
 214             log("spr key", key)
 215         key = special_sprs.get(key, key)
 216         if key == 'HSRR0':  # HACK!
 217             self.__setitem__('SRR0', value)
 218         if key == 'HSRR1':  # HACK!
 219             self.__setitem__('SRR1', value)
 220         log("setting spr", key, value)
 221         dict.__setitem__(self, key, value)
 222
 223     def __call__(self, ridx):
 224         return self[ridx]
 225
 226     def dump(self, printout=True):
 227         res = []
 228         keys = list(self.keys())
 229         # keys.sort()
 230         for k in keys:
 231             sprname = spr_dict.get(k, None)
 232             if sprname is None:
 233                 sprname = k
 234             else:
 235                 sprname = sprname.SPR
 236             res.append((sprname, self[k].value))
 237         if printout:
 238             for sprname, value in res:
 239                 print("    ", sprname, hex(value))
 240         return res
 241
 242
 243 class PC:
 244     def __init__(self, pc_init=0):
 245         self.CIA = SelectableInt(pc_init, 64)
 246         self.NIA = self.CIA + SelectableInt(4, 64)  # only true for v3.0B!
 247
 248     def update_nia(self, is_svp64):
 249         increment = 8 if is_svp64 else 4
 250         self.NIA = self.CIA + SelectableInt(increment, 64)
 251
 252     def update(self, namespace, is_svp64):
 253         """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
 254         """
 255         self.CIA = namespace['NIA'].narrow(64)
 256         self.update_nia(is_svp64)
 257         namespace['CIA'] = self.CIA
 258         namespace['NIA'] = self.NIA
 259
 260
 261 # SVP64 ReMap field
 262 class SVP64RMFields(SelectableIntMapping, bits=24, fields={
 263     "spr": range(24),
 264     # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
 265     "mmode": (0,),
 266     "mask": range(1, 4),
 267     "elwidth": range(4, 6),
 268     "ewsrc": range(6, 8),
 269     "subvl": range(8, 10),
 270     "extra": range(10, 19),
 271     "mode": range(19, 24),
 272     # these cover the same extra field, split into parts as EXTRA2
 273     "extra2": dict(enumerate([
 274         range(10, 12),
 275         range(12, 14),
 276         range(14, 16),
 277         range(16, 18),
 278     ])),
 279     "smask": range(16, 19),
 280     # and here as well, but EXTRA3
 281     "extra3": dict(enumerate([
 282         range(10, 13),
 283         range(13, 16),
 284         range(16, 19),
 285     ])),
 286 }):
 287
 288     def __init__(self, value=0):
 289         super().__init__(value=value)
 290         self.spr = self
 291
 292
 293 SVP64RM_MMODE_SIZE = len(SVP64RMFields.mmode)
 294 SVP64RM_MASK_SIZE = len(SVP64RMFields.mask)
 295 SVP64RM_ELWIDTH_SIZE = len(SVP64RMFields.elwidth)
 296 SVP64RM_EWSRC_SIZE = len(SVP64RMFields.ewsrc)
 297 SVP64RM_SUBVL_SIZE = len(SVP64RMFields.subvl)
 298 SVP64RM_EXTRA2_SPEC_SIZE = len(SVP64RMFields.extra2[0])
 299 SVP64RM_EXTRA3_SPEC_SIZE = len(SVP64RMFields.extra3[0])
 300 SVP64RM_SMASK_SIZE = len(SVP64RMFields.smask)
 301 SVP64RM_MODE_SIZE = len(SVP64RMFields.mode)
 302
 303
 304 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
 305 class SVP64PrefixFields(SelectableIntMapping, bits=32, fields={
 306     "insn": range(32),
 307     # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
 308     "major": range(0, 6),
 309     "pid": (7, 9),
 310     # SVP64 24-bit RM (ReMap)
 311     "rm": ((6, 8) + tuple(range(10, 32))),
 312 }):
 313
 314     def __init__(self, value=0):
 315         super().__init__(value=value)
 316         self.insn = self
 317
 318
 319 SV64P_MAJOR_SIZE = len(SVP64PrefixFields.major)
 320 SV64P_PID_SIZE = len(SVP64PrefixFields.pid)
 321 SV64P_RM_SIZE = len(SVP64PrefixFields.rm)
 322
 323
 324 # CR register fields
 325 # See PowerISA Version 3.0 B Book 1
 326 # Section 2.3.1 Condition Register pages 30 - 31
 327 class CRFields:
 328     LT = FL = 0  # negative, less than, floating-point less than
 329     GT = FG = 1  # positive, greater than, floating-point greater than
 330     EQ = FE = 2  # equal, floating-point equal
 331     SO = FU = 3  # summary overflow, floating-point unordered
 332
 333     def __init__(self, init=0):
 334         # rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
 335         # self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
 336         self.cr = SelectableInt(init, 64)  # underlying reg
 337         # field-selectable versions of Condition Register TODO check bitranges?
 338         self.crl = []
 339         for i in range(8):
 340             bits = tuple(range(i*4+32, (i+1)*4+32))
 341             _cr = FieldSelectableInt(self.cr, bits)
 342             self.crl.append(_cr)
 343
 344 # decode SVP64 predicate integer to reg number and invert
 345
 346
 347 def get_predint(gpr, mask):
 348     r10 = gpr(10)
 349     r30 = gpr(30)
 350     log("get_predint", mask, SVP64PredInt.ALWAYS.value)
 351     if mask == SVP64PredInt.ALWAYS.value:
 352         return 0xffff_ffff_ffff_ffff  # 64 bits of 1
 353     if mask == SVP64PredInt.R3_UNARY.value:
 354         return 1 << (gpr(3).value & 0b111111)
 355     if mask == SVP64PredInt.R3.value:
 356         return gpr(3).value
 357     if mask == SVP64PredInt.R3_N.value:
 358         return ~gpr(3).value
 359     if mask == SVP64PredInt.R10.value:
 360         return gpr(10).value
 361     if mask == SVP64PredInt.R10_N.value:
 362         return ~gpr(10).value
 363     if mask == SVP64PredInt.R30.value:
 364         return gpr(30).value
 365     if mask == SVP64PredInt.R30_N.value:
 366         return ~gpr(30).value
 367
 368 # decode SVP64 predicate CR to reg number and invert status
 369
 370
 371 def _get_predcr(mask):
 372     if mask == SVP64PredCR.LT.value:
 373         return 0, 1
 374     if mask == SVP64PredCR.GE.value:
 375         return 0, 0
 376     if mask == SVP64PredCR.GT.value:
 377         return 1, 1
 378     if mask == SVP64PredCR.LE.value:
 379         return 1, 0
 380     if mask == SVP64PredCR.EQ.value:
 381         return 2, 1
 382     if mask == SVP64PredCR.NE.value:
 383         return 2, 0
 384     if mask == SVP64PredCR.SO.value:
 385         return 3, 1
 386     if mask == SVP64PredCR.NS.value:
 387         return 3, 0
 388
 389 # read individual CR fields (0..VL-1), extract the required bit
 390 # and construct the mask
 391
 392
 393 def get_predcr(crl, mask, vl):
 394     idx, noninv = _get_predcr(mask)
 395     mask = 0
 396     for i in range(vl):
 397         cr = crl[i+SVP64CROffs.CRPred]
 398         if cr[idx].value == noninv:
 399             mask |= (1 << i)
 400     return mask
 401
 402
 403 # TODO, really should just be using PowerDecoder2
 404 def get_pdecode_idx_in(dec2, name):
 405     op = dec2.dec.op
 406     in1_sel = yield op.in1_sel
 407     in2_sel = yield op.in2_sel
 408     in3_sel = yield op.in3_sel
 409     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 410     in1 = yield dec2.e.read_reg1.data
 411     in2 = yield dec2.e.read_reg2.data
 412     in3 = yield dec2.e.read_reg3.data
 413     in1_isvec = yield dec2.in1_isvec
 414     in2_isvec = yield dec2.in2_isvec
 415     in3_isvec = yield dec2.in3_isvec
 416     log("get_pdecode_idx_in in1", name, in1_sel, In1Sel.RA.value,
 417         in1, in1_isvec)
 418     log("get_pdecode_idx_in in2", name, in2_sel, In2Sel.RB.value,
 419         in2, in2_isvec)
 420     log("get_pdecode_idx_in in3", name, in3_sel, In3Sel.RS.value,
 421         in3, in3_isvec)
 422     log("get_pdecode_idx_in FRS in3", name, in3_sel, In3Sel.FRS.value,
 423         in3, in3_isvec)
 424     log("get_pdecode_idx_in FRB in2", name, in2_sel, In2Sel.FRB.value,
 425         in2, in2_isvec)
 426     log("get_pdecode_idx_in FRC in3", name, in3_sel, In3Sel.FRC.value,
 427         in3, in3_isvec)
 428     # identify which regnames map to in1/2/3
 429     if name == 'RA':
 430         if (in1_sel == In1Sel.RA.value or
 431                 (in1_sel == In1Sel.RA_OR_ZERO.value and in1 != 0)):
 432             return in1, in1_isvec
 433         if in1_sel == In1Sel.RA_OR_ZERO.value:
 434             return in1, in1_isvec
 435     elif name == 'RB':
 436         if in2_sel == In2Sel.RB.value:
 437             return in2, in2_isvec
 438         if in3_sel == In3Sel.RB.value:
 439             return in3, in3_isvec
 440     # XXX TODO, RC doesn't exist yet!
 441     elif name == 'RC':
 442         assert False, "RC does not exist yet"
 443     elif name == 'RS':
 444         if in1_sel == In1Sel.RS.value:
 445             return in1, in1_isvec
 446         if in2_sel == In2Sel.RS.value:
 447             return in2, in2_isvec
 448         if in3_sel == In3Sel.RS.value:
 449             return in3, in3_isvec
 450     elif name == 'FRA':
 451         if in1_sel == In1Sel.FRA.value:
 452             return in1, in1_isvec
 453     elif name == 'FRB':
 454         if in2_sel == In2Sel.FRB.value:
 455             return in2, in2_isvec
 456     elif name == 'FRC':
 457         if in3_sel == In3Sel.FRC.value:
 458             return in3, in3_isvec
 459     elif name == 'FRS':
 460         if in1_sel == In1Sel.FRS.value:
 461             return in1, in1_isvec
 462         if in3_sel == In3Sel.FRS.value:
 463             return in3, in3_isvec
 464     return None, False
 465
 466
 467 # TODO, really should just be using PowerDecoder2
 468 def get_pdecode_cr_in(dec2, name):
 469     op = dec2.dec.op
 470     in_sel = yield op.cr_in
 471     in_bitfield = yield dec2.dec_cr_in.cr_bitfield.data
 472     sv_cr_in = yield op.sv_cr_in
 473     spec = yield dec2.crin_svdec.spec
 474     sv_override = yield dec2.dec_cr_in.sv_override
 475     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 476     in1 = yield dec2.e.read_cr1.data
 477     cr_isvec = yield dec2.cr_in_isvec
 478     log("get_pdecode_cr_in", in_sel, CROutSel.CR0.value, in1, cr_isvec)
 479     log("    sv_cr_in", sv_cr_in)
 480     log("    cr_bf", in_bitfield)
 481     log("    spec", spec)
 482     log("    override", sv_override)
 483     # identify which regnames map to in / o2
 484     if name == 'BI':
 485         if in_sel == CRInSel.BI.value:
 486             return in1, cr_isvec
 487     log("get_pdecode_cr_in not found", name)
 488     return None, False
 489
 490
 491 # TODO, really should just be using PowerDecoder2
 492 def get_pdecode_cr_out(dec2, name):
 493     op = dec2.dec.op
 494     out_sel = yield op.cr_out
 495     out_bitfield = yield dec2.dec_cr_out.cr_bitfield.data
 496     sv_cr_out = yield op.sv_cr_out
 497     spec = yield dec2.crout_svdec.spec
 498     sv_override = yield dec2.dec_cr_out.sv_override
 499     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 500     out = yield dec2.e.write_cr.data
 501     o_isvec = yield dec2.o_isvec
 502     log("get_pdecode_cr_out", out_sel, CROutSel.CR0.value, out, o_isvec)
 503     log("    sv_cr_out", sv_cr_out)
 504     log("    cr_bf", out_bitfield)
 505     log("    spec", spec)
 506     log("    override", sv_override)
 507     # identify which regnames map to out / o2
 508     if name == 'CR0':
 509         if out_sel == CROutSel.CR0.value:
 510             return out, o_isvec
 511     log("get_pdecode_cr_out not found", name)
 512     return None, False
 513
 514
 515 # TODO, really should just be using PowerDecoder2
 516 def get_pdecode_idx_out(dec2, name):
 517     op = dec2.dec.op
 518     out_sel = yield op.out_sel
 519     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 520     out = yield dec2.e.write_reg.data
 521     o_isvec = yield dec2.o_isvec
 522     # identify which regnames map to out / o2
 523     if name == 'RA':
 524         log("get_pdecode_idx_out", out_sel, OutSel.RA.value, out, o_isvec)
 525         if out_sel == OutSel.RA.value:
 526             return out, o_isvec
 527     elif name == 'RT':
 528         log("get_pdecode_idx_out", out_sel, OutSel.RT.value,
 529             OutSel.RT_OR_ZERO.value, out, o_isvec,
 530             dec2.dec.RT)
 531         if out_sel == OutSel.RT.value:
 532             return out, o_isvec
 533     elif name == 'RT_OR_ZERO':
 534         log("get_pdecode_idx_out", out_sel, OutSel.RT.value,
 535             OutSel.RT_OR_ZERO.value, out, o_isvec,
 536             dec2.dec.RT)
 537         if out_sel == OutSel.RT_OR_ZERO.value:
 538             return out, o_isvec
 539     elif name == 'FRA':
 540         log("get_pdecode_idx_out", out_sel, OutSel.FRA.value, out, o_isvec)
 541         if out_sel == OutSel.FRA.value:
 542             return out, o_isvec
 543     elif name == 'FRT':
 544         log("get_pdecode_idx_out", out_sel, OutSel.FRT.value,
 545             OutSel.FRT.value, out, o_isvec)
 546         if out_sel == OutSel.FRT.value:
 547             return out, o_isvec
 548     log("get_pdecode_idx_out not found", name, out_sel, out, o_isvec)
 549     return None, False
 550
 551
 552 # TODO, really should just be using PowerDecoder2
 553 def get_pdecode_idx_out2(dec2, name):
 554     # check first if register is activated for write
 555     op = dec2.dec.op
 556     out_sel = yield op.out_sel
 557     out = yield dec2.e.write_ea.data
 558     o_isvec = yield dec2.o2_isvec
 559     out_ok = yield dec2.e.write_ea.ok
 560     log("get_pdecode_idx_out2", name, out_sel, out, out_ok, o_isvec)
 561     if not out_ok:
 562         return None, False
 563
 564     if name == 'RA':
 565         if hasattr(op, "upd"):
 566             # update mode LD/ST uses read-reg A also as an output
 567             upd = yield op.upd
 568             log("get_pdecode_idx_out2", upd, LDSTMode.update.value,
 569                 out_sel, OutSel.RA.value,
 570                 out, o_isvec)
 571             if upd == LDSTMode.update.value:
 572                 return out, o_isvec
 573     if name == 'FRS':
 574         int_op = yield dec2.dec.op.internal_op
 575         fft_en = yield dec2.use_svp64_fft
 576         # if int_op == MicrOp.OP_FP_MADD.value and fft_en:
 577         if fft_en:
 578             log("get_pdecode_idx_out2", out_sel, OutSel.FRS.value,
 579                 out, o_isvec)
 580             return out, o_isvec
 581     return None, False
 582
 583
 584 class ISACaller(ISACallerHelper, ISAFPHelpers):
 585     # decoder2 - an instance of power_decoder2
 586     # regfile - a list of initial values for the registers
 587     # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
 588     # respect_pc - tracks the program counter.  requires initial_insns
 589     def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
 590                  initial_mem=None, initial_msr=0,
 591                  initial_svstate=0,
 592                  initial_insns=None,
 593                  fpregfile=None,
 594                  respect_pc=False,
 595                  disassembly=None,
 596                  initial_pc=0,
 597                  bigendian=False,
 598                  mmu=False,
 599                  icachemmu=False):
 600
 601         self.bigendian = bigendian
 602         self.halted = False
 603         self.is_svp64_mode = False
 604         self.respect_pc = respect_pc
 605         if initial_sprs is None:
 606             initial_sprs = {}
 607         if initial_mem is None:
 608             initial_mem = {}
 609         if fpregfile is None:
 610             fpregfile = [0] * 32
 611         if initial_insns is None:
 612             initial_insns = {}
 613             assert self.respect_pc == False, "instructions required to honor pc"
 614
 615         log("ISACaller insns", respect_pc, initial_insns, disassembly)
 616         log("ISACaller initial_msr", initial_msr)
 617
 618         # "fake program counter" mode (for unit testing)
 619         self.fake_pc = 0
 620         disasm_start = 0
 621         if not respect_pc:
 622             if isinstance(initial_mem, tuple):
 623                 self.fake_pc = initial_mem[0]
 624                 disasm_start = self.fake_pc
 625         else:
 626             disasm_start = initial_pc
 627
 628         # disassembly: we need this for now (not given from the decoder)
 629         self.disassembly = {}
 630         if disassembly:
 631             for i, code in enumerate(disassembly):
 632                 self.disassembly[i*4 + disasm_start] = code
 633
 634         # set up registers, instruction memory, data memory, PC, SPRs, MSR, CR
 635         self.svp64rm = SVP64RM()
 636         if initial_svstate is None:
 637             initial_svstate = 0
 638         if isinstance(initial_svstate, int):
 639             initial_svstate = SVP64State(initial_svstate)
 640         # SVSTATE, MSR and PC
 641         self.svstate = initial_svstate
 642         self.msr = SelectableInt(initial_msr, 64)  # underlying reg
 643         self.pc = PC()
 644         # GPR FPR SPR registers
 645         initial_sprs = deepcopy(initial_sprs)  # so as not to get modified
 646         self.gpr = GPR(decoder2, self, self.svstate, regfile)
 647         self.fpr = GPR(decoder2, self, self.svstate, fpregfile)
 648         self.spr = SPR(decoder2, initial_sprs)  # initialise SPRs before MMU
 649
 650         # set up 4 dummy SVSHAPEs if they aren't already set up
 651         for i in range(4):
 652             sname = 'SVSHAPE%d' % i
 653             if sname not in self.spr:
 654                 val = 0
 655             else:
 656                 val = self.spr[sname].value
 657             # make sure it's an SVSHAPE
 658             self.spr[sname] = SVSHAPE(val, self.gpr)
 659         self.last_op_svshape = False
 660
 661         # "raw" memory
 662         self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
 663         self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
 664         # MMU mode, redirect underlying Mem through RADIX
 665         if mmu:
 666             self.mem = RADIX(self.mem, self)
 667             if icachemmu:
 668                 self.imem = RADIX(self.imem, self)
 669
 670         # TODO, needed here:
 671         # FPR (same as GPR except for FP nums)
 672         # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
 673         #            note that mffs, mcrfs, mtfsf "manage" this FPSCR
 674         # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
 675         #         note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
 676         #         -- Done
 677         # 2.3.2 LR   (actually SPR #8) -- Done
 678         # 2.3.3 CTR  (actually SPR #9) -- Done
 679         # 2.3.4 TAR  (actually SPR #815)
 680         # 3.2.2 p45 XER  (actually SPR #1) -- Done
 681         # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
 682
 683         # create CR then allow portions of it to be "selectable" (below)
 684         self.cr_fields = CRFields(initial_cr)
 685         self.cr = self.cr_fields.cr
 686
 687         # "undefined", just set to variable-bit-width int (use exts "max")
 688         # self.undefined = SelectableInt(0, 256)  # TODO, not hard-code 256!
 689
 690         self.namespace = {}
 691         self.namespace.update(self.spr)
 692         self.namespace.update({'GPR': self.gpr,
 693                                'FPR': self.fpr,
 694                                'MEM': self.mem,
 695                                'SPR': self.spr,
 696                                'memassign': self.memassign,
 697                                'NIA': self.pc.NIA,
 698                                'CIA': self.pc.CIA,
 699                                'SVSTATE': self.svstate,
 700                                'SVSHAPE0': self.spr['SVSHAPE0'],
 701                                'SVSHAPE1': self.spr['SVSHAPE1'],
 702                                'SVSHAPE2': self.spr['SVSHAPE2'],
 703                                'SVSHAPE3': self.spr['SVSHAPE3'],
 704                                'CR': self.cr,
 705                                'MSR': self.msr,
 706                                'undefined': undefined,
 707                                'mode_is_64bit': True,
 708                                'SO': XER_bits['SO'],
 709                                'XLEN': 64  # elwidth overrides, later
 710                                })
 711
 712         # update pc to requested start point
 713         self.set_pc(initial_pc)
 714
 715         # field-selectable versions of Condition Register
 716         self.crl = self.cr_fields.crl
 717         for i in range(8):
 718             self.namespace["CR%d" % i] = self.crl[i]
 719
 720         self.decoder = decoder2.dec
 721         self.dec2 = decoder2
 722
 723         super().__init__(XLEN=self.namespace["XLEN"])
 724
 725     @property
 726     def XLEN(self):
 727         return self.namespace["XLEN"]
 728
 729     def call_trap(self, trap_addr, trap_bit):
 730         """calls TRAP and sets up NIA to the new execution location.
 731         next instruction will begin at trap_addr.
 732         """
 733         self.TRAP(trap_addr, trap_bit)
 734         self.namespace['NIA'] = self.trap_nia
 735         self.pc.update(self.namespace, self.is_svp64_mode)
 736
 737     def TRAP(self, trap_addr=0x700, trap_bit=PIb.TRAP):
 738         """TRAP> saves PC, MSR (and TODO SVSTATE), and updates MSR
 739
 740         TRAP function is callable from inside the pseudocode itself,
 741         hence the default arguments.  when calling from inside ISACaller
 742         it is best to use call_trap()
 743         """
 744         # https://bugs.libre-soc.org/show_bug.cgi?id=859
 745         kaivb = self.spr['KAIVB'].value
 746         msr = self.namespace['MSR'].value
 747         log("TRAP:", hex(trap_addr), hex(msr), "kaivb", hex(kaivb))
 748         # store CIA(+4?) in SRR0, set NIA to 0x700
 749         # store MSR in SRR1, set MSR to um errr something, have to check spec
 750         # store SVSTATE (if enabled) in SVSRR0
 751         self.spr['SRR0'].value = self.pc.CIA.value
 752         self.spr['SRR1'].value = msr
 753         if self.is_svp64_mode:
 754             self.spr['SVSRR0'] = self.namespace['SVSTATE'].value
 755         self.trap_nia = SelectableInt(trap_addr | (kaivb&~0x1fff), 64)
 756         self.spr['SRR1'][trap_bit] = 1  # change *copy* of MSR in SRR1
 757
 758         # set exception bits.  TODO: this should, based on the address
 759         # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
 760         # bits appropriately.  however it turns out that *for now* in all
 761         # cases (all trap_addrs) the exact same thing is needed.
 762         self.msr[MSRb.IR] = 0
 763         self.msr[MSRb.DR] = 0
 764         self.msr[MSRb.FE0] = 0
 765         self.msr[MSRb.FE1] = 0
 766         self.msr[MSRb.EE] = 0
 767         self.msr[MSRb.RI] = 0
 768         self.msr[MSRb.SF] = 1
 769         self.msr[MSRb.TM] = 0
 770         self.msr[MSRb.VEC] = 0
 771         self.msr[MSRb.VSX] = 0
 772         self.msr[MSRb.PR] = 0
 773         self.msr[MSRb.FP] = 0
 774         self.msr[MSRb.PMM] = 0
 775         self.msr[MSRb.TEs] = 0
 776         self.msr[MSRb.TEe] = 0
 777         self.msr[MSRb.UND] = 0
 778         self.msr[MSRb.LE] = 1
 779
 780     def memassign(self, ea, sz, val):
 781         self.mem.memassign(ea, sz, val)
 782
 783     def prep_namespace(self, insn_name, formname, op_fields):
 784         # TODO: get field names from form in decoder*1* (not decoder2)
 785         # decoder2 is hand-created, and decoder1.sigform is auto-generated
 786         # from spec
 787         # then "yield" fields only from op_fields rather than hard-coded
 788         # list, here.
 789         fields = self.decoder.sigforms[formname]
 790         log("prep_namespace", formname, op_fields)
 791         for name in op_fields:
 792             # CR immediates. deal with separately.  needs modifying
 793             # pseudocode
 794             if self.is_svp64_mode and name in ['BI']:  # TODO, more CRs
 795                 # BI is a 5-bit, must reconstruct the value
 796                 regnum, is_vec = yield from get_pdecode_cr_in(self.dec2, name)
 797                 sig = getattr(fields, name)
 798                 val = yield sig
 799                 # low 2 LSBs (CR field selector) remain same, CR num extended
 800                 assert regnum <= 7, "sigh, TODO, 128 CR fields"
 801                 val = (val & 0b11) | (regnum << 2)
 802             else:
 803                 if name == 'spr':
 804                     sig = getattr(fields, name.upper())
 805                 else:
 806                     sig = getattr(fields, name)
 807                 val = yield sig
 808             # these are all opcode fields involved in index-selection of CR,
 809             # and need to do "standard" arithmetic.  CR[BA+32] for example
 810             # would, if using SelectableInt, only be 5-bit.
 811             if name in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
 812                 self.namespace[name] = val
 813             else:
 814                 self.namespace[name] = SelectableInt(val, sig.width)
 815
 816         self.namespace['XER'] = self.spr['XER']
 817         self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
 818         self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
 819
 820         # add some SVSTATE convenience variables
 821         vl = self.svstate.vl
 822         srcstep = self.svstate.srcstep
 823         self.namespace['VL'] = vl
 824         self.namespace['srcstep'] = srcstep
 825
 826         # sv.bc* need some extra fields
 827         if self.is_svp64_mode and insn_name.startswith("sv.bc"):
 828             # blegh grab bits manually
 829             mode = yield self.dec2.rm_dec.rm_in.mode
 830             bc_vlset = (mode & SVP64MODE.BC_VLSET) != 0
 831             bc_vli = (mode & SVP64MODE.BC_VLI) != 0
 832             bc_snz = (mode & SVP64MODE.BC_SNZ) != 0
 833             bc_vsb = yield self.dec2.rm_dec.bc_vsb
 834             bc_lru = yield self.dec2.rm_dec.bc_lru
 835             bc_gate = yield self.dec2.rm_dec.bc_gate
 836             sz = yield self.dec2.rm_dec.pred_sz
 837             self.namespace['ALL'] = SelectableInt(bc_gate, 1)
 838             self.namespace['VSb'] = SelectableInt(bc_vsb, 1)
 839             self.namespace['LRu'] = SelectableInt(bc_lru, 1)
 840             self.namespace['VLSET'] = SelectableInt(bc_vlset, 1)
 841             self.namespace['VLI'] = SelectableInt(bc_vli, 1)
 842             self.namespace['sz'] = SelectableInt(sz, 1)
 843             self.namespace['SNZ'] = SelectableInt(bc_snz, 1)
 844
 845     def handle_carry_(self, inputs, outputs, already_done):
 846         inv_a = yield self.dec2.e.do.invert_in
 847         if inv_a:
 848             inputs[0] = ~inputs[0]
 849
 850         imm_ok = yield self.dec2.e.do.imm_data.ok
 851         if imm_ok:
 852             imm = yield self.dec2.e.do.imm_data.data
 853             inputs.append(SelectableInt(imm, 64))
 854         assert len(outputs) >= 1
 855         log("outputs", repr(outputs))
 856         if isinstance(outputs, list) or isinstance(outputs, tuple):
 857             output = outputs[0]
 858         else:
 859             output = outputs
 860         gts = []
 861         for x in inputs:
 862             log("gt input", x, output)
 863             gt = (gtu(x, output))
 864             gts.append(gt)
 865         log(gts)
 866         cy = 1 if any(gts) else 0
 867         log("CA", cy, gts)
 868         if not (1 & already_done):
 869             self.spr['XER'][XER_bits['CA']] = cy
 870
 871         log("inputs", already_done, inputs)
 872         # 32 bit carry
 873         # ARGH... different for OP_ADD... *sigh*...
 874         op = yield self.dec2.e.do.insn_type
 875         if op == MicrOp.OP_ADD.value:
 876             res32 = (output.value & (1 << 32)) != 0
 877             a32 = (inputs[0].value & (1 << 32)) != 0
 878             if len(inputs) >= 2:
 879                 b32 = (inputs[1].value & (1 << 32)) != 0
 880             else:
 881                 b32 = False
 882             cy32 = res32 ^ a32 ^ b32
 883             log("CA32 ADD", cy32)
 884         else:
 885             gts = []
 886             for x in inputs:
 887                 log("input", x, output)
 888                 log("     x[32:64]", x, x[32:64])
 889                 log("     o[32:64]", output, output[32:64])
 890                 gt = (gtu(x[32:64], output[32:64])) == SelectableInt(1, 1)
 891                 gts.append(gt)
 892             cy32 = 1 if any(gts) else 0
 893             log("CA32", cy32, gts)
 894         if not (2 & already_done):
 895             self.spr['XER'][XER_bits['CA32']] = cy32
 896
 897     def handle_overflow(self, inputs, outputs, div_overflow):
 898         if hasattr(self.dec2.e.do, "invert_in"):
 899             inv_a = yield self.dec2.e.do.invert_in
 900             if inv_a:
 901                 inputs[0] = ~inputs[0]
 902
 903         imm_ok = yield self.dec2.e.do.imm_data.ok
 904         if imm_ok:
 905             imm = yield self.dec2.e.do.imm_data.data
 906             inputs.append(SelectableInt(imm, 64))
 907         assert len(outputs) >= 1
 908         log("handle_overflow", inputs, outputs, div_overflow)
 909         if len(inputs) < 2 and div_overflow is None:
 910             return
 911
 912         # div overflow is different: it's returned by the pseudo-code
 913         # because it's more complex than can be done by analysing the output
 914         if div_overflow is not None:
 915             ov, ov32 = div_overflow, div_overflow
 916         # arithmetic overflow can be done by analysing the input and output
 917         elif len(inputs) >= 2:
 918             output = outputs[0]
 919
 920             # OV (64-bit)
 921             input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
 922             output_sgn = exts(output.value, output.bits) < 0
 923             ov = 1 if input_sgn[0] == input_sgn[1] and \
 924                 output_sgn != input_sgn[0] else 0
 925
 926             # OV (32-bit)
 927             input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
 928             output32_sgn = exts(output.value, 32) < 0
 929             ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
 930                 output32_sgn != input32_sgn[0] else 0
 931
 932         self.spr['XER'][XER_bits['OV']] = ov
 933         self.spr['XER'][XER_bits['OV32']] = ov32
 934         so = self.spr['XER'][XER_bits['SO']]
 935         so = so | ov
 936         self.spr['XER'][XER_bits['SO']] = so
 937
 938     def handle_comparison(self, outputs, cr_idx=0):
 939         out = outputs[0]
 940         assert isinstance(out, SelectableInt), \
 941             "out zero not a SelectableInt %s" % repr(outputs)
 942         log("handle_comparison", out.bits, hex(out.value))
 943         # TODO - XXX *processor* in 32-bit mode
 944         # https://bugs.libre-soc.org/show_bug.cgi?id=424
 945         # if is_32bit:
 946         #    o32 = exts(out.value, 32)
 947         #    print ("handle_comparison exts 32 bit", hex(o32))
 948         out = exts(out.value, out.bits)
 949         log("handle_comparison exts", hex(out))
 950         zero = SelectableInt(out == 0, 1)
 951         positive = SelectableInt(out > 0, 1)
 952         negative = SelectableInt(out < 0, 1)
 953         SO = self.spr['XER'][XER_bits['SO']]
 954         log("handle_comparison SO", SO)
 955         cr_field = selectconcat(negative, positive, zero, SO)
 956         log("handle_comparison cr_field", self.cr, cr_idx, cr_field)
 957         self.crl[cr_idx].eq(cr_field)
 958
 959     def set_pc(self, pc_val):
 960         self.namespace['NIA'] = SelectableInt(pc_val, 64)
 961         self.pc.update(self.namespace, self.is_svp64_mode)
 962
 963     def get_next_insn(self):
 964         """check instruction
 965         """
 966         if self.respect_pc:
 967             pc = self.pc.CIA.value
 968         else:
 969             pc = self.fake_pc
 970         ins = self.imem.ld(pc, 4, False, True, instr_fetch=True)
 971         if ins is None:
 972             raise KeyError("no instruction at 0x%x" % pc)
 973         return pc, ins
 974
 975     def setup_one(self):
 976         """set up one instruction
 977         """
 978         pc, insn = self.get_next_insn()
 979         yield from self.setup_next_insn(pc, insn)
 980
 981     def setup_next_insn(self, pc, ins):
 982         """set up next instruction
 983         """
 984         self._pc = pc
 985         log("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
 986         log("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
 987
 988         yield self.dec2.sv_rm.eq(0)
 989         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
 990         yield self.dec2.dec.bigendian.eq(self.bigendian)
 991         yield self.dec2.state.msr.eq(self.msr.value)
 992         yield self.dec2.state.pc.eq(pc)
 993         if self.svstate is not None:
 994             yield self.dec2.state.svstate.eq(self.svstate.value)
 995
 996         # SVP64.  first, check if the opcode is EXT001, and SVP64 id bits set
 997         yield Settle()
 998         opcode = yield self.dec2.dec.opcode_in
 999         pfx = SVP64PrefixFields()  # TODO should probably use SVP64PrefixDecoder
1000         pfx.insn.value = opcode
1001         major = pfx.major.asint(msb0=True)  # MSB0 inversion
1002         log("prefix test: opcode:", major, bin(major),
1003             pfx.insn[7] == 0b1, pfx.insn[9] == 0b1)
1004         self.is_svp64_mode = ((major == 0b000001) and
1005                               pfx.insn[7].value == 0b1 and
1006                               pfx.insn[9].value == 0b1)
1007         self.pc.update_nia(self.is_svp64_mode)
1008         # set SVP64 decode
1009         yield self.dec2.is_svp64_mode.eq(self.is_svp64_mode)
1010         self.namespace['NIA'] = self.pc.NIA
1011         self.namespace['SVSTATE'] = self.svstate
1012         if not self.is_svp64_mode:
1013             return
1014
1015         # in SVP64 mode.  decode/print out svp64 prefix, get v3.0B instruction
1016         log("svp64.rm", bin(pfx.rm.asint(msb0=True)))
1017         log("    svstate.vl", self.svstate.vl)
1018         log("    svstate.mvl", self.svstate.maxvl)
1019         sv_rm = pfx.rm.asint(msb0=True)
1020         ins = self.imem.ld(pc+4, 4, False, True, instr_fetch=True)
1021         log("     svsetup: 0x%x 0x%x %s" % (pc+4, ins & 0xffffffff, bin(ins)))
1022         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)  # v3.0B suffix
1023         yield self.dec2.sv_rm.eq(sv_rm)                        # svp64 prefix
1024         yield Settle()
1025
1026     def execute_one(self):
1027         """execute one instruction
1028         """
1029         # get the disassembly code for this instruction
1030         if self.is_svp64_mode:
1031             if not self.disassembly:
1032                 code = yield from self.get_assembly_name()
1033             else:
1034                 code = self.disassembly[self._pc+4]
1035             log("    svp64 sim-execute", hex(self._pc), code)
1036         else:
1037             if not self.disassembly:
1038                 code = yield from self.get_assembly_name()
1039             else:
1040                 code = self.disassembly[self._pc]
1041             log("sim-execute", hex(self._pc), code)
1042         opname = code.split(' ')[0]
1043         try:
1044             yield from self.call(opname)         # execute the instruction
1045         except MemException as e:                # check for memory errors
1046             if e.args[0] == 'unaligned':         # alignment error
1047                # run a Trap but set DAR first
1048                 print("memory unaligned exception, DAR", e.dar)
1049                 self.spr['DAR'] = SelectableInt(e.dar, 64)
1050                 self.call_trap(0x600, PIb.PRIV)    # 0x600, privileged
1051                 return
1052             elif e.args[0] == 'invalid':         # invalid
1053                # run a Trap but set DAR first
1054                 log("RADIX MMU memory invalid error, mode %s" % e.mode)
1055                 if e.mode == 'EXECUTE':
1056                     # XXX TODO: must set a few bits in SRR1,
1057                     # see microwatt loadstore1.vhdl
1058                     # if m_in.segerr = '0' then
1059                     #     v.srr1(47 - 33) := m_in.invalid;
1060                     #     v.srr1(47 - 35) := m_in.perm_error; -- noexec fault
1061                     #     v.srr1(47 - 44) := m_in.badtree;
1062                     #     v.srr1(47 - 45) := m_in.rc_error;
1063                     #     v.intr_vec := 16#400#;
1064                     # else
1065                     #     v.intr_vec := 16#480#;
1066                     self.call_trap(0x400, PIb.PRIV)    # 0x400, privileged
1067                 else:
1068                     self.call_trap(0x300, PIb.PRIV)    # 0x300, privileged
1069                 return
1070             # not supported yet:
1071             raise e                          # ... re-raise
1072
1073         # don't use this except in special circumstances
1074         if not self.respect_pc:
1075             self.fake_pc += 4
1076
1077         log("execute one, CIA NIA", hex(self.pc.CIA.value),
1078             hex(self.pc.NIA.value))
1079
1080     def get_assembly_name(self):
1081         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
1082         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
1083         dec_insn = yield self.dec2.e.do.insn
1084         insn_1_11 = yield self.dec2.e.do.insn[1:11]
1085         asmcode = yield self.dec2.dec.op.asmcode
1086         int_op = yield self.dec2.dec.op.internal_op
1087         log("get assembly name asmcode", asmcode, int_op,
1088             hex(dec_insn), bin(insn_1_11))
1089         asmop = insns.get(asmcode, None)
1090
1091         # sigh reconstruct the assembly instruction name
1092         if hasattr(self.dec2.e.do, "oe"):
1093             ov_en = yield self.dec2.e.do.oe.oe
1094             ov_ok = yield self.dec2.e.do.oe.ok
1095         else:
1096             ov_en = False
1097             ov_ok = False
1098         if hasattr(self.dec2.e.do, "rc"):
1099             rc_en = yield self.dec2.e.do.rc.rc
1100             rc_ok = yield self.dec2.e.do.rc.ok
1101         else:
1102             rc_en = False
1103             rc_ok = False
1104         # grrrr have to special-case MUL op (see DecodeOE)
1105         log("ov %d en %d rc %d en %d op %d" %
1106             (ov_ok, ov_en, rc_ok, rc_en, int_op))
1107         if int_op in [MicrOp.OP_MUL_H64.value, MicrOp.OP_MUL_H32.value]:
1108             log("mul op")
1109             if rc_en & rc_ok:
1110                 asmop += "."
1111         else:
1112             if not asmop.endswith("."):  # don't add "." to "andis."
1113                 if rc_en & rc_ok:
1114                     asmop += "."
1115         if hasattr(self.dec2.e.do, "lk"):
1116             lk = yield self.dec2.e.do.lk
1117             if lk:
1118                 asmop += "l"
1119         log("int_op", int_op)
1120         if int_op in [MicrOp.OP_B.value, MicrOp.OP_BC.value]:
1121             AA = yield self.dec2.dec.fields.FormI.AA[0:-1]
1122             log("AA", AA)
1123             if AA:
1124                 asmop += "a"
1125         spr_msb = yield from self.get_spr_msb()
1126         if int_op == MicrOp.OP_MFCR.value:
1127             if spr_msb:
1128                 asmop = 'mfocrf'
1129             else:
1130                 asmop = 'mfcr'
1131         # XXX TODO: for whatever weird reason this doesn't work
1132         # https://bugs.libre-soc.org/show_bug.cgi?id=390
1133         if int_op == MicrOp.OP_MTCRF.value:
1134             if spr_msb:
1135                 asmop = 'mtocrf'
1136             else:
1137                 asmop = 'mtcrf'
1138         return asmop
1139
1140     def get_remap_indices(self):
1141         """WARNING, this function stores remap_idxs and remap_loopends
1142         in the class for later use.  this to avoid problems with yield
1143         """
1144         # go through all iterators in lock-step, advance to next remap_idx
1145         srcstep, dststep, ssubstep, dsubstep = self.get_src_dststeps()
1146         # get four SVSHAPEs. here we are hard-coding
1147         SVSHAPE0 = self.spr['SVSHAPE0']
1148         SVSHAPE1 = self.spr['SVSHAPE1']
1149         SVSHAPE2 = self.spr['SVSHAPE2']
1150         SVSHAPE3 = self.spr['SVSHAPE3']
1151         # set up the iterators
1152         remaps = [(SVSHAPE0, SVSHAPE0.get_iterator()),
1153                   (SVSHAPE1, SVSHAPE1.get_iterator()),
1154                   (SVSHAPE2, SVSHAPE2.get_iterator()),
1155                   (SVSHAPE3, SVSHAPE3.get_iterator()),
1156                   ]
1157
1158         self.remap_loopends = [0] * 4
1159         self.remap_idxs = [0, 1, 2, 3]
1160         dbg = []
1161         for i, (shape, remap) in enumerate(remaps):
1162             # zero is "disabled"
1163             if shape.value == 0x0:
1164                 self.remap_idxs[i] = 0
1165             # pick src or dststep depending on reg num (0-2=in, 3-4=out)
1166             step = dststep if (i in [3, 4]) else srcstep
1167             # this is terrible.  O(N^2) looking for the match. but hey.
1168             for idx, (remap_idx, loopends) in enumerate(remap):
1169                 if idx == step:
1170                     break
1171             self.remap_idxs[i] = remap_idx
1172             self.remap_loopends[i] = loopends
1173             dbg.append((i, step, remap_idx, loopends))
1174         for (i, step, remap_idx, loopends) in dbg:
1175             log("SVSHAPE %d idx, end" % i, step, remap_idx, bin(loopends))
1176         return remaps
1177
1178     def get_spr_msb(self):
1179         dec_insn = yield self.dec2.e.do.insn
1180         return dec_insn & (1 << 20) != 0  # sigh - XFF.spr[-1]?
1181
1182     def call(self, name):
1183         """call(opcode) - the primary execution point for instructions
1184         """
1185         self.last_st_addr = None  # reset the last known store address
1186         self.last_ld_addr = None  # etc.
1187
1188         ins_name = name.strip()  # remove spaces if not already done so
1189         if self.halted:
1190             log("halted - not executing", ins_name)
1191             return
1192
1193         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
1194         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
1195         asmop = yield from self.get_assembly_name()
1196         log("call", ins_name, asmop)
1197
1198         # check privileged
1199         int_op = yield self.dec2.dec.op.internal_op
1200         spr_msb = yield from self.get_spr_msb()
1201
1202         instr_is_privileged = False
1203         if int_op in [MicrOp.OP_ATTN.value,
1204                       MicrOp.OP_MFMSR.value,
1205                       MicrOp.OP_MTMSR.value,
1206                       MicrOp.OP_MTMSRD.value,
1207                       # TODO: OP_TLBIE
1208                       MicrOp.OP_RFID.value]:
1209             instr_is_privileged = True
1210         if int_op in [MicrOp.OP_MFSPR.value,
1211                       MicrOp.OP_MTSPR.value] and spr_msb:
1212             instr_is_privileged = True
1213
1214         log("is priv", instr_is_privileged, hex(self.msr.value),
1215             self.msr[MSRb.PR])
1216         # check MSR priv bit and whether op is privileged: if so, throw trap
1217         if instr_is_privileged and self.msr[MSRb.PR] == 1:
1218             self.call_trap(0x700, PIb.PRIV)
1219             return
1220
1221         # check halted condition
1222         if ins_name == 'attn':
1223             self.halted = True
1224             return
1225
1226         # check illegal instruction
1227         illegal = False
1228         if ins_name not in ['mtcrf', 'mtocrf']:
1229             illegal = ins_name != asmop
1230
1231         # list of instructions not being supported by binutils (.long)
1232         dotstrp = asmop[:-1] if asmop[-1] == '.' else asmop
1233         if dotstrp in [ 'fsins', 'fcoss',
1234                     'ffmadds', 'fdmadds', 'ffadds',
1235                      'mins', 'maxs', 'minu', 'maxu',
1236                     'setvl', 'svindex', 'svremap', 'svstep', 'svshape',
1237                     'grev', 'ternlogi', 'bmask', 'cprop',
1238                     'absdu', 'absds', 'absdacs', 'absdacu', 'avgadd',
1239                     ]:
1240             illegal = False
1241             ins_name = dotstrp
1242
1243         # branch-conditional redirects to sv.bc
1244         if asmop.startswith('bc') and self.is_svp64_mode:
1245             ins_name = 'sv.%s' % ins_name
1246
1247         log("   post-processed name", dotstrp, ins_name, asmop)
1248
1249         # illegal instructions call TRAP at 0x700
1250         if illegal:
1251             print("illegal", ins_name, asmop)
1252             self.call_trap(0x700, PIb.ILLEG)
1253             print("name %s != %s - calling ILLEGAL trap, PC: %x" %
1254                   (ins_name, asmop, self.pc.CIA.value))
1255             return
1256
1257         # this is for setvl "Vertical" mode: if set true,
1258         # srcstep/dststep is explicitly advanced. mode says which SVSTATE to
1259         # test for Rc=1 end condition.  3 bits of all 3 loops are put into CR0
1260         self.allow_next_step_inc = False
1261         self.svstate_next_mode = 0
1262
1263         # nop has to be supported, we could let the actual op calculate
1264         # but PowerDecoder has a pattern for nop
1265         if ins_name is 'nop':
1266             self.update_pc_next()
1267             return
1268
1269         # look up instruction in ISA.instrs, prepare namespace
1270         info = self.instrs[ins_name]
1271         yield from self.prep_namespace(ins_name, info.form, info.op_fields)
1272
1273         # preserve order of register names
1274         input_names = create_args(list(info.read_regs) +
1275                                   list(info.uninit_regs))
1276         log("input names", input_names)
1277
1278         # get SVP64 entry for the current instruction
1279         sv_rm = self.svp64rm.instrs.get(ins_name)
1280         if sv_rm is not None:
1281             dest_cr, src_cr, src_byname, dest_byname = decode_extra(sv_rm)
1282         else:
1283             dest_cr, src_cr, src_byname, dest_byname = False, False, {}, {}
1284         log("sv rm", sv_rm, dest_cr, src_cr, src_byname, dest_byname)
1285
1286         # see if srcstep/dststep need skipping over masked-out predicate bits
1287         if (self.is_svp64_mode or ins_name == 'setvl' or
1288                 ins_name in ['svremap', 'svstate']):
1289             yield from self.svstate_pre_inc()
1290         if self.is_svp64_mode:
1291             pre = yield from self.update_new_svstate_steps()
1292             if pre:
1293                 self.svp64_reset_loop()
1294                 self.update_nia()
1295                 self.update_pc_next()
1296                 return
1297             srcstep, dststep, ssubstep, dsubstep = self.get_src_dststeps()
1298             pred_dst_zero = self.pred_dst_zero
1299             pred_src_zero = self.pred_src_zero
1300             vl = self.svstate.vl
1301             subvl = yield self.dec2.rm_dec.rm_in.subvl
1302
1303         # VL=0 in SVP64 mode means "do nothing: skip instruction"
1304         if self.is_svp64_mode and vl == 0:
1305             self.pc.update(self.namespace, self.is_svp64_mode)
1306             log("SVP64: VL=0, end of call", self.namespace['CIA'],
1307                 self.namespace['NIA'])
1308             return
1309
1310         # for when SVREMAP is active, using pre-arranged schedule.
1311         # note: modifying PowerDecoder2 needs to "settle"
1312         remap_en = self.svstate.SVme
1313         persist = self.svstate.RMpst
1314         active = (persist or self.last_op_svshape) and remap_en != 0
1315         if self.is_svp64_mode:
1316             yield self.dec2.remap_active.eq(remap_en if active else 0)
1317         yield Settle()
1318         if persist or self.last_op_svshape:
1319             remaps = self.get_remap_indices()
1320         if self.is_svp64_mode and (persist or self.last_op_svshape):
1321             yield from self.remap_debug(remaps)
1322         # after that, settle down (combinatorial) to let Vector reg numbers
1323         # work themselves out
1324         yield Settle()
1325         if self.is_svp64_mode:
1326             remap_active = yield self.dec2.remap_active
1327         else:
1328             remap_active = False
1329         log("remap active", bin(remap_active))
1330
1331         # main input registers (RT, RA ...)
1332         inputs = []
1333         for name in input_names:
1334             regval = (yield from self.get_input(name))
1335             inputs.append(regval)
1336
1337         # arrrrgh, awful hack, to get _RT into namespace
1338         if ins_name in ['setvl', 'svstep']:
1339             regname = "_RT"
1340             RT = yield self.dec2.dec.RT
1341             self.namespace[regname] = SelectableInt(RT, 5)
1342             if RT == 0:
1343                 self.namespace["RT"] = SelectableInt(0, 5)
1344             regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, "RT")
1345             log('hack input reg %s %s' % (name, str(regnum)), is_vec)
1346
1347         # in SVP64 mode for LD/ST work out immediate
1348         # XXX TODO: replace_ds for DS-Form rather than D-Form.
1349         # use info.form to detect
1350         if self.is_svp64_mode:
1351             yield from self.check_replace_d(info, remap_active)
1352
1353         # "special" registers
1354         for special in info.special_regs:
1355             if special in special_sprs:
1356                 inputs.append(self.spr[special])
1357             else:
1358                 inputs.append(self.namespace[special])
1359
1360         # clear trap (trap) NIA
1361         self.trap_nia = None
1362
1363         # check if this was an sv.bc* and create an indicator that
1364         # this is the last check to be made as a loop.  combined with
1365         # the ALL/ANY mode we can early-exit
1366         if self.is_svp64_mode and ins_name.startswith("sv.bc"):
1367             no_in_vec = yield self.dec2.no_in_vec  # BI is scalar
1368             end_loop = no_in_vec or srcstep == vl-1 or dststep == vl-1
1369             self.namespace['end_loop'] = SelectableInt(end_loop, 1)
1370
1371         # execute actual instruction here (finally)
1372         log("inputs", inputs)
1373         results = info.func(self, *inputs)
1374         log("results", results)
1375
1376         # "inject" decorator takes namespace from function locals: we need to
1377         # overwrite NIA being overwritten (sigh)
1378         if self.trap_nia is not None:
1379             self.namespace['NIA'] = self.trap_nia
1380
1381         log("after func", self.namespace['CIA'], self.namespace['NIA'])
1382
1383         # check if op was a LD/ST so that debugging can check the
1384         # address
1385         if int_op in [MicrOp.OP_STORE.value,
1386                       ]:
1387             self.last_st_addr = self.mem.last_st_addr
1388         if int_op in [MicrOp.OP_LOAD.value,
1389                       ]:
1390             self.last_ld_addr = self.mem.last_ld_addr
1391         log("op", int_op, MicrOp.OP_STORE.value, MicrOp.OP_LOAD.value,
1392             self.last_st_addr, self.last_ld_addr)
1393
1394         # detect if CA/CA32 already in outputs (sra*, basically)
1395         already_done = 0
1396         if info.write_regs:
1397             output_names = create_args(info.write_regs)
1398             for name in output_names:
1399                 if name == 'CA':
1400                     already_done |= 1
1401                 if name == 'CA32':
1402                     already_done |= 2
1403
1404         log("carry already done?", bin(already_done))
1405         if hasattr(self.dec2.e.do, "output_carry"):
1406             carry_en = yield self.dec2.e.do.output_carry
1407         else:
1408             carry_en = False
1409         if carry_en:
1410             yield from self.handle_carry_(inputs, results, already_done)
1411
1412         if not self.is_svp64_mode:  # yeah just no. not in parallel processing
1413             # detect if overflow was in return result
1414             overflow = None
1415             if info.write_regs:
1416                 for name, output in zip(output_names, results):
1417                     if name == 'overflow':
1418                         overflow = output
1419
1420             if hasattr(self.dec2.e.do, "oe"):
1421                 ov_en = yield self.dec2.e.do.oe.oe
1422                 ov_ok = yield self.dec2.e.do.oe.ok
1423             else:
1424                 ov_en = False
1425                 ov_ok = False
1426             log("internal overflow", overflow, ov_en, ov_ok)
1427             if ov_en & ov_ok:
1428                 yield from self.handle_overflow(inputs, results, overflow)
1429
1430         # only do SVP64 dest predicated Rc=1 if dest-pred is not enabled
1431         rc_en = False
1432         if not self.is_svp64_mode or not pred_dst_zero:
1433             if hasattr(self.dec2.e.do, "rc"):
1434                 rc_en = yield self.dec2.e.do.rc.rc
1435         if rc_en and ins_name not in ['svstep']:
1436             regnum, is_vec = yield from get_pdecode_cr_out(self.dec2, "CR0")
1437             self.handle_comparison(results, regnum)
1438
1439         # any modified return results?
1440         if info.write_regs:
1441             for name, output in zip(output_names, results):
1442                 yield from self.check_write(info, name, output)
1443
1444         nia_update = (yield from self.check_step_increment(results, rc_en,
1445                                                            asmop, ins_name))
1446         if nia_update:
1447             self.update_pc_next()
1448
1449     def check_replace_d(self, info, remap_active):
1450         replace_d = False  # update / replace constant in pseudocode
1451         ldstmode = yield self.dec2.rm_dec.ldstmode
1452         # shift mode reads SVD (or SVDS - TODO)
1453         # *BUT*... because this is "overloading" of LD operations,
1454         # it gets *STORED* into D (or DS, TODO)
1455         if ldstmode == SVP64LDSTmode.SHIFT.value:
1456             imm = yield self.dec2.dec.fields.FormSVD.SVD[0:11]
1457             imm = exts(imm, 11)  # sign-extend to integer
1458             log("shift SVD", imm)
1459             replace_d = True
1460         else:
1461             if info.form == 'DS':
1462                 # DS-Form, multiply by 4 then knock 2 bits off after
1463                 imm = yield self.dec2.dec.fields.FormDS.DS[0:14] * 4
1464             else:
1465                 imm = yield self.dec2.dec.fields.FormD.D[0:16]
1466             imm = exts(imm, 16)  # sign-extend to integer
1467         # get the right step. LD is from srcstep, ST is dststep
1468         op = yield self.dec2.e.do.insn_type
1469         offsmul = 0
1470         if op == MicrOp.OP_LOAD.value:
1471             if remap_active:
1472                 offsmul = yield self.dec2.in1_step
1473                 log("D-field REMAP src", imm, offsmul)
1474             else:
1475                 offsmul = (srcstep * (subvl+1)) + ssubstep
1476                 log("D-field src", imm, offsmul)
1477         elif op == MicrOp.OP_STORE.value:
1478             # XXX NOTE! no bit-reversed STORE! this should not ever be used
1479             offsmul = (dststep * (subvl+1)) + dsubstep
1480             log("D-field dst", imm, offsmul)
1481         # bit-reverse mode, rev already done through get_src_dst_steps()
1482         if ldstmode == SVP64LDSTmode.SHIFT.value:
1483             # manually look up RC, sigh
1484             RC = yield self.dec2.dec.RC[0:5]
1485             RC = self.gpr(RC)
1486             log("LD-SHIFT:", "VL", vl,
1487                 "RC", RC.value, "imm", imm,
1488                 "offs", bin(offsmul),
1489                 )
1490             imm = SelectableInt((imm * offsmul) << RC.value, 32)
1491         # Unit-Strided LD/ST adds offset*width to immediate
1492         elif ldstmode == SVP64LDSTmode.UNITSTRIDE.value:
1493             ldst_len = yield self.dec2.e.do.data_len
1494             imm = SelectableInt(imm + offsmul * ldst_len, 32)
1495             replace_d = True
1496         # Element-strided multiplies the immediate by element step
1497         elif ldstmode == SVP64LDSTmode.ELSTRIDE.value:
1498             imm = SelectableInt(imm * offsmul, 32)
1499             replace_d = True
1500         if replace_d:
1501             ldst_ra_vec = yield self.dec2.rm_dec.ldst_ra_vec
1502             ldst_imz_in = yield self.dec2.rm_dec.ldst_imz_in
1503             log("LDSTmode", SVP64LDSTmode(ldstmode),
1504                 offsmul, imm, ldst_ra_vec, ldst_imz_in)
1505         # new replacement D... errr.. DS
1506         if replace_d:
1507             if info.form == 'DS':
1508                 # TODO: assert 2 LSBs are zero?
1509                 log("DS-Form, TODO, assert 2 LSBs zero?", bin(imm.value))
1510                 imm.value = imm.value >> 2
1511                 self.namespace['DS'] = imm
1512             else:
1513                 self.namespace['D'] = imm
1514
1515     def get_input(self, name):
1516         # using PowerDecoder2, first, find the decoder index.
1517         # (mapping name RA RB RC RS to in1, in2, in3)
1518         regnum, is_vec = yield from get_pdecode_idx_in(self.dec2, name)
1519         if regnum is None:
1520             # doing this is not part of svp64, it's because output
1521             # registers, to be modified, need to be in the namespace.
1522             regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, name)
1523         if regnum is None:
1524             regnum, is_vec = yield from get_pdecode_idx_out2(self.dec2, name)
1525
1526         # in case getting the register number is needed, _RA, _RB
1527         regname = "_" + name
1528         self.namespace[regname] = regnum
1529         if not self.is_svp64_mode or not self.pred_src_zero:
1530             log('reading reg %s %s' % (name, str(regnum)), is_vec)
1531             if name in fregs:
1532                 reg_val = SelectableInt(self.fpr(regnum))
1533             elif name is not None:
1534                 reg_val = SelectableInt(self.gpr(regnum))
1535         else:
1536             log('zero input reg %s %s' % (name, str(regnum)), is_vec)
1537             reg_val = 0
1538         return reg_val
1539
1540     def remap_debug(self, remaps):
1541         # just some convenient debug info
1542         for i in range(4):
1543             sname = 'SVSHAPE%d' % i
1544             shape = self.spr[sname]
1545             log(sname, bin(shape.value))
1546             log("    lims", shape.lims)
1547             log("    mode", shape.mode)
1548             log("    skip", shape.skip)
1549
1550         # set up the list of steps to remap
1551         mi0 = self.svstate.mi0
1552         mi1 = self.svstate.mi1
1553         mi2 = self.svstate.mi2
1554         mo0 = self.svstate.mo0
1555         mo1 = self.svstate.mo1
1556         steps = [(self.dec2.in1_step, mi0),  # RA
1557                  (self.dec2.in2_step, mi1),  # RB
1558                  (self.dec2.in3_step, mi2),  # RC
1559                  (self.dec2.o_step, mo0),   # RT
1560                  (self.dec2.o2_step, mo1),   # EA
1561                  ]
1562         remap_idxs = self.remap_idxs
1563         rremaps = []
1564         # now cross-index the required SHAPE for each of 3-in 2-out regs
1565         rnames = ['RA', 'RB', 'RC', 'RT', 'EA']
1566         for i, (dstep, shape_idx) in enumerate(steps):
1567             (shape, remap) = remaps[shape_idx]
1568             remap_idx = remap_idxs[shape_idx]
1569             # zero is "disabled"
1570             if shape.value == 0x0:
1571                 continue
1572             # now set the actual requested step to the current index
1573             yield dstep.eq(remap_idx)
1574
1575             # debug printout info
1576             rremaps.append((shape.mode, i, rnames[i], shape_idx, remap_idx))
1577         for x in rremaps:
1578             log("shape remap", x)
1579
1580     def check_write(self, info, name, output):
1581         if name == 'overflow':  # ignore, done already (above)
1582             return
1583         if isinstance(output, int):
1584             output = SelectableInt(output, 256)
1585         if name in ['CA', 'CA32']:
1586             if carry_en:
1587                 log("writing %s to XER" % name, output)
1588                 self.spr['XER'][XER_bits[name]] = output.value
1589             else:
1590                 log("NOT writing %s to XER" % name, output)
1591         elif name in info.special_regs:
1592             log('writing special %s' % name, output, special_sprs)
1593             if name in special_sprs:
1594                 self.spr[name] = output
1595             else:
1596                 self.namespace[name].eq(output)
1597             if name == 'MSR':
1598                 log('msr written', hex(self.msr.value))
1599         else:
1600             regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, name)
1601             if regnum is None:
1602                 regnum, is_vec = yield from get_pdecode_idx_out2(
1603                     self.dec2, name)
1604             if regnum is None:
1605                 # temporary hack for not having 2nd output
1606                 regnum = yield getattr(self.decoder, name)
1607                 is_vec = False
1608             if self.is_svp64_mode and self.pred_dst_zero:
1609                 log('zeroing reg %d %s' % (regnum, str(output)),
1610                     is_vec)
1611                 output = SelectableInt(0, 256)
1612             else:
1613                 if name in fregs:
1614                     ftype = 'fpr'
1615                 else:
1616                     ftype = 'gpr'
1617                 log('writing %s %s %s' % (ftype, regnum, str(output)),
1618                     is_vec)
1619             if output.bits > 64:
1620                 output = SelectableInt(output.value, 64)
1621             if name in fregs:
1622                 self.fpr[regnum] = output
1623             else:
1624                 self.gpr[regnum] = output
1625
1626     def check_step_increment(self, results, rc_en, asmop, ins_name):
1627         # check if it is the SVSTATE.src/dest step that needs incrementing
1628         # this is our Sub-Program-Counter loop from 0 to VL-1
1629         pre = False
1630         post = False
1631         if self.allow_next_step_inc:
1632             log("SVSTATE_NEXT: inc requested, mode",
1633                 self.svstate_next_mode, self.allow_next_step_inc)
1634             yield from self.svstate_pre_inc()
1635             pre = yield from self.update_new_svstate_steps()
1636             if pre:
1637                 # reset at end of loop including exit Vertical Mode
1638                 log("SVSTATE_NEXT: end of loop, reset")
1639                 self.svp64_reset_loop()
1640                 self.svstate.vfirst = 0
1641                 self.update_nia()
1642                 if not rc_en:
1643                     return True
1644                 results = [SelectableInt(0, 64)]
1645                 self.handle_comparison(results)  # CR0
1646                 return True
1647             if self.allow_next_step_inc == 2:
1648                 log("SVSTATE_NEXT: read")
1649                 nia_update = (yield from self.svstate_post_inc(ins_name))
1650             else:
1651                 log("SVSTATE_NEXT: post-inc")
1652             # use actual src/dst-step here to check end, do NOT
1653             # use bit-reversed version
1654             srcstep, dststep = self.new_srcstep, self.new_dststep
1655             ssubstep, dsubstep = self.new_ssubstep, self.new_dsubstep
1656             remaps = self.get_remap_indices()
1657             remap_idxs = self.remap_idxs
1658             vl = self.svstate.vl
1659             subvl = yield self.dec2.rm_dec.rm_in.subvl
1660             end_src = srcstep == vl-1
1661             end_dst = dststep == vl-1
1662             if self.allow_next_step_inc != 2:
1663                 yield from self.advance_svstate_steps(end_src, end_dst)
1664             self.namespace['SVSTATE'] = self.svstate.spr
1665             # set CR0 (if Rc=1) based on end
1666             if rc_en:
1667                 endtest = 1 if (end_src or end_dst) else 0
1668                 #results = [SelectableInt(endtest, 64)]
1669                 # self.handle_comparison(results) # CR0
1670
1671                 # see if svstep was requested, if so, which SVSTATE
1672                 endings = 0b111
1673                 if self.svstate_next_mode > 0:
1674                     shape_idx = self.svstate_next_mode.value-1
1675                     endings = self.remap_loopends[shape_idx]
1676                 cr_field = SelectableInt((~endings) << 1 | endtest, 4)
1677                 print("svstep Rc=1, CR0", cr_field)
1678                 self.crl[0].eq(cr_field)  # CR0
1679             if end_src or end_dst:
1680                 # reset at end of loop including exit Vertical Mode
1681                 log("SVSTATE_NEXT: after increments, reset")
1682                 self.svp64_reset_loop()
1683                 self.svstate.vfirst = 0
1684             return nia_update
1685
1686         if self.is_svp64_mode:
1687             return (yield from self.svstate_post_inc(ins_name))
1688
1689         # XXX only in non-SVP64 mode!
1690         # record state of whether the current operation was an svshape,
1691         # OR svindex!
1692         # to be able to know if it should apply in the next instruction.
1693         # also (if going to use this instruction) should disable ability
1694         # to interrupt in between. sigh.
1695         self.last_op_svshape = asmop in ['svremap', 'svindex']
1696
1697         return True
1698
1699     def SVSTATE_NEXT(self, mode, submode):
1700         """explicitly moves srcstep/dststep on to next element, for
1701         "Vertical-First" mode.  this function is called from
1702         setvl pseudo-code, as a pseudo-op "svstep"
1703
1704         WARNING: this function uses information that was created EARLIER
1705         due to it being in the middle of a yield, but this function is
1706         *NOT* called from yield (it's called from compiled pseudocode).
1707         """
1708         self.allow_next_step_inc = submode.value + 1
1709         log("SVSTATE_NEXT mode", mode, submode, self.allow_next_step_inc)
1710         self.svstate_next_mode = mode
1711         if self.svstate_next_mode > 0 and self.svstate_next_mode < 5:
1712             shape_idx = self.svstate_next_mode.value-1
1713             return SelectableInt(self.remap_idxs[shape_idx], 7)
1714         if self.svstate_next_mode == 5:
1715             self.svstate_next_mode = 0
1716             return SelectableInt(self.svstate.srcstep, 7)
1717         if self.svstate_next_mode == 6:
1718             self.svstate_next_mode = 0
1719             return SelectableInt(self.svstate.dststep, 7)
1720         return SelectableInt(0, 7)
1721
1722     def svstate_pre_inc(self):
1723         """check if srcstep/dststep need to skip over masked-out predicate bits
1724         note that this is not supposed to do anything to substep,
1725         it is purely for skipping masked-out bits
1726         """
1727         # get SVSTATE VL (oh and print out some debug stuff)
1728         vl = self.svstate.vl
1729         subvl = yield self.dec2.rm_dec.rm_in.subvl
1730         srcstep = self.svstate.srcstep
1731         dststep = self.svstate.dststep
1732         ssubstep = self.svstate.ssubstep
1733         dsubstep = self.svstate.dsubstep
1734         sv_a_nz = yield self.dec2.sv_a_nz
1735         fft_mode = yield self.dec2.use_svp64_fft
1736         in1 = yield self.dec2.e.read_reg1.data
1737         log("SVP64: VL, subvl, srcstep, dststep, ssubstep, dsybstep, sv_a_nz, "
1738             "in1 fft, svp64",
1739             vl, subvl, srcstep, dststep, ssubstep, dsubstep,
1740             sv_a_nz, in1, fft_mode,
1741             self.is_svp64_mode)
1742
1743         # get predicate mask (all 64 bits)
1744         srcmask = dstmask = 0xffff_ffff_ffff_ffff
1745
1746         pmode = yield self.dec2.rm_dec.predmode
1747         reverse_gear = yield self.dec2.rm_dec.reverse_gear
1748         sv_ptype = yield self.dec2.dec.op.SV_Ptype
1749         srcpred = yield self.dec2.rm_dec.srcpred
1750         dstpred = yield self.dec2.rm_dec.dstpred
1751         pred_src_zero = yield self.dec2.rm_dec.pred_sz
1752         pred_dst_zero = yield self.dec2.rm_dec.pred_dz
1753         if pmode == SVP64PredMode.INT.value:
1754             srcmask = dstmask = get_predint(self.gpr, dstpred)
1755             if sv_ptype == SVPtype.P2.value:
1756                 srcmask = get_predint(self.gpr, srcpred)
1757         elif pmode == SVP64PredMode.CR.value:
1758             srcmask = dstmask = get_predcr(self.crl, dstpred, vl)
1759             if sv_ptype == SVPtype.P2.value:
1760                 srcmask = get_predcr(self.crl, srcpred, vl)
1761         # work out if the ssubsteps are completed
1762         ssubstart = ssubstep == 0
1763         dsubstart = dsubstep == 0
1764         log("    pmode", pmode)
1765         log("    reverse", reverse_gear)
1766         log("    ptype", sv_ptype)
1767         log("    srcpred", bin(srcpred))
1768         log("    dstpred", bin(dstpred))
1769         log("    srcmask", bin(srcmask))
1770         log("    dstmask", bin(dstmask))
1771         log("    pred_sz", bin(pred_src_zero))
1772         log("    pred_dz", bin(pred_dst_zero))
1773         log("    ssubstart", ssubstart)
1774         log("    dsubstart", dsubstart)
1775
1776         # okaaay, so here we simply advance srcstep (TODO dststep)
1777         # this can ONLY be done at the beginning of the "for" loop
1778         # (this is all actually a FSM so it's hell to keep track sigh)
1779         if ssubstart:
1780             # until the predicate mask has a "1" bit... or we run out of VL
1781             # let srcstep==VL be the indicator to move to next instruction
1782             if not pred_src_zero:
1783                 while (((1 << srcstep) & srcmask) == 0) and (srcstep != vl):
1784                     log("      sskip", bin(1 << srcstep))
1785                     srcstep += 1
1786         if dsubstart:
1787             # same for dststep
1788             if not pred_dst_zero:
1789                 while (((1 << dststep) & dstmask) == 0) and (dststep != vl):
1790                     log("      dskip", bin(1 << dststep))
1791                     dststep += 1
1792
1793         # now work out if the relevant mask bits require zeroing
1794         if pred_dst_zero:
1795             pred_dst_zero = ((1 << dststep) & dstmask) == 0
1796         if pred_src_zero:
1797             pred_src_zero = ((1 << srcstep) & srcmask) == 0
1798
1799         # store new srcstep / dststep
1800         self.new_srcstep, self.new_dststep = (srcstep, dststep)
1801         self.new_ssubstep, self.new_dsubstep = (ssubstep, dsubstep)
1802         self.pred_dst_zero, self.pred_src_zero = (pred_dst_zero, pred_src_zero)
1803         log("    new srcstep", srcstep)
1804         log("    new dststep", dststep)
1805         log("    new ssubstep", ssubstep)
1806         log("    new dsubstep", dsubstep)
1807
1808     def get_src_dststeps(self):
1809         """gets srcstep, dststep, and ssubstep, dsubstep
1810         """
1811         return (self.new_srcstep, self.new_dststep,
1812                 self.new_ssubstep, self.new_dsubstep)
1813
1814     def update_new_svstate_steps(self):
1815         # note, do not get the bit-reversed srcstep here!
1816         srcstep, dststep = self.new_srcstep, self.new_dststep
1817         ssubstep, dsubstep = self.new_ssubstep, self.new_dsubstep
1818
1819         # update SVSTATE with new srcstep
1820         self.svstate.srcstep = srcstep
1821         self.svstate.dststep = dststep
1822         self.svstate.ssubstep = ssubstep
1823         self.svstate.dsubstep = dsubstep
1824         self.namespace['SVSTATE'] = self.svstate
1825         yield self.dec2.state.svstate.eq(self.svstate.value)
1826         yield Settle()  # let decoder update
1827         srcstep = self.svstate.srcstep
1828         dststep = self.svstate.dststep
1829         ssubstep = self.svstate.ssubstep
1830         dsubstep = self.svstate.dsubstep
1831         vl = self.svstate.vl
1832         subvl = yield self.dec2.rm_dec.rm_in.subvl
1833         log("    srcstep", srcstep)
1834         log("    dststep", dststep)
1835         log("    ssubstep", ssubstep)
1836         log("    dsubstep", dsubstep)
1837         log("         vl", vl)
1838         log("      subvl", subvl)
1839
1840         # check if end reached (we let srcstep overrun, above)
1841         # nothing needs doing (TODO zeroing): just do next instruction
1842         return srcstep == vl or dststep == vl
1843
1844     def svstate_post_inc(self, insn_name, vf=0):
1845         # check if SV "Vertical First" mode is enabled
1846         vfirst = self.svstate.vfirst
1847         log("    SV Vertical First", vf, vfirst)
1848         if not vf and vfirst == 1:
1849             self.update_nia()
1850             return True
1851
1852         # check if it is the SVSTATE.src/dest step that needs incrementing
1853         # this is our Sub-Program-Counter loop from 0 to VL-1
1854         # XXX twin predication TODO
1855         vl = self.svstate.vl
1856         subvl = yield self.dec2.rm_dec.rm_in.subvl
1857         mvl = self.svstate.maxvl
1858         srcstep = self.svstate.srcstep
1859         dststep = self.svstate.dststep
1860         ssubstep = self.svstate.ssubstep
1861         dsubstep = self.svstate.dsubstep
1862         rm_mode = yield self.dec2.rm_dec.mode
1863         reverse_gear = yield self.dec2.rm_dec.reverse_gear
1864         sv_ptype = yield self.dec2.dec.op.SV_Ptype
1865         out_vec = not (yield self.dec2.no_out_vec)
1866         in_vec = not (yield self.dec2.no_in_vec)
1867         log("    svstate.vl", vl)
1868         log("    svstate.mvl", mvl)
1869         log("         rm.subvl", subvl)
1870         log("    svstate.srcstep", srcstep)
1871         log("    svstate.dststep", dststep)
1872         log("    svstate.ssubstep", ssubstep)
1873         log("    svstate.dsubstep", dsubstep)
1874         log("    mode", rm_mode)
1875         log("    reverse", reverse_gear)
1876         log("    out_vec", out_vec)
1877         log("    in_vec", in_vec)
1878         log("    sv_ptype", sv_ptype, sv_ptype == SVPtype.P2.value)
1879         # check if srcstep needs incrementing by one, stop PC advancing
1880         # svp64 loop can end early if the dest is scalar for single-pred
1881         # but for 2-pred both src/dest have to be checked.
1882         # XXX this might not be true! it may just be LD/ST
1883         if sv_ptype == SVPtype.P2.value:
1884             svp64_is_vector = (out_vec or in_vec)
1885         else:
1886             svp64_is_vector = out_vec
1887         # check if this was an sv.bc* and if so did it succeed
1888         if self.is_svp64_mode and insn_name.startswith("sv.bc"):
1889             end_loop = self.namespace['end_loop']
1890             log("branch %s end_loop" % insn_name, end_loop)
1891             if end_loop.value:
1892                 self.svp64_reset_loop()
1893                 self.update_pc_next()
1894                 return False
1895         # loops end at the first "hit" (source or dest)
1896         loopend = ((srcstep == vl-1 and ssubstep == subvl) or
1897                    (dststep == vl-1 and dsubstep == subvl))
1898         if svp64_is_vector and not loopend:
1899             yield from self.advance_svstate_steps()
1900             self.namespace['SVSTATE'] = self.svstate
1901             # not an SVP64 branch, so fix PC (NIA==CIA) for next loop
1902             # (by default, NIA is CIA+4 if v3.0B or CIA+8 if SVP64)
1903             # this way we keep repeating the same instruction (with new steps)
1904             self.pc.NIA.value = self.pc.CIA.value
1905             self.namespace['NIA'] = self.pc.NIA
1906             log("end of sub-pc call", self.namespace['CIA'],
1907                 self.namespace['NIA'])
1908             return False  # DO NOT allow PC update whilst Sub-PC loop running
1909
1910         # reset loop to zero and update NIA
1911         self.svp64_reset_loop()
1912         self.update_nia()
1913
1914         return True
1915
1916     def advance_svstate_steps(self, end_src=False, end_dst=False):
1917         """ advance sub/steps. note that Pack/Unpack *INVERTS* the order.
1918         TODO when Pack/Unpack is set, substep becomes the *outer* loop
1919         """
1920         subvl = yield self.dec2.rm_dec.rm_in.subvl
1921         # first source step
1922         ssubstep = self.svstate.ssubstep
1923         end_sub = ssubstep == subvl
1924         if end_sub:
1925             if not end_src:
1926                 self.svstate.srcstep += SelectableInt(1, 7)
1927             self.svstate.ssubstep = SelectableInt(0, 2)  # reset
1928         else:
1929             self.svstate.ssubstep += SelectableInt(1, 2) # advance ssubstep
1930         # now dest step
1931         dsubstep = self.svstate.dsubstep
1932         end_sub = dsubstep == subvl
1933         if end_sub:
1934             if not end_dst:
1935                 self.svstate.dststep += SelectableInt(1, 7)
1936             self.svstate.dsubstep = SelectableInt(0, 2)  # reset
1937         else:
1938             self.svstate.dsubstep += SelectableInt(1, 2) # advance ssubstep
1939
1940     def update_pc_next(self):
1941         # UPDATE program counter
1942         self.pc.update(self.namespace, self.is_svp64_mode)
1943         self.svstate.spr = self.namespace['SVSTATE']
1944         log("end of call", self.namespace['CIA'],
1945             self.namespace['NIA'],
1946             self.namespace['SVSTATE'])
1947
1948     def svp64_reset_loop(self):
1949         self.svstate.srcstep = 0
1950         self.svstate.dststep = 0
1951         self.svstate.ssubstep = 0
1952         self.svstate.dsubstep = 0
1953         log("    svstate.srcstep loop end (PC to update)")
1954         self.namespace['SVSTATE'] = self.svstate
1955
1956     def update_nia(self):
1957         self.pc.update_nia(self.is_svp64_mode)
1958         self.namespace['NIA'] = self.pc.NIA
1959
1960
1961 def inject():
1962     """Decorator factory.
1963
1964     this decorator will "inject" variables into the function's namespace,
1965     from the *dictionary* in self.namespace.  it therefore becomes possible
1966     to make it look like a whole stack of variables which would otherwise
1967     need "self." inserted in front of them (*and* for those variables to be
1968     added to the instance) "appear" in the function.
1969
1970     "self.namespace['SI']" for example becomes accessible as just "SI" but
1971     *only* inside the function, when decorated.
1972     """
1973     def variable_injector(func):
1974         @wraps(func)
1975         def decorator(*args, **kwargs):
1976             try:
1977                 func_globals = func.__globals__  # Python 2.6+
1978             except AttributeError:
1979                 func_globals = func.func_globals  # Earlier versions.
1980
1981             context = args[0].namespace  # variables to be injected
1982             saved_values = func_globals.copy()  # Shallow copy of dict.
1983             log("globals before", context.keys())
1984             func_globals.update(context)
1985             result = func(*args, **kwargs)
1986             log("globals after", func_globals['CIA'], func_globals['NIA'])
1987             log("args[0]", args[0].namespace['CIA'],
1988                 args[0].namespace['NIA'],
1989                 args[0].namespace['SVSTATE'])
1990             if 'end_loop' in func_globals:
1991                 log("args[0] end_loop", func_globals['end_loop'])
1992             args[0].namespace = func_globals
1993             #exec (func.__code__, func_globals)
1994
1995             # finally:
1996             #    func_globals = saved_values  # Undo changes.
1997
1998             return result
1999
2000         return decorator
2001
2002     return variable_injector