src/soc/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 from nmigen.back.pysim import Settle
  17 from functools import wraps
  18 from copy import copy
  19 from soc.decoder.orderedset import OrderedSet
  20 from soc.decoder.selectable_int import (FieldSelectableInt, SelectableInt,
  21                                         selectconcat)
  22 from soc.decoder.power_enums import (spr_dict, spr_byname, XER_bits,
  23                                      insns, MicrOp, In1Sel, In2Sel, In3Sel,
  24                                      OutSel, CROutSel,
  25                                      SVP64RMMode, SVP64PredMode,
  26                                      SVP64PredInt, SVP64PredCR)
  27
  28 from soc.decoder.power_enums import SVPtype
  29
  30 from soc.decoder.helpers import exts, gtu, ltu, undefined
  31 from soc.consts import PIb, MSRb  # big-endian (PowerISA versions)
  32 from soc.consts import SVP64CROffs
  33 from soc.decoder.power_svp64 import SVP64RM, decode_extra
  34
  35 from soc.decoder.isa.radixmmu import RADIX
  36 from soc.decoder.isa.mem import Mem, swap_order
  37
  38 from collections import namedtuple
  39 import math
  40 import sys
  41
  42 instruction_info = namedtuple('instruction_info',
  43                               'func read_regs uninit_regs write_regs ' +
  44                               'special_regs op_fields form asmregs')
  45
  46 special_sprs = {
  47     'LR': 8,
  48     'CTR': 9,
  49     'TAR': 815,
  50     'XER': 1,
  51     'VRSAVE': 256}
  52
  53
  54 REG_SORT_ORDER = {
  55     # TODO (lkcl): adjust other registers that should be in a particular order
  56     # probably CA, CA32, and CR
  57     "RT": 0,
  58     "RA": 0,
  59     "RB": 0,
  60     "RS": 0,
  61     "CR": 0,
  62     "LR": 0,
  63     "CTR": 0,
  64     "TAR": 0,
  65     "CA": 0,
  66     "CA32": 0,
  67     "MSR": 0,
  68     "SVSTATE": 0,
  69
  70     "overflow": 1,
  71 }
  72
  73
  74 def create_args(reglist, extra=None):
  75     retval = list(OrderedSet(reglist))
  76     retval.sort(key=lambda reg: REG_SORT_ORDER[reg])
  77     if extra is not None:
  78         return [extra] + retval
  79     return retval
  80
  81
  82
  83 class GPR(dict):
  84     def __init__(self, decoder, isacaller, svstate, regfile):
  85         dict.__init__(self)
  86         self.sd = decoder
  87         self.isacaller = isacaller
  88         self.svstate = svstate
  89         for i in range(32):
  90             self[i] = SelectableInt(regfile[i], 64)
  91
  92     def __call__(self, ridx):
  93         return self[ridx]
  94
  95     def set_form(self, form):
  96         self.form = form
  97
  98     def getz(self, rnum):
  99         # rnum = rnum.value # only SelectableInt allowed
 100         print("GPR getzero?", rnum)
 101         if rnum == 0:
 102             return SelectableInt(0, 64)
 103         return self[rnum]
 104
 105     def _get_regnum(self, attr):
 106         getform = self.sd.sigforms[self.form]
 107         rnum = getattr(getform, attr)
 108         return rnum
 109
 110     def ___getitem__(self, attr):
 111         """ XXX currently not used
 112         """
 113         rnum = self._get_regnum(attr)
 114         offs = self.svstate.srcstep
 115         print("GPR getitem", attr, rnum, "srcoffs", offs)
 116         return self.regfile[rnum]
 117
 118     def dump(self):
 119         for i in range(0, len(self), 8):
 120             s = []
 121             for j in range(8):
 122                 s.append("%08x" % self[i+j].value)
 123             s = ' '.join(s)
 124             print("reg", "%2d" % i, s)
 125
 126
 127 class SPR(dict):
 128     def __init__(self, dec2, initial_sprs={}):
 129         self.sd = dec2
 130         dict.__init__(self)
 131         for key, v in initial_sprs.items():
 132             if isinstance(key, SelectableInt):
 133                 key = key.value
 134             key = special_sprs.get(key, key)
 135             if isinstance(key, int):
 136                 info = spr_dict[key]
 137             else:
 138                 info = spr_byname[key]
 139             if not isinstance(v, SelectableInt):
 140                 v = SelectableInt(v, info.length)
 141             self[key] = v
 142
 143     def __getitem__(self, key):
 144         print("get spr", key)
 145         print("dict", self.items())
 146         # if key in special_sprs get the special spr, otherwise return key
 147         if isinstance(key, SelectableInt):
 148             key = key.value
 149         if isinstance(key, int):
 150             key = spr_dict[key].SPR
 151         key = special_sprs.get(key, key)
 152         if key == 'HSRR0':  # HACK!
 153             key = 'SRR0'
 154         if key == 'HSRR1':  # HACK!
 155             key = 'SRR1'
 156         if key in self:
 157             res = dict.__getitem__(self, key)
 158         else:
 159             if isinstance(key, int):
 160                 info = spr_dict[key]
 161             else:
 162                 info = spr_byname[key]
 163             dict.__setitem__(self, key, SelectableInt(0, info.length))
 164             res = dict.__getitem__(self, key)
 165         print("spr returning", key, res)
 166         return res
 167
 168     def __setitem__(self, key, value):
 169         if isinstance(key, SelectableInt):
 170             key = key.value
 171         if isinstance(key, int):
 172             key = spr_dict[key].SPR
 173             print("spr key", key)
 174         key = special_sprs.get(key, key)
 175         if key == 'HSRR0':  # HACK!
 176             self.__setitem__('SRR0', value)
 177         if key == 'HSRR1':  # HACK!
 178             self.__setitem__('SRR1', value)
 179         print("setting spr", key, value)
 180         dict.__setitem__(self, key, value)
 181
 182     def __call__(self, ridx):
 183         return self[ridx]
 184
 185
 186 class PC:
 187     def __init__(self, pc_init=0):
 188         self.CIA = SelectableInt(pc_init, 64)
 189         self.NIA = self.CIA + SelectableInt(4, 64) # only true for v3.0B!
 190
 191     def update_nia(self, is_svp64):
 192         increment = 8 if is_svp64 else 4
 193         self.NIA = self.CIA + SelectableInt(increment, 64)
 194
 195     def update(self, namespace, is_svp64):
 196         """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
 197         """
 198         self.CIA = namespace['NIA'].narrow(64)
 199         self.update_nia(is_svp64)
 200         namespace['CIA'] = self.CIA
 201         namespace['NIA'] = self.NIA
 202
 203
 204 # Simple-V: see https://libre-soc.org/openpower/sv
 205 class SVP64State:
 206     def __init__(self, init=0):
 207         self.spr = SelectableInt(init, 32)
 208         # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
 209         self.maxvl = FieldSelectableInt(self.spr, tuple(range(0,7)))
 210         self.vl = FieldSelectableInt(self.spr, tuple(range(7,14)))
 211         self.srcstep = FieldSelectableInt(self.spr, tuple(range(14,21)))
 212         self.dststep = FieldSelectableInt(self.spr, tuple(range(21,28)))
 213         self.subvl = FieldSelectableInt(self.spr, tuple(range(28,30)))
 214         self.svstep = FieldSelectableInt(self.spr, tuple(range(30,32)))
 215
 216
 217 # SVP64 ReMap field
 218 class SVP64RMFields:
 219     def __init__(self, init=0):
 220         self.spr = SelectableInt(init, 24)
 221         # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
 222         self.mmode = FieldSelectableInt(self.spr, [0])
 223         self.mask = FieldSelectableInt(self.spr, tuple(range(1,4)))
 224         self.elwidth = FieldSelectableInt(self.spr, tuple(range(4,6)))
 225         self.ewsrc = FieldSelectableInt(self.spr, tuple(range(6,8)))
 226         self.subvl = FieldSelectableInt(self.spr, tuple(range(8,10)))
 227         self.extra = FieldSelectableInt(self.spr, tuple(range(10,19)))
 228         self.mode = FieldSelectableInt(self.spr, tuple(range(19,24)))
 229         # these cover the same extra field, split into parts as EXTRA2
 230         self.extra2 = list(range(4))
 231         self.extra2[0] = FieldSelectableInt(self.spr, tuple(range(10,12)))
 232         self.extra2[1] = FieldSelectableInt(self.spr, tuple(range(12,14)))
 233         self.extra2[2] = FieldSelectableInt(self.spr, tuple(range(14,16)))
 234         self.extra2[3] = FieldSelectableInt(self.spr, tuple(range(16,18)))
 235         self.smask = FieldSelectableInt(self.spr, tuple(range(16,19)))
 236         # and here as well, but EXTRA3
 237         self.extra3 = list(range(3))
 238         self.extra3[0] = FieldSelectableInt(self.spr, tuple(range(10,13)))
 239         self.extra3[1] = FieldSelectableInt(self.spr, tuple(range(13,16)))
 240         self.extra3[2] = FieldSelectableInt(self.spr, tuple(range(16,19)))
 241
 242
 243 SVP64RM_MMODE_SIZE = len(SVP64RMFields().mmode.br)
 244 SVP64RM_MASK_SIZE = len(SVP64RMFields().mask.br)
 245 SVP64RM_ELWIDTH_SIZE = len(SVP64RMFields().elwidth.br)
 246 SVP64RM_EWSRC_SIZE = len(SVP64RMFields().ewsrc.br)
 247 SVP64RM_SUBVL_SIZE = len(SVP64RMFields().subvl.br)
 248 SVP64RM_EXTRA2_SPEC_SIZE = len(SVP64RMFields().extra2[0].br)
 249 SVP64RM_EXTRA3_SPEC_SIZE = len(SVP64RMFields().extra3[0].br)
 250 SVP64RM_SMASK_SIZE = len(SVP64RMFields().smask.br)
 251 SVP64RM_MODE_SIZE = len(SVP64RMFields().mode.br)
 252
 253
 254 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
 255 class SVP64PrefixFields:
 256     def __init__(self):
 257         self.insn = SelectableInt(0, 32)
 258         # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
 259         self.major = FieldSelectableInt(self.insn, tuple(range(0,6)))
 260         self.pid = FieldSelectableInt(self.insn, (7, 9)) # must be 0b11
 261         rmfields = [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
 262         self.rm = FieldSelectableInt(self.insn, rmfields)
 263
 264
 265 SV64P_MAJOR_SIZE = len(SVP64PrefixFields().major.br)
 266 SV64P_PID_SIZE = len(SVP64PrefixFields().pid.br)
 267 SV64P_RM_SIZE = len(SVP64PrefixFields().rm.br)
 268
 269 # decode SVP64 predicate integer to reg number and invert
 270 def get_predint(gpr, mask):
 271     r10 = gpr(10)
 272     r30 = gpr(30)
 273     print ("get_predint", mask, SVP64PredInt.ALWAYS.value)
 274     if mask == SVP64PredInt.ALWAYS.value:
 275         return 0xffff_ffff_ffff_ffff
 276     if mask == SVP64PredInt.R3_UNARY.value:
 277         return 1 << (gpr(3).value & 0b111111)
 278     if mask == SVP64PredInt.R3.value:
 279         return gpr(3).value
 280     if mask == SVP64PredInt.R3_N.value:
 281         return ~gpr(3).value
 282     if mask == SVP64PredInt.R10.value:
 283         return gpr(10).value
 284     if mask == SVP64PredInt.R10_N.value:
 285         return ~gpr(10).value
 286     if mask == SVP64PredInt.R30.value:
 287         return gpr(30).value
 288     if mask == SVP64PredInt.R30_N.value:
 289         return ~gpr(30).value
 290
 291 # decode SVP64 predicate CR to reg number and invert status
 292 def _get_predcr(mask):
 293     if mask == SVP64PredCR.LT.value:
 294         return 0, 1
 295     if mask == SVP64PredCR.GE.value:
 296         return 0, 0
 297     if mask == SVP64PredCR.GT.value:
 298         return 1, 1
 299     if mask == SVP64PredCR.LE.value:
 300         return 1, 0
 301     if mask == SVP64PredCR.EQ.value:
 302         return 2, 1
 303     if mask == SVP64PredCR.NE.value:
 304         return 2, 0
 305     if mask == SVP64PredCR.SO.value:
 306         return 3, 1
 307     if mask == SVP64PredCR.NS.value:
 308         return 3, 0
 309
 310 # read individual CR fields (0..VL-1), extract the required bit
 311 # and construct the mask
 312 def get_predcr(crl, mask, vl):
 313     idx, noninv = _get_predcr(mask)
 314     mask = 0
 315     for i in range(vl):
 316         cr = crl[i+SVP64CROffs.CRPred]
 317         if cr[idx].value == noninv:
 318             mask |= (1<<i)
 319     return mask
 320
 321
 322 def get_pdecode_idx_in(dec2, name):
 323     op = dec2.dec.op
 324     in1_sel = yield op.in1_sel
 325     in2_sel = yield op.in2_sel
 326     in3_sel = yield op.in3_sel
 327     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 328     in1 = yield dec2.e.read_reg1.data
 329     in2 = yield dec2.e.read_reg2.data
 330     in3 = yield dec2.e.read_reg3.data
 331     in1_isvec = yield dec2.in1_isvec
 332     in2_isvec = yield dec2.in2_isvec
 333     in3_isvec = yield dec2.in3_isvec
 334     print ("get_pdecode_idx_in in1", name, in1_sel, In1Sel.RA.value,
 335                                      in1, in1_isvec)
 336     print ("get_pdecode_idx_in in2", name, in2_sel, In2Sel.RB.value,
 337                                      in2, in2_isvec)
 338     print ("get_pdecode_idx_in in3", name, in3_sel, In3Sel.RS.value,
 339                                      in3, in3_isvec)
 340     # identify which regnames map to in1/2/3
 341     if name == 'RA':
 342         if (in1_sel == In1Sel.RA.value or
 343             (in1_sel == In1Sel.RA_OR_ZERO.value and in1 != 0)):
 344             return in1, in1_isvec
 345         if in1_sel == In1Sel.RA_OR_ZERO.value:
 346             return in1, in1_isvec
 347     elif name == 'RB':
 348         if in2_sel == In2Sel.RB.value:
 349             return in2, in2_isvec
 350         if in3_sel == In3Sel.RB.value:
 351             return in3, in3_isvec
 352     # XXX TODO, RC doesn't exist yet!
 353     elif name == 'RC':
 354         assert False, "RC does not exist yet"
 355     elif name == 'RS':
 356         if in1_sel == In1Sel.RS.value:
 357             return in1, in1_isvec
 358         if in2_sel == In2Sel.RS.value:
 359             return in2, in2_isvec
 360         if in3_sel == In3Sel.RS.value:
 361             return in3, in3_isvec
 362     return None, False
 363
 364
 365 def get_pdecode_cr_out(dec2, name):
 366     op = dec2.dec.op
 367     out_sel = yield op.cr_out
 368     out_bitfield = yield dec2.dec_cr_out.cr_bitfield.data
 369     sv_cr_out = yield op.sv_cr_out
 370     spec = yield dec2.crout_svdec.spec
 371     sv_override = yield dec2.dec_cr_out.sv_override
 372     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 373     out = yield dec2.e.write_cr.data
 374     o_isvec = yield dec2.o_isvec
 375     print ("get_pdecode_cr_out", out_sel, CROutSel.CR0.value, out, o_isvec)
 376     print ("    sv_cr_out", sv_cr_out)
 377     print ("    cr_bf", out_bitfield)
 378     print ("    spec", spec)
 379     print ("    override", sv_override)
 380     # identify which regnames map to out / o2
 381     if name == 'CR0':
 382         if out_sel == CROutSel.CR0.value:
 383             return out, o_isvec
 384     print ("get_pdecode_idx_out not found", name)
 385     return None, False
 386
 387
 388 def get_pdecode_idx_out(dec2, name):
 389     op = dec2.dec.op
 390     out_sel = yield op.out_sel
 391     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 392     out = yield dec2.e.write_reg.data
 393     o_isvec = yield dec2.o_isvec
 394     # identify which regnames map to out / o2
 395     if name == 'RA':
 396         print ("get_pdecode_idx_out", out_sel, OutSel.RA.value, out, o_isvec)
 397         if out_sel == OutSel.RA.value:
 398             return out, o_isvec
 399     elif name == 'RT':
 400         print ("get_pdecode_idx_out", out_sel, OutSel.RT.value,
 401                                       OutSel.RT_OR_ZERO.value, out, o_isvec)
 402         if out_sel == OutSel.RT.value:
 403             return out, o_isvec
 404     print ("get_pdecode_idx_out not found", name)
 405     return None, False
 406
 407
 408 # XXX TODO
 409 def get_pdecode_idx_out2(dec2, name):
 410     op = dec2.dec.op
 411     print ("TODO: get_pdecode_idx_out2", name)
 412     return None, False
 413
 414
 415 class ISACaller:
 416     # decoder2 - an instance of power_decoder2
 417     # regfile - a list of initial values for the registers
 418     # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
 419     # respect_pc - tracks the program counter.  requires initial_insns
 420     def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
 421                  initial_mem=None, initial_msr=0,
 422                  initial_svstate=0,
 423                  initial_insns=None, respect_pc=False,
 424                  disassembly=None,
 425                  initial_pc=0,
 426                  bigendian=False,
 427                  mmu=False):
 428
 429         self.bigendian = bigendian
 430         self.halted = False
 431         self.is_svp64_mode = False
 432         self.respect_pc = respect_pc
 433         if initial_sprs is None:
 434             initial_sprs = {}
 435         if initial_mem is None:
 436             initial_mem = {}
 437         if initial_insns is None:
 438             initial_insns = {}
 439             assert self.respect_pc == False, "instructions required to honor pc"
 440
 441         print("ISACaller insns", respect_pc, initial_insns, disassembly)
 442         print("ISACaller initial_msr", initial_msr)
 443
 444         # "fake program counter" mode (for unit testing)
 445         self.fake_pc = 0
 446         disasm_start = 0
 447         if not respect_pc:
 448             if isinstance(initial_mem, tuple):
 449                 self.fake_pc = initial_mem[0]
 450                 disasm_start = self.fake_pc
 451         else:
 452             disasm_start = initial_pc
 453
 454         # disassembly: we need this for now (not given from the decoder)
 455         self.disassembly = {}
 456         if disassembly:
 457             for i, code in enumerate(disassembly):
 458                 self.disassembly[i*4 + disasm_start] = code
 459
 460         # set up registers, instruction memory, data memory, PC, SPRs, MSR
 461         self.svp64rm = SVP64RM()
 462         if initial_svstate is None:
 463             initial_svstate = 0
 464         if isinstance(initial_svstate, int):
 465             initial_svstate = SVP64State(initial_svstate)
 466         self.svstate = initial_svstate
 467         self.gpr = GPR(decoder2, self, self.svstate, regfile)
 468         self.spr = SPR(decoder2, initial_sprs) # initialise SPRs before MMU
 469         self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
 470         self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
 471         # MMU mode, redirect underlying Mem through RADIX
 472         if mmu:
 473             self.mem = RADIX(self.mem, self)
 474             self.imem = RADIX(self.imem, self)
 475         self.pc = PC()
 476         self.msr = SelectableInt(initial_msr, 64)  # underlying reg
 477
 478         # TODO, needed here:
 479         # FPR (same as GPR except for FP nums)
 480         # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
 481         #            note that mffs, mcrfs, mtfsf "manage" this FPSCR
 482         # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
 483         #         note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
 484         #         -- Done
 485         # 2.3.2 LR   (actually SPR #8) -- Done
 486         # 2.3.3 CTR  (actually SPR #9) -- Done
 487         # 2.3.4 TAR  (actually SPR #815)
 488         # 3.2.2 p45 XER  (actually SPR #1) -- Done
 489         # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
 490
 491         # create CR then allow portions of it to be "selectable" (below)
 492         #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
 493         self.cr = SelectableInt(initial_cr, 64)  # underlying reg
 494         #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
 495
 496         # "undefined", just set to variable-bit-width int (use exts "max")
 497         #self.undefined = SelectableInt(0, 256)  # TODO, not hard-code 256!
 498
 499         self.namespace = {}
 500         self.namespace.update(self.spr)
 501         self.namespace.update({'GPR': self.gpr,
 502                                'MEM': self.mem,
 503                                'SPR': self.spr,
 504                                'memassign': self.memassign,
 505                                'NIA': self.pc.NIA,
 506                                'CIA': self.pc.CIA,
 507                                'SVSTATE': self.svstate.spr,
 508                                'CR': self.cr,
 509                                'MSR': self.msr,
 510                                'undefined': undefined,
 511                                'mode_is_64bit': True,
 512                                'SO': XER_bits['SO']
 513                                })
 514
 515         # update pc to requested start point
 516         self.set_pc(initial_pc)
 517
 518         # field-selectable versions of Condition Register TODO check bitranges?
 519         self.crl = []
 520         for i in range(8):
 521             bits = tuple(range(i*4+32, (i+1)*4+32))  # errr... maybe?
 522             _cr = FieldSelectableInt(self.cr, bits)
 523             self.crl.append(_cr)
 524             self.namespace["CR%d" % i] = _cr
 525
 526         self.decoder = decoder2.dec
 527         self.dec2 = decoder2
 528
 529     def TRAP(self, trap_addr=0x700, trap_bit=PIb.TRAP):
 530         print("TRAP:", hex(trap_addr), hex(self.namespace['MSR'].value))
 531         # store CIA(+4?) in SRR0, set NIA to 0x700
 532         # store MSR in SRR1, set MSR to um errr something, have to check spec
 533         self.spr['SRR0'].value = self.pc.CIA.value
 534         self.spr['SRR1'].value = self.namespace['MSR'].value
 535         self.trap_nia = SelectableInt(trap_addr, 64)
 536         self.spr['SRR1'][trap_bit] = 1  # change *copy* of MSR in SRR1
 537
 538         # set exception bits.  TODO: this should, based on the address
 539         # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
 540         # bits appropriately.  however it turns out that *for now* in all
 541         # cases (all trap_addrs) the exact same thing is needed.
 542         self.msr[MSRb.IR] = 0
 543         self.msr[MSRb.DR] = 0
 544         self.msr[MSRb.FE0] = 0
 545         self.msr[MSRb.FE1] = 0
 546         self.msr[MSRb.EE] = 0
 547         self.msr[MSRb.RI] = 0
 548         self.msr[MSRb.SF] = 1
 549         self.msr[MSRb.TM] = 0
 550         self.msr[MSRb.VEC] = 0
 551         self.msr[MSRb.VSX] = 0
 552         self.msr[MSRb.PR] = 0
 553         self.msr[MSRb.FP] = 0
 554         self.msr[MSRb.PMM] = 0
 555         self.msr[MSRb.TEs] = 0
 556         self.msr[MSRb.TEe] = 0
 557         self.msr[MSRb.UND] = 0
 558         self.msr[MSRb.LE] = 1
 559
 560     def memassign(self, ea, sz, val):
 561         self.mem.memassign(ea, sz, val)
 562
 563     def prep_namespace(self, formname, op_fields):
 564         # TODO: get field names from form in decoder*1* (not decoder2)
 565         # decoder2 is hand-created, and decoder1.sigform is auto-generated
 566         # from spec
 567         # then "yield" fields only from op_fields rather than hard-coded
 568         # list, here.
 569         fields = self.decoder.sigforms[formname]
 570         for name in op_fields:
 571             if name == 'spr':
 572                 sig = getattr(fields, name.upper())
 573             else:
 574                 sig = getattr(fields, name)
 575             val = yield sig
 576             # these are all opcode fields involved in index-selection of CR,
 577             # and need to do "standard" arithmetic.  CR[BA+32] for example
 578             # would, if using SelectableInt, only be 5-bit.
 579             if name in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
 580                 self.namespace[name] = val
 581             else:
 582                 self.namespace[name] = SelectableInt(val, sig.width)
 583
 584         self.namespace['XER'] = self.spr['XER']
 585         self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
 586         self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
 587
 588     def handle_carry_(self, inputs, outputs, already_done):
 589         inv_a = yield self.dec2.e.do.invert_in
 590         if inv_a:
 591             inputs[0] = ~inputs[0]
 592
 593         imm_ok = yield self.dec2.e.do.imm_data.ok
 594         if imm_ok:
 595             imm = yield self.dec2.e.do.imm_data.data
 596             inputs.append(SelectableInt(imm, 64))
 597         assert len(outputs) >= 1
 598         print("outputs", repr(outputs))
 599         if isinstance(outputs, list) or isinstance(outputs, tuple):
 600             output = outputs[0]
 601         else:
 602             output = outputs
 603         gts = []
 604         for x in inputs:
 605             print("gt input", x, output)
 606             gt = (gtu(x, output))
 607             gts.append(gt)
 608         print(gts)
 609         cy = 1 if any(gts) else 0
 610         print("CA", cy, gts)
 611         if not (1 & already_done):
 612             self.spr['XER'][XER_bits['CA']] = cy
 613
 614         print("inputs", already_done, inputs)
 615         # 32 bit carry
 616         # ARGH... different for OP_ADD... *sigh*...
 617         op = yield self.dec2.e.do.insn_type
 618         if op == MicrOp.OP_ADD.value:
 619             res32 = (output.value & (1 << 32)) != 0
 620             a32 = (inputs[0].value & (1 << 32)) != 0
 621             if len(inputs) >= 2:
 622                 b32 = (inputs[1].value & (1 << 32)) != 0
 623             else:
 624                 b32 = False
 625             cy32 = res32 ^ a32 ^ b32
 626             print("CA32 ADD", cy32)
 627         else:
 628             gts = []
 629             for x in inputs:
 630                 print("input", x, output)
 631                 print("     x[32:64]", x, x[32:64])
 632                 print("     o[32:64]", output, output[32:64])
 633                 gt = (gtu(x[32:64], output[32:64])) == SelectableInt(1, 1)
 634                 gts.append(gt)
 635             cy32 = 1 if any(gts) else 0
 636             print("CA32", cy32, gts)
 637         if not (2 & already_done):
 638             self.spr['XER'][XER_bits['CA32']] = cy32
 639
 640     def handle_overflow(self, inputs, outputs, div_overflow):
 641         if hasattr(self.dec2.e.do, "invert_in"):
 642             inv_a = yield self.dec2.e.do.invert_in
 643             if inv_a:
 644                 inputs[0] = ~inputs[0]
 645
 646         imm_ok = yield self.dec2.e.do.imm_data.ok
 647         if imm_ok:
 648             imm = yield self.dec2.e.do.imm_data.data
 649             inputs.append(SelectableInt(imm, 64))
 650         assert len(outputs) >= 1
 651         print("handle_overflow", inputs, outputs, div_overflow)
 652         if len(inputs) < 2 and div_overflow is None:
 653             return
 654
 655         # div overflow is different: it's returned by the pseudo-code
 656         # because it's more complex than can be done by analysing the output
 657         if div_overflow is not None:
 658             ov, ov32 = div_overflow, div_overflow
 659         # arithmetic overflow can be done by analysing the input and output
 660         elif len(inputs) >= 2:
 661             output = outputs[0]
 662
 663             # OV (64-bit)
 664             input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
 665             output_sgn = exts(output.value, output.bits) < 0
 666             ov = 1 if input_sgn[0] == input_sgn[1] and \
 667                 output_sgn != input_sgn[0] else 0
 668
 669             # OV (32-bit)
 670             input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
 671             output32_sgn = exts(output.value, 32) < 0
 672             ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
 673                 output32_sgn != input32_sgn[0] else 0
 674
 675         self.spr['XER'][XER_bits['OV']] = ov
 676         self.spr['XER'][XER_bits['OV32']] = ov32
 677         so = self.spr['XER'][XER_bits['SO']]
 678         so = so | ov
 679         self.spr['XER'][XER_bits['SO']] = so
 680
 681     def handle_comparison(self, outputs, cr_idx=0):
 682         out = outputs[0]
 683         assert isinstance(out, SelectableInt), \
 684             "out zero not a SelectableInt %s" % repr(outputs)
 685         print("handle_comparison", out.bits, hex(out.value))
 686         # TODO - XXX *processor* in 32-bit mode
 687         # https://bugs.libre-soc.org/show_bug.cgi?id=424
 688         # if is_32bit:
 689         #    o32 = exts(out.value, 32)
 690         #    print ("handle_comparison exts 32 bit", hex(o32))
 691         out = exts(out.value, out.bits)
 692         print("handle_comparison exts", hex(out))
 693         zero = SelectableInt(out == 0, 1)
 694         positive = SelectableInt(out > 0, 1)
 695         negative = SelectableInt(out < 0, 1)
 696         SO = self.spr['XER'][XER_bits['SO']]
 697         print("handle_comparison SO", SO)
 698         cr_field = selectconcat(negative, positive, zero, SO)
 699         self.crl[cr_idx].eq(cr_field)
 700
 701     def set_pc(self, pc_val):
 702         self.namespace['NIA'] = SelectableInt(pc_val, 64)
 703         self.pc.update(self.namespace, self.is_svp64_mode)
 704
 705     def setup_one(self):
 706         """set up one instruction
 707         """
 708         if self.respect_pc:
 709             pc = self.pc.CIA.value
 710         else:
 711             pc = self.fake_pc
 712         self._pc = pc
 713         ins = self.imem.ld(pc, 4, False, True, instr_fetch=True)
 714         if ins is None:
 715             raise KeyError("no instruction at 0x%x" % pc)
 716         print("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
 717         print("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
 718
 719         yield self.dec2.sv_rm.eq(0)
 720         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
 721         yield self.dec2.dec.bigendian.eq(self.bigendian)
 722         yield self.dec2.state.msr.eq(self.msr.value)
 723         yield self.dec2.state.pc.eq(pc)
 724         if self.svstate is not None:
 725             yield self.dec2.state.svstate.eq(self.svstate.spr.value)
 726
 727         # SVP64.  first, check if the opcode is EXT001, and SVP64 id bits set
 728         yield Settle()
 729         opcode = yield self.dec2.dec.opcode_in
 730         pfx = SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
 731         pfx.insn.value = opcode
 732         major = pfx.major.asint(msb0=True) # MSB0 inversion
 733         print ("prefix test: opcode:", major, bin(major),
 734                 pfx.insn[7] == 0b1, pfx.insn[9] == 0b1)
 735         self.is_svp64_mode = ((major == 0b000001) and
 736                               pfx.insn[7].value == 0b1 and
 737                               pfx.insn[9].value == 0b1)
 738         self.pc.update_nia(self.is_svp64_mode)
 739         self.namespace['NIA'] = self.pc.NIA
 740         self.namespace['SVSTATE'] = self.svstate.spr
 741         if not self.is_svp64_mode:
 742             return
 743
 744         # in SVP64 mode.  decode/print out svp64 prefix, get v3.0B instruction
 745         print ("svp64.rm", bin(pfx.rm.asint(msb0=True)))
 746         print ("    svstate.vl", self.svstate.vl.asint(msb0=True))
 747         print ("    svstate.mvl", self.svstate.maxvl.asint(msb0=True))
 748         sv_rm = pfx.rm.asint(msb0=True)
 749         ins = self.imem.ld(pc+4, 4, False, True, instr_fetch=True)
 750         print("     svsetup: 0x%x 0x%x %s" % (pc+4, ins & 0xffffffff, bin(ins)))
 751         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff) # v3.0B suffix
 752         yield self.dec2.sv_rm.eq(sv_rm)                        # svp64 prefix
 753         yield Settle()
 754
 755     def execute_one(self):
 756         """execute one instruction
 757         """
 758         # get the disassembly code for this instruction
 759         if self.is_svp64_mode:
 760             code = self.disassembly[self._pc+4]
 761             print("    svp64 sim-execute", hex(self._pc), code)
 762         else:
 763             code = self.disassembly[self._pc]
 764             print("sim-execute", hex(self._pc), code)
 765         opname = code.split(' ')[0]
 766         yield from self.call(opname)
 767
 768         # don't use this except in special circumstances
 769         if not self.respect_pc:
 770             self.fake_pc += 4
 771
 772         print("execute one, CIA NIA", self.pc.CIA.value, self.pc.NIA.value)
 773
 774     def get_assembly_name(self):
 775         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 776         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 777         dec_insn = yield self.dec2.e.do.insn
 778         asmcode = yield self.dec2.dec.op.asmcode
 779         print("get assembly name asmcode", asmcode, hex(dec_insn))
 780         asmop = insns.get(asmcode, None)
 781         int_op = yield self.dec2.dec.op.internal_op
 782
 783         # sigh reconstruct the assembly instruction name
 784         if hasattr(self.dec2.e.do, "oe"):
 785             ov_en = yield self.dec2.e.do.oe.oe
 786             ov_ok = yield self.dec2.e.do.oe.ok
 787         else:
 788             ov_en = False
 789             ov_ok = False
 790         if hasattr(self.dec2.e.do, "rc"):
 791             rc_en = yield self.dec2.e.do.rc.rc
 792             rc_ok = yield self.dec2.e.do.rc.ok
 793         else:
 794             rc_en = False
 795             rc_ok = False
 796         # grrrr have to special-case MUL op (see DecodeOE)
 797         print("ov %d en %d rc %d en %d op %d" %
 798               (ov_ok, ov_en, rc_ok, rc_en, int_op))
 799         if int_op in [MicrOp.OP_MUL_H64.value, MicrOp.OP_MUL_H32.value]:
 800             print("mul op")
 801             if rc_en & rc_ok:
 802                 asmop += "."
 803         else:
 804             if not asmop.endswith("."):  # don't add "." to "andis."
 805                 if rc_en & rc_ok:
 806                     asmop += "."
 807         if hasattr(self.dec2.e.do, "lk"):
 808             lk = yield self.dec2.e.do.lk
 809             if lk:
 810                 asmop += "l"
 811         print("int_op", int_op)
 812         if int_op in [MicrOp.OP_B.value, MicrOp.OP_BC.value]:
 813             AA = yield self.dec2.dec.fields.FormI.AA[0:-1]
 814             print("AA", AA)
 815             if AA:
 816                 asmop += "a"
 817         spr_msb = yield from self.get_spr_msb()
 818         if int_op == MicrOp.OP_MFCR.value:
 819             if spr_msb:
 820                 asmop = 'mfocrf'
 821             else:
 822                 asmop = 'mfcr'
 823         # XXX TODO: for whatever weird reason this doesn't work
 824         # https://bugs.libre-soc.org/show_bug.cgi?id=390
 825         if int_op == MicrOp.OP_MTCRF.value:
 826             if spr_msb:
 827                 asmop = 'mtocrf'
 828             else:
 829                 asmop = 'mtcrf'
 830         return asmop
 831
 832     def get_spr_msb(self):
 833         dec_insn = yield self.dec2.e.do.insn
 834         return dec_insn & (1 << 20) != 0  # sigh - XFF.spr[-1]?
 835
 836     def call(self, name):
 837         """call(opcode) - the primary execution point for instructions
 838         """
 839         name = name.strip()  # remove spaces if not already done so
 840         if self.halted:
 841             print("halted - not executing", name)
 842             return
 843
 844         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 845         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 846         asmop = yield from self.get_assembly_name()
 847         print("call", name, asmop)
 848
 849         # check privileged
 850         int_op = yield self.dec2.dec.op.internal_op
 851         spr_msb = yield from self.get_spr_msb()
 852
 853         instr_is_privileged = False
 854         if int_op in [MicrOp.OP_ATTN.value,
 855                       MicrOp.OP_MFMSR.value,
 856                       MicrOp.OP_MTMSR.value,
 857                       MicrOp.OP_MTMSRD.value,
 858                       # TODO: OP_TLBIE
 859                       MicrOp.OP_RFID.value]:
 860             instr_is_privileged = True
 861         if int_op in [MicrOp.OP_MFSPR.value,
 862                       MicrOp.OP_MTSPR.value] and spr_msb:
 863             instr_is_privileged = True
 864
 865         print("is priv", instr_is_privileged, hex(self.msr.value),
 866               self.msr[MSRb.PR])
 867         # check MSR priv bit and whether op is privileged: if so, throw trap
 868         if instr_is_privileged and self.msr[MSRb.PR] == 1:
 869             self.TRAP(0x700, PIb.PRIV)
 870             self.namespace['NIA'] = self.trap_nia
 871             self.pc.update(self.namespace, self.is_svp64_mode)
 872             return
 873
 874         # check halted condition
 875         if name == 'attn':
 876             self.halted = True
 877             return
 878
 879         # check illegal instruction
 880         illegal = False
 881         if name not in ['mtcrf', 'mtocrf']:
 882             illegal = name != asmop
 883
 884         # sigh deal with setvl not being supported by binutils (.long)
 885         if asmop.startswith('setvl'):
 886             illegal = False
 887             name = 'setvl'
 888
 889         if illegal:
 890             print("illegal", name, asmop)
 891             self.TRAP(0x700, PIb.ILLEG)
 892             self.namespace['NIA'] = self.trap_nia
 893             self.pc.update(self.namespace, self.is_svp64_mode)
 894             print("name %s != %s - calling ILLEGAL trap, PC: %x" %
 895                   (name, asmop, self.pc.CIA.value))
 896             return
 897
 898         info = self.instrs[name]
 899         yield from self.prep_namespace(info.form, info.op_fields)
 900
 901         # preserve order of register names
 902         input_names = create_args(list(info.read_regs) +
 903                                   list(info.uninit_regs))
 904         print(input_names)
 905
 906         # get SVP64 entry for the current instruction
 907         sv_rm = self.svp64rm.instrs.get(name)
 908         if sv_rm is not None:
 909             dest_cr, src_cr, src_byname, dest_byname = decode_extra(sv_rm)
 910         else:
 911             dest_cr, src_cr, src_byname, dest_byname = False, False, {}, {}
 912         print ("sv rm", sv_rm, dest_cr, src_cr, src_byname, dest_byname)
 913
 914         # get SVSTATE VL (oh and print out some debug stuff)
 915         if self.is_svp64_mode:
 916             vl = self.svstate.vl.asint(msb0=True)
 917             srcstep = self.svstate.srcstep.asint(msb0=True)
 918             dststep = self.svstate.srcstep.asint(msb0=True)
 919             sv_a_nz = yield self.dec2.sv_a_nz
 920             in1 = yield self.dec2.e.read_reg1.data
 921             print ("SVP64: VL, srcstep, dststep, sv_a_nz, in1",
 922                     vl, srcstep, dststep, sv_a_nz, in1)
 923
 924         # get predicate mask
 925         srcmask = dstmask = 0xffff_ffff_ffff_ffff
 926         if self.is_svp64_mode:
 927             pmode = yield self.dec2.rm_dec.predmode
 928             sv_ptype = yield self.dec2.dec.op.SV_Ptype
 929             srcpred = yield self.dec2.rm_dec.srcpred
 930             dstpred = yield self.dec2.rm_dec.dstpred
 931             pred_src_zero = yield self.dec2.rm_dec.pred_sz
 932             pred_dst_zero = yield self.dec2.rm_dec.pred_dz
 933             if pmode == SVP64PredMode.INT.value:
 934                 srcmask = dstmask = get_predint(self.gpr, dstpred)
 935                 if sv_ptype == SVPtype.P2.value:
 936                     srcmask = get_predint(self.gpr, srcpred)
 937             elif pmode == SVP64PredMode.CR.value:
 938                 srcmask = dstmask = get_predcr(self.crl, dstpred, vl)
 939                 if sv_ptype == SVPtype.P2.value:
 940                     srcmask = get_predcr(self.crl, srcpred, vl)
 941             print ("    pmode", pmode)
 942             print ("    ptype", sv_ptype)
 943             print ("    srcpred", bin(srcpred))
 944             print ("    dstpred", bin(dstpred))
 945             print ("    srcmask", bin(srcmask))
 946             print ("    dstmask", bin(dstmask))
 947             print ("    pred_sz", bin(pred_src_zero))
 948             print ("    pred_dz", bin(pred_dst_zero))
 949
 950             # okaaay, so here we simply advance srcstep (TODO dststep)
 951             # until the predicate mask has a "1" bit... or we run out of VL
 952             # let srcstep==VL be the indicator to move to next instruction
 953             if not pred_src_zero:
 954                 while (((1<<srcstep) & srcmask) == 0) and (srcstep != vl):
 955                     print ("      skip", bin(1<<srcstep))
 956                     srcstep += 1
 957             # same for dststep
 958             if not pred_dst_zero:
 959                 while (((1<<dststep) & dstmask) == 0) and (dststep != vl):
 960                     print ("      skip", bin(1<<dststep))
 961                     dststep += 1
 962
 963             # now work out if the relevant mask bits require zeroing
 964             if pred_dst_zero:
 965                 pred_dst_zero = ((1<<dststep) & dstmask) == 0
 966             if pred_src_zero:
 967                 pred_src_zero = ((1<<srcstep) & srcmask) == 0
 968
 969             # update SVSTATE with new srcstep
 970             self.svstate.srcstep[0:7] = srcstep
 971             self.svstate.dststep[0:7] = dststep
 972             self.namespace['SVSTATE'] = self.svstate.spr
 973             yield self.dec2.state.svstate.eq(self.svstate.spr.value)
 974             yield Settle() # let decoder update
 975             srcstep = self.svstate.srcstep.asint(msb0=True)
 976             dststep = self.svstate.dststep.asint(msb0=True)
 977             print ("    srcstep", srcstep)
 978             print ("    dststep", dststep)
 979
 980             # check if end reached (we let srcstep overrun, above)
 981             # nothing needs doing (TODO zeroing): just do next instruction
 982             if srcstep == vl:
 983                 self.svp64_reset_loop()
 984                 self.update_pc_next()
 985                 return
 986
 987         # VL=0 in SVP64 mode means "do nothing: skip instruction"
 988         if self.is_svp64_mode and vl == 0:
 989             self.pc.update(self.namespace, self.is_svp64_mode)
 990             print("SVP64: VL=0, end of call", self.namespace['CIA'],
 991                                        self.namespace['NIA'])
 992             return
 993
 994         # main input registers (RT, RA ...)
 995         inputs = []
 996         for name in input_names:
 997             # using PowerDecoder2, first, find the decoder index.
 998             # (mapping name RA RB RC RS to in1, in2, in3)
 999             regnum, is_vec = yield from get_pdecode_idx_in(self.dec2, name)
1000             if regnum is None:
1001                 # doing this is not part of svp64, it's because output
1002                 # registers, to be modified, need to be in the namespace.
1003                 regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, name)
1004
1005             # in case getting the register number is needed, _RA, _RB
1006             regname = "_" + name
1007             self.namespace[regname] = regnum
1008             if not self.is_svp64_mode or not pred_src_zero:
1009                 print('reading reg %s %s' % (name, str(regnum)), is_vec)
1010                 reg_val = self.gpr(regnum)
1011             else:
1012                 print('zero input reg %s %s' % (name, str(regnum)), is_vec)
1013                 reg_val = 0
1014             inputs.append(reg_val)
1015
1016         # "special" registers
1017         for special in info.special_regs:
1018             if special in special_sprs:
1019                 inputs.append(self.spr[special])
1020             else:
1021                 inputs.append(self.namespace[special])
1022
1023         # clear trap (trap) NIA
1024         self.trap_nia = None
1025
1026         # execute actual instruction here
1027         print("inputs", inputs)
1028         results = info.func(self, *inputs)
1029         print("results", results)
1030
1031         # "inject" decorator takes namespace from function locals: we need to
1032         # overwrite NIA being overwritten (sigh)
1033         if self.trap_nia is not None:
1034             self.namespace['NIA'] = self.trap_nia
1035
1036         print("after func", self.namespace['CIA'], self.namespace['NIA'])
1037
1038         # detect if CA/CA32 already in outputs (sra*, basically)
1039         already_done = 0
1040         if info.write_regs:
1041             output_names = create_args(info.write_regs)
1042             for name in output_names:
1043                 if name == 'CA':
1044                     already_done |= 1
1045                 if name == 'CA32':
1046                     already_done |= 2
1047
1048         print("carry already done?", bin(already_done))
1049         if hasattr(self.dec2.e.do, "output_carry"):
1050             carry_en = yield self.dec2.e.do.output_carry
1051         else:
1052             carry_en = False
1053         if carry_en:
1054             yield from self.handle_carry_(inputs, results, already_done)
1055
1056         if not self.is_svp64_mode: # yeah just no. not in parallel processing
1057             # detect if overflow was in return result
1058             overflow = None
1059             if info.write_regs:
1060                 for name, output in zip(output_names, results):
1061                     if name == 'overflow':
1062                         overflow = output
1063
1064             if hasattr(self.dec2.e.do, "oe"):
1065                 ov_en = yield self.dec2.e.do.oe.oe
1066                 ov_ok = yield self.dec2.e.do.oe.ok
1067             else:
1068                 ov_en = False
1069                 ov_ok = False
1070             print("internal overflow", overflow, ov_en, ov_ok)
1071             if ov_en & ov_ok:
1072                 yield from self.handle_overflow(inputs, results, overflow)
1073
1074         # only do SVP64 dest predicated Rc=1 if dest-pred is not enabled
1075         rc_en = False
1076         if not self.is_svp64_mode or not pred_dst_zero:
1077             if hasattr(self.dec2.e.do, "rc"):
1078                 rc_en = yield self.dec2.e.do.rc.rc
1079         if rc_en:
1080             regnum, is_vec = yield from get_pdecode_cr_out(self.dec2, "CR0")
1081             self.handle_comparison(results, regnum)
1082
1083         # any modified return results?
1084         if info.write_regs:
1085             for name, output in zip(output_names, results):
1086                 if name == 'overflow':  # ignore, done already (above)
1087                     continue
1088                 if isinstance(output, int):
1089                     output = SelectableInt(output, 256)
1090                 if name in ['CA', 'CA32']:
1091                     if carry_en:
1092                         print("writing %s to XER" % name, output)
1093                         self.spr['XER'][XER_bits[name]] = output.value
1094                     else:
1095                         print("NOT writing %s to XER" % name, output)
1096                 elif name in info.special_regs:
1097                     print('writing special %s' % name, output, special_sprs)
1098                     if name in special_sprs:
1099                         self.spr[name] = output
1100                     else:
1101                         self.namespace[name].eq(output)
1102                     if name == 'MSR':
1103                         print('msr written', hex(self.msr.value))
1104                 else:
1105                     regnum, is_vec = yield from get_pdecode_idx_out(self.dec2,
1106                                                 name)
1107                     if regnum is None:
1108                         # temporary hack for not having 2nd output
1109                         regnum = yield getattr(self.decoder, name)
1110                         is_vec = False
1111                     if self.is_svp64_mode and pred_dst_zero:
1112                         print('zeroing reg %d %s' % (regnum, str(output)),
1113                                                      is_vec)
1114                         output = SelectableInt(0, 256)
1115                     else:
1116                         print('writing reg %d %s' % (regnum, str(output)),
1117                                                      is_vec)
1118                     if output.bits > 64:
1119                         output = SelectableInt(output.value, 64)
1120                     self.gpr[regnum] = output
1121
1122         # check if it is the SVSTATE.src/dest step that needs incrementing
1123         # this is our Sub-Program-Counter loop from 0 to VL-1
1124         if self.is_svp64_mode:
1125             # XXX twin predication TODO
1126             vl = self.svstate.vl.asint(msb0=True)
1127             mvl = self.svstate.maxvl.asint(msb0=True)
1128             srcstep = self.svstate.srcstep.asint(msb0=True)
1129             dststep = self.svstate.dststep.asint(msb0=True)
1130             sv_ptype = yield self.dec2.dec.op.SV_Ptype
1131             no_out_vec = not (yield self.dec2.no_out_vec)
1132             no_in_vec = not (yield self.dec2.no_in_vec)
1133             print ("    svstate.vl", vl)
1134             print ("    svstate.mvl", mvl)
1135             print ("    svstate.srcstep", srcstep)
1136             print ("    svstate.dststep", dststep)
1137             print ("    no_out_vec", no_out_vec)
1138             print ("    no_in_vec", no_in_vec)
1139             print ("    sv_ptype", sv_ptype, sv_ptype == SVPtype.P2.value)
1140             # check if srcstep needs incrementing by one, stop PC advancing
1141             # svp64 loop can end early if the dest is scalar for single-pred
1142             # but for 2-pred both src/dest have to be checked.
1143             # XXX this might not be true! it may just be LD/ST
1144             if sv_ptype == SVPtype.P2.value:
1145                 svp64_is_vector = (no_out_vec or no_in_vec)
1146             else:
1147                 svp64_is_vector = no_out_vec
1148             if svp64_is_vector and srcstep != vl-1:
1149                 self.svstate.srcstep += SelectableInt(1, 7)
1150                 self.svstate.dststep += SelectableInt(1, 7)
1151                 self.pc.NIA.value = self.pc.CIA.value
1152                 self.namespace['NIA'] = self.pc.NIA
1153                 self.namespace['SVSTATE'] = self.svstate.spr
1154                 print("end of sub-pc call", self.namespace['CIA'],
1155                                      self.namespace['NIA'])
1156                 return # DO NOT allow PC to update whilst Sub-PC loop running
1157             # reset loop to zero
1158             self.svp64_reset_loop()
1159
1160         self.update_pc_next()
1161
1162     def update_pc_next(self):
1163         # UPDATE program counter
1164         self.pc.update(self.namespace, self.is_svp64_mode)
1165         self.svstate.spr = self.namespace['SVSTATE']
1166         print("end of call", self.namespace['CIA'],
1167                              self.namespace['NIA'],
1168                              self.namespace['SVSTATE'])
1169
1170     def svp64_reset_loop(self):
1171         self.svstate.srcstep[0:7] = 0
1172         self.svstate.dststep[0:7] = 0
1173         print ("    svstate.srcstep loop end (PC to update)")
1174         self.pc.update_nia(self.is_svp64_mode)
1175         self.namespace['NIA'] = self.pc.NIA
1176         self.namespace['SVSTATE'] = self.svstate.spr
1177
1178 def inject():
1179     """Decorator factory.
1180
1181     this decorator will "inject" variables into the function's namespace,
1182     from the *dictionary* in self.namespace.  it therefore becomes possible
1183     to make it look like a whole stack of variables which would otherwise
1184     need "self." inserted in front of them (*and* for those variables to be
1185     added to the instance) "appear" in the function.
1186
1187     "self.namespace['SI']" for example becomes accessible as just "SI" but
1188     *only* inside the function, when decorated.
1189     """
1190     def variable_injector(func):
1191         @wraps(func)
1192         def decorator(*args, **kwargs):
1193             try:
1194                 func_globals = func.__globals__  # Python 2.6+
1195             except AttributeError:
1196                 func_globals = func.func_globals  # Earlier versions.
1197
1198             context = args[0].namespace  # variables to be injected
1199             saved_values = func_globals.copy()  # Shallow copy of dict.
1200             func_globals.update(context)
1201             result = func(*args, **kwargs)
1202             print("globals after", func_globals['CIA'], func_globals['NIA'])
1203             print("args[0]", args[0].namespace['CIA'],
1204                   args[0].namespace['NIA'],
1205                   args[0].namespace['SVSTATE'])
1206             args[0].namespace = func_globals
1207             #exec (func.__code__, func_globals)
1208
1209             # finally:
1210             #    func_globals = saved_values  # Undo changes.
1211
1212             return result
1213
1214         return decorator
1215
1216     return variable_injector
1217
1218