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
  26 from soc.decoder.power_enums import SPR as DEC_SPR
  27
  28 from soc.decoder.helpers import exts, gtu, ltu, undefined
  29 from soc.consts import PIb, MSRb  # big-endian (PowerISA versions)
  30 from soc.decoder.power_svp64 import SVP64RM, decode_extra
  31
  32 from soc.decoder.isa.radixmmu import RADIX
  33 from soc.decoder.isa.mem import Mem, swap_order
  34
  35 from collections import namedtuple
  36 import math
  37 import sys
  38
  39 instruction_info = namedtuple('instruction_info',
  40                               'func read_regs uninit_regs write_regs ' +
  41                               'special_regs op_fields form asmregs')
  42
  43 special_sprs = {
  44     'LR': 8,
  45     'CTR': 9,
  46     'TAR': 815,
  47     'XER': 1,
  48     'VRSAVE': 256}
  49
  50
  51 REG_SORT_ORDER = {
  52     # TODO (lkcl): adjust other registers that should be in a particular order
  53     # probably CA, CA32, and CR
  54     "RT": 0,
  55     "RA": 0,
  56     "RB": 0,
  57     "RS": 0,
  58     "CR": 0,
  59     "LR": 0,
  60     "CTR": 0,
  61     "TAR": 0,
  62     "CA": 0,
  63     "CA32": 0,
  64     "MSR": 0,
  65
  66     "overflow": 1,
  67 }
  68
  69
  70 def create_args(reglist, extra=None):
  71     retval = list(OrderedSet(reglist))
  72     retval.sort(key=lambda reg: REG_SORT_ORDER[reg])
  73     if extra is not None:
  74         return [extra] + retval
  75     return retval
  76
  77
  78
  79 class GPR(dict):
  80     def __init__(self, decoder, isacaller, svstate, regfile):
  81         dict.__init__(self)
  82         self.sd = decoder
  83         self.isacaller = isacaller
  84         self.svstate = svstate
  85         for i in range(32):
  86             self[i] = SelectableInt(regfile[i], 64)
  87
  88     def __call__(self, ridx):
  89         return self[ridx]
  90
  91     def set_form(self, form):
  92         self.form = form
  93
  94     def getz(self, rnum):
  95         # rnum = rnum.value # only SelectableInt allowed
  96         print("GPR getzero", rnum)
  97         if rnum == 0:
  98             return SelectableInt(0, 64)
  99         return self[rnum]
 100
 101     def _get_regnum(self, attr):
 102         getform = self.sd.sigforms[self.form]
 103         rnum = getattr(getform, attr)
 104         return rnum
 105
 106     def ___getitem__(self, attr):
 107         """ XXX currently not used
 108         """
 109         rnum = self._get_regnum(attr)
 110         offs = self.svstate.srcstep
 111         print("GPR getitem", attr, rnum, "srcoffs", offs)
 112         return self.regfile[rnum]
 113
 114     def dump(self):
 115         for i in range(0, len(self), 8):
 116             s = []
 117             for j in range(8):
 118                 s.append("%08x" % self[i+j].value)
 119             s = ' '.join(s)
 120             print("reg", "%2d" % i, s)
 121
 122
 123 class PC:
 124     def __init__(self, pc_init=0):
 125         self.CIA = SelectableInt(pc_init, 64)
 126         self.NIA = self.CIA + SelectableInt(4, 64) # only true for v3.0B!
 127
 128     def update_nia(self, is_svp64):
 129         increment = 8 if is_svp64 else 4
 130         self.NIA = self.CIA + SelectableInt(increment, 64)
 131
 132     def update(self, namespace, is_svp64):
 133         """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
 134         """
 135         self.CIA = namespace['NIA'].narrow(64)
 136         self.update_nia(is_svp64)
 137         namespace['CIA'] = self.CIA
 138         namespace['NIA'] = self.NIA
 139
 140
 141 # Simple-V: see https://libre-soc.org/openpower/sv
 142 class SVP64State:
 143     def __init__(self, init=0):
 144         self.spr = SelectableInt(init, 32)
 145         # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
 146         self.maxvl = FieldSelectableInt(self.spr, tuple(range(0,7)))
 147         self.vl = FieldSelectableInt(self.spr, tuple(range(7,14)))
 148         self.srcstep = FieldSelectableInt(self.spr, tuple(range(14,21)))
 149         self.dststep = FieldSelectableInt(self.spr, tuple(range(21,28)))
 150         self.subvl = FieldSelectableInt(self.spr, tuple(range(28,30)))
 151         self.svstep = FieldSelectableInt(self.spr, tuple(range(30,32)))
 152
 153
 154 # SVP64 ReMap field
 155 class SVP64RMFields:
 156     def __init__(self, init=0):
 157         self.spr = SelectableInt(init, 24)
 158         # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
 159         self.mmode = FieldSelectableInt(self.spr, [0])
 160         self.mask = FieldSelectableInt(self.spr, tuple(range(1,4)))
 161         self.elwidth = FieldSelectableInt(self.spr, tuple(range(4,6)))
 162         self.ewsrc = FieldSelectableInt(self.spr, tuple(range(6,8)))
 163         self.subvl = FieldSelectableInt(self.spr, tuple(range(8,10)))
 164         self.extra = FieldSelectableInt(self.spr, tuple(range(10,19)))
 165         self.mode = FieldSelectableInt(self.spr, tuple(range(19,24)))
 166         # these cover the same extra field, split into parts as EXTRA2
 167         self.extra2 = list(range(4))
 168         self.extra2[0] = FieldSelectableInt(self.spr, tuple(range(10,12)))
 169         self.extra2[1] = FieldSelectableInt(self.spr, tuple(range(12,14)))
 170         self.extra2[2] = FieldSelectableInt(self.spr, tuple(range(14,16)))
 171         self.extra2[3] = FieldSelectableInt(self.spr, tuple(range(16,18)))
 172         self.smask = FieldSelectableInt(self.spr, tuple(range(16,19)))
 173         # and here as well, but EXTRA3
 174         self.extra3 = list(range(3))
 175         self.extra3[0] = FieldSelectableInt(self.spr, tuple(range(10,13)))
 176         self.extra3[1] = FieldSelectableInt(self.spr, tuple(range(13,16)))
 177         self.extra3[2] = FieldSelectableInt(self.spr, tuple(range(16,19)))
 178
 179
 180 SVP64RM_MMODE_SIZE = len(SVP64RMFields().mmode.br)
 181 SVP64RM_MASK_SIZE = len(SVP64RMFields().mask.br)
 182 SVP64RM_ELWIDTH_SIZE = len(SVP64RMFields().elwidth.br)
 183 SVP64RM_EWSRC_SIZE = len(SVP64RMFields().ewsrc.br)
 184 SVP64RM_SUBVL_SIZE = len(SVP64RMFields().subvl.br)
 185 SVP64RM_EXTRA2_SPEC_SIZE = len(SVP64RMFields().extra2[0].br)
 186 SVP64RM_EXTRA3_SPEC_SIZE = len(SVP64RMFields().extra3[0].br)
 187 SVP64RM_SMASK_SIZE = len(SVP64RMFields().smask.br)
 188 SVP64RM_MODE_SIZE = len(SVP64RMFields().mode.br)
 189
 190
 191 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
 192 class SVP64PrefixFields:
 193     def __init__(self):
 194         self.insn = SelectableInt(0, 32)
 195         # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
 196         self.major = FieldSelectableInt(self.insn, tuple(range(0,6)))
 197         self.pid = FieldSelectableInt(self.insn, (7, 9)) # must be 0b11
 198         rmfields = [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
 199         self.rm = FieldSelectableInt(self.insn, rmfields)
 200
 201
 202 SV64P_MAJOR_SIZE = len(SVP64PrefixFields().major.br)
 203 SV64P_PID_SIZE = len(SVP64PrefixFields().pid.br)
 204 SV64P_RM_SIZE = len(SVP64PrefixFields().rm.br)
 205
 206
 207 class SPR(dict):
 208     def __init__(self, dec2, initial_sprs={}):
 209         self.sd = dec2
 210         dict.__init__(self)
 211         for key, v in initial_sprs.items():
 212             if isinstance(key, SelectableInt):
 213                 key = key.value
 214             key = special_sprs.get(key, key)
 215             if isinstance(key, int):
 216                 info = spr_dict[key]
 217             else:
 218                 info = spr_byname[key]
 219             if not isinstance(v, SelectableInt):
 220                 v = SelectableInt(v, info.length)
 221             self[key] = v
 222
 223     def __getitem__(self, key):
 224         print("get spr", key)
 225         print("dict", self.items())
 226         # if key in special_sprs get the special spr, otherwise return key
 227         if isinstance(key, SelectableInt):
 228             key = key.value
 229         if isinstance(key, int):
 230             key = spr_dict[key].SPR
 231         key = special_sprs.get(key, key)
 232         if key == 'HSRR0':  # HACK!
 233             key = 'SRR0'
 234         if key == 'HSRR1':  # HACK!
 235             key = 'SRR1'
 236         if key in self:
 237             res = dict.__getitem__(self, key)
 238         else:
 239             if isinstance(key, int):
 240                 info = spr_dict[key]
 241             else:
 242                 info = spr_byname[key]
 243             dict.__setitem__(self, key, SelectableInt(0, info.length))
 244             res = dict.__getitem__(self, key)
 245         print("spr returning", key, res)
 246         return res
 247
 248     def __setitem__(self, key, value):
 249         if isinstance(key, SelectableInt):
 250             key = key.value
 251         if isinstance(key, int):
 252             key = spr_dict[key].SPR
 253             print("spr key", key)
 254         key = special_sprs.get(key, key)
 255         if key == 'HSRR0':  # HACK!
 256             self.__setitem__('SRR0', value)
 257         if key == 'HSRR1':  # HACK!
 258             self.__setitem__('SRR1', value)
 259         print("setting spr", key, value)
 260         dict.__setitem__(self, key, value)
 261
 262     def __call__(self, ridx):
 263         return self[ridx]
 264
 265 def get_pdecode_idx_in(dec2, name):
 266     op = dec2.dec.op
 267     in1_sel = yield op.in1_sel
 268     in2_sel = yield op.in2_sel
 269     in3_sel = yield op.in3_sel
 270     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 271     in1 = yield dec2.e.read_reg1.data
 272     in2 = yield dec2.e.read_reg2.data
 273     in3 = yield dec2.e.read_reg3.data
 274     in1_isvec = yield dec2.in1_isvec
 275     in2_isvec = yield dec2.in2_isvec
 276     in3_isvec = yield dec2.in3_isvec
 277     print ("get_pdecode_idx", in1_sel, In1Sel.RA.value, in1, in1_isvec)
 278     # identify which regnames map to in1/2/3
 279     if name == 'RA':
 280         if (in1_sel == In1Sel.RA.value or
 281             (in1_sel == In1Sel.RA_OR_ZERO.value and in1 != 0)):
 282             return in1, in1_isvec
 283         if in1_sel == In1Sel.RA_OR_ZERO.value:
 284             return in1, in1_isvec
 285     elif name == 'RB':
 286         if in2_sel == In2Sel.RB.value:
 287             return in2, in2_isvec
 288         if in3_sel == In3Sel.RB.value:
 289             return in3, in3_isvec
 290     # XXX TODO, RC doesn't exist yet!
 291     elif name == 'RC':
 292         assert False, "RC does not exist yet"
 293     elif name == 'RS':
 294         if in1_sel == In1Sel.RS.value:
 295             return in1, in1_isvec
 296         if in2_sel == In2Sel.RS.value:
 297             return in2, in2_isvec
 298         if in3_sel == In3Sel.RS.value:
 299             return in3, in3_isvec
 300     return None, False
 301
 302
 303 def get_pdecode_cr_out(dec2, name):
 304     op = dec2.dec.op
 305     out_sel = yield op.cr_out
 306     out_bitfield = yield dec2.dec_cr_out.cr_bitfield.data
 307     sv_cr_out = yield op.sv_cr_out
 308     spec = yield dec2.crout_svdec.spec
 309     sv_override = yield dec2.dec_cr_out.sv_override
 310     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 311     out = yield dec2.e.write_cr.data
 312     o_isvec = yield dec2.o_isvec
 313     print ("get_pdecode_cr_out", out_sel, CROutSel.CR0.value, out, o_isvec)
 314     print ("    sv_cr_out", sv_cr_out)
 315     print ("    cr_bf", out_bitfield)
 316     print ("    spec", spec)
 317     print ("    override", sv_override)
 318     # identify which regnames map to out / o2
 319     if name == 'CR0':
 320         if out_sel == CROutSel.CR0.value:
 321             return out, o_isvec
 322     print ("get_pdecode_idx_out not found", name)
 323     return None, False
 324
 325
 326 def get_pdecode_idx_out(dec2, name):
 327     op = dec2.dec.op
 328     out_sel = yield op.out_sel
 329     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 330     out = yield dec2.e.write_reg.data
 331     o_isvec = yield dec2.o_isvec
 332     print ("get_pdecode_idx_out", out_sel, OutSel.RA.value, out, o_isvec)
 333     # identify which regnames map to out / o2
 334     if name == 'RA':
 335         if out_sel == OutSel.RA.value:
 336             return out, o_isvec
 337     elif name == 'RT':
 338         if out_sel == OutSel.RT.value:
 339             return out, o_isvec
 340     print ("get_pdecode_idx_out not found", name)
 341     return None, False
 342
 343
 344 # XXX TODO
 345 def get_pdecode_idx_out2(dec2, name):
 346     op = dec2.dec.op
 347     print ("TODO: get_pdecode_idx_out2", name)
 348     return None, False
 349
 350
 351 class ISACaller:
 352     # decoder2 - an instance of power_decoder2
 353     # regfile - a list of initial values for the registers
 354     # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
 355     # respect_pc - tracks the program counter.  requires initial_insns
 356     def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
 357                  initial_mem=None, initial_msr=0,
 358                  initial_svstate=0,
 359                  initial_insns=None, respect_pc=False,
 360                  disassembly=None,
 361                  initial_pc=0,
 362                  bigendian=False,
 363                  mmu=False):
 364
 365         self.bigendian = bigendian
 366         self.halted = False
 367         self.is_svp64_mode = False
 368         self.respect_pc = respect_pc
 369         if initial_sprs is None:
 370             initial_sprs = {}
 371         if initial_mem is None:
 372             initial_mem = {}
 373         if initial_insns is None:
 374             initial_insns = {}
 375             assert self.respect_pc == False, "instructions required to honor pc"
 376
 377         print("ISACaller insns", respect_pc, initial_insns, disassembly)
 378         print("ISACaller initial_msr", initial_msr)
 379
 380         # "fake program counter" mode (for unit testing)
 381         self.fake_pc = 0
 382         disasm_start = 0
 383         if not respect_pc:
 384             if isinstance(initial_mem, tuple):
 385                 self.fake_pc = initial_mem[0]
 386                 disasm_start = self.fake_pc
 387         else:
 388             disasm_start = initial_pc
 389
 390         # disassembly: we need this for now (not given from the decoder)
 391         self.disassembly = {}
 392         if disassembly:
 393             for i, code in enumerate(disassembly):
 394                 self.disassembly[i*4 + disasm_start] = code
 395
 396         # set up registers, instruction memory, data memory, PC, SPRs, MSR
 397         self.svp64rm = SVP64RM()
 398         if isinstance(initial_svstate, int):
 399             initial_svstate = SVP64State(initial_svstate)
 400         self.svstate = initial_svstate
 401         self.gpr = GPR(decoder2, self, self.svstate, regfile)
 402         self.spr = SPR(decoder2, initial_sprs) # initialise SPRs before MMU
 403         self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
 404         if mmu:
 405             self.mem = RADIX(self.mem, self)
 406         self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
 407         self.pc = PC()
 408         self.msr = SelectableInt(initial_msr, 64)  # underlying reg
 409
 410         # TODO, needed here:
 411         # FPR (same as GPR except for FP nums)
 412         # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
 413         #            note that mffs, mcrfs, mtfsf "manage" this FPSCR
 414         # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
 415         #         note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
 416         #         -- Done
 417         # 2.3.2 LR   (actually SPR #8) -- Done
 418         # 2.3.3 CTR  (actually SPR #9) -- Done
 419         # 2.3.4 TAR  (actually SPR #815)
 420         # 3.2.2 p45 XER  (actually SPR #1) -- Done
 421         # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
 422
 423         # create CR then allow portions of it to be "selectable" (below)
 424         #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
 425         self.cr = SelectableInt(initial_cr, 64)  # underlying reg
 426         #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
 427
 428         # "undefined", just set to variable-bit-width int (use exts "max")
 429         #self.undefined = SelectableInt(0, 256)  # TODO, not hard-code 256!
 430
 431         self.namespace = {}
 432         self.namespace.update(self.spr)
 433         self.namespace.update({'GPR': self.gpr,
 434                                'MEM': self.mem,
 435                                'SPR': self.spr,
 436                                'memassign': self.memassign,
 437                                'NIA': self.pc.NIA,
 438                                'CIA': self.pc.CIA,
 439                                'CR': self.cr,
 440                                'MSR': self.msr,
 441                                'undefined': undefined,
 442                                'mode_is_64bit': True,
 443                                'SO': XER_bits['SO']
 444                                })
 445
 446         # update pc to requested start point
 447         self.set_pc(initial_pc)
 448
 449         # field-selectable versions of Condition Register TODO check bitranges?
 450         self.crl = []
 451         for i in range(8):
 452             bits = tuple(range(i*4+32, (i+1)*4+32))  # errr... maybe?
 453             _cr = FieldSelectableInt(self.cr, bits)
 454             self.crl.append(_cr)
 455             self.namespace["CR%d" % i] = _cr
 456
 457         self.decoder = decoder2.dec
 458         self.dec2 = decoder2
 459
 460     def TRAP(self, trap_addr=0x700, trap_bit=PIb.TRAP):
 461         print("TRAP:", hex(trap_addr), hex(self.namespace['MSR'].value))
 462         # store CIA(+4?) in SRR0, set NIA to 0x700
 463         # store MSR in SRR1, set MSR to um errr something, have to check spec
 464         self.spr['SRR0'].value = self.pc.CIA.value
 465         self.spr['SRR1'].value = self.namespace['MSR'].value
 466         self.trap_nia = SelectableInt(trap_addr, 64)
 467         self.spr['SRR1'][trap_bit] = 1  # change *copy* of MSR in SRR1
 468
 469         # set exception bits.  TODO: this should, based on the address
 470         # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
 471         # bits appropriately.  however it turns out that *for now* in all
 472         # cases (all trap_addrs) the exact same thing is needed.
 473         self.msr[MSRb.IR] = 0
 474         self.msr[MSRb.DR] = 0
 475         self.msr[MSRb.FE0] = 0
 476         self.msr[MSRb.FE1] = 0
 477         self.msr[MSRb.EE] = 0
 478         self.msr[MSRb.RI] = 0
 479         self.msr[MSRb.SF] = 1
 480         self.msr[MSRb.TM] = 0
 481         self.msr[MSRb.VEC] = 0
 482         self.msr[MSRb.VSX] = 0
 483         self.msr[MSRb.PR] = 0
 484         self.msr[MSRb.FP] = 0
 485         self.msr[MSRb.PMM] = 0
 486         self.msr[MSRb.TEs] = 0
 487         self.msr[MSRb.TEe] = 0
 488         self.msr[MSRb.UND] = 0
 489         self.msr[MSRb.LE] = 1
 490
 491     def memassign(self, ea, sz, val):
 492         self.mem.memassign(ea, sz, val)
 493
 494     def prep_namespace(self, formname, op_fields):
 495         # TODO: get field names from form in decoder*1* (not decoder2)
 496         # decoder2 is hand-created, and decoder1.sigform is auto-generated
 497         # from spec
 498         # then "yield" fields only from op_fields rather than hard-coded
 499         # list, here.
 500         fields = self.decoder.sigforms[formname]
 501         for name in op_fields:
 502             if name == 'spr':
 503                 sig = getattr(fields, name.upper())
 504             else:
 505                 sig = getattr(fields, name)
 506             val = yield sig
 507             # these are all opcode fields involved in index-selection of CR,
 508             # and need to do "standard" arithmetic.  CR[BA+32] for example
 509             # would, if using SelectableInt, only be 5-bit.
 510             if name in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
 511                 self.namespace[name] = val
 512             else:
 513                 self.namespace[name] = SelectableInt(val, sig.width)
 514
 515         self.namespace['XER'] = self.spr['XER']
 516         self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
 517         self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
 518
 519     def handle_carry_(self, inputs, outputs, already_done):
 520         inv_a = yield self.dec2.e.do.invert_in
 521         if inv_a:
 522             inputs[0] = ~inputs[0]
 523
 524         imm_ok = yield self.dec2.e.do.imm_data.ok
 525         if imm_ok:
 526             imm = yield self.dec2.e.do.imm_data.data
 527             inputs.append(SelectableInt(imm, 64))
 528         assert len(outputs) >= 1
 529         print("outputs", repr(outputs))
 530         if isinstance(outputs, list) or isinstance(outputs, tuple):
 531             output = outputs[0]
 532         else:
 533             output = outputs
 534         gts = []
 535         for x in inputs:
 536             print("gt input", x, output)
 537             gt = (gtu(x, output))
 538             gts.append(gt)
 539         print(gts)
 540         cy = 1 if any(gts) else 0
 541         print("CA", cy, gts)
 542         if not (1 & already_done):
 543             self.spr['XER'][XER_bits['CA']] = cy
 544
 545         print("inputs", already_done, inputs)
 546         # 32 bit carry
 547         # ARGH... different for OP_ADD... *sigh*...
 548         op = yield self.dec2.e.do.insn_type
 549         if op == MicrOp.OP_ADD.value:
 550             res32 = (output.value & (1 << 32)) != 0
 551             a32 = (inputs[0].value & (1 << 32)) != 0
 552             if len(inputs) >= 2:
 553                 b32 = (inputs[1].value & (1 << 32)) != 0
 554             else:
 555                 b32 = False
 556             cy32 = res32 ^ a32 ^ b32
 557             print("CA32 ADD", cy32)
 558         else:
 559             gts = []
 560             for x in inputs:
 561                 print("input", x, output)
 562                 print("     x[32:64]", x, x[32:64])
 563                 print("     o[32:64]", output, output[32:64])
 564                 gt = (gtu(x[32:64], output[32:64])) == SelectableInt(1, 1)
 565                 gts.append(gt)
 566             cy32 = 1 if any(gts) else 0
 567             print("CA32", cy32, gts)
 568         if not (2 & already_done):
 569             self.spr['XER'][XER_bits['CA32']] = cy32
 570
 571     def handle_overflow(self, inputs, outputs, div_overflow):
 572         if hasattr(self.dec2.e.do, "invert_in"):
 573             inv_a = yield self.dec2.e.do.invert_in
 574             if inv_a:
 575                 inputs[0] = ~inputs[0]
 576
 577         imm_ok = yield self.dec2.e.do.imm_data.ok
 578         if imm_ok:
 579             imm = yield self.dec2.e.do.imm_data.data
 580             inputs.append(SelectableInt(imm, 64))
 581         assert len(outputs) >= 1
 582         print("handle_overflow", inputs, outputs, div_overflow)
 583         if len(inputs) < 2 and div_overflow is None:
 584             return
 585
 586         # div overflow is different: it's returned by the pseudo-code
 587         # because it's more complex than can be done by analysing the output
 588         if div_overflow is not None:
 589             ov, ov32 = div_overflow, div_overflow
 590         # arithmetic overflow can be done by analysing the input and output
 591         elif len(inputs) >= 2:
 592             output = outputs[0]
 593
 594             # OV (64-bit)
 595             input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
 596             output_sgn = exts(output.value, output.bits) < 0
 597             ov = 1 if input_sgn[0] == input_sgn[1] and \
 598                 output_sgn != input_sgn[0] else 0
 599
 600             # OV (32-bit)
 601             input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
 602             output32_sgn = exts(output.value, 32) < 0
 603             ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
 604                 output32_sgn != input32_sgn[0] else 0
 605
 606         self.spr['XER'][XER_bits['OV']] = ov
 607         self.spr['XER'][XER_bits['OV32']] = ov32
 608         so = self.spr['XER'][XER_bits['SO']]
 609         so = so | ov
 610         self.spr['XER'][XER_bits['SO']] = so
 611
 612     def handle_comparison(self, outputs, cr_idx=0):
 613         out = outputs[0]
 614         assert isinstance(out, SelectableInt), \
 615             "out zero not a SelectableInt %s" % repr(outputs)
 616         print("handle_comparison", out.bits, hex(out.value))
 617         # TODO - XXX *processor* in 32-bit mode
 618         # https://bugs.libre-soc.org/show_bug.cgi?id=424
 619         # if is_32bit:
 620         #    o32 = exts(out.value, 32)
 621         #    print ("handle_comparison exts 32 bit", hex(o32))
 622         out = exts(out.value, out.bits)
 623         print("handle_comparison exts", hex(out))
 624         zero = SelectableInt(out == 0, 1)
 625         positive = SelectableInt(out > 0, 1)
 626         negative = SelectableInt(out < 0, 1)
 627         SO = self.spr['XER'][XER_bits['SO']]
 628         print("handle_comparison SO", SO)
 629         cr_field = selectconcat(negative, positive, zero, SO)
 630         self.crl[cr_idx].eq(cr_field)
 631
 632     def set_pc(self, pc_val):
 633         self.namespace['NIA'] = SelectableInt(pc_val, 64)
 634         self.pc.update(self.namespace, self.is_svp64_mode)
 635
 636     def setup_one(self):
 637         """set up one instruction
 638         """
 639         if self.respect_pc:
 640             pc = self.pc.CIA.value
 641         else:
 642             pc = self.fake_pc
 643         self._pc = pc
 644         ins = self.imem.ld(pc, 4, False, True)
 645         if ins is None:
 646             raise KeyError("no instruction at 0x%x" % pc)
 647         print("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
 648         print("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
 649
 650         yield self.dec2.sv_rm.eq(0)
 651         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
 652         yield self.dec2.dec.bigendian.eq(self.bigendian)
 653         yield self.dec2.state.msr.eq(self.msr.value)
 654         yield self.dec2.state.pc.eq(pc)
 655         yield self.dec2.state.svstate.eq(self.svstate.spr.value)
 656
 657         # SVP64.  first, check if the opcode is EXT001, and SVP64 id bits set
 658         yield Settle()
 659         opcode = yield self.dec2.dec.opcode_in
 660         pfx = SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
 661         pfx.insn.value = opcode
 662         major = pfx.major.asint(msb0=True) # MSB0 inversion
 663         print ("prefix test: opcode:", major, bin(major),
 664                 pfx.insn[7] == 0b1, pfx.insn[9] == 0b1)
 665         self.is_svp64_mode = ((major == 0b000001) and
 666                               pfx.insn[7].value == 0b1 and
 667                               pfx.insn[9].value == 0b1)
 668         self.pc.update_nia(self.is_svp64_mode)
 669         self.namespace['NIA'] = self.pc.NIA
 670         if not self.is_svp64_mode:
 671             return
 672
 673         # in SVP64 mode.  decode/print out svp64 prefix, get v3.0B instruction
 674         print ("svp64.rm", bin(pfx.rm.asint(msb0=True)))
 675         print ("    svstate.vl", self.svstate.vl.asint(msb0=True))
 676         print ("    svstate.mvl", self.svstate.maxvl.asint(msb0=True))
 677         sv_rm = pfx.rm.asint(msb0=True)
 678         ins = self.imem.ld(pc+4, 4, False, True)
 679         print("     svsetup: 0x%x 0x%x %s" % (pc+4, ins & 0xffffffff, bin(ins)))
 680         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff) # v3.0B suffix
 681         yield self.dec2.sv_rm.eq(sv_rm)                        # svp64 prefix
 682         yield Settle()
 683
 684     def execute_one(self):
 685         """execute one instruction
 686         """
 687         # get the disassembly code for this instruction
 688         if self.is_svp64_mode:
 689             code = self.disassembly[self._pc+4]
 690             print("    svp64 sim-execute", hex(self._pc), code)
 691         else:
 692             code = self.disassembly[self._pc]
 693             print("sim-execute", hex(self._pc), code)
 694         opname = code.split(' ')[0]
 695         yield from self.call(opname)
 696
 697         # don't use this except in special circumstances
 698         if not self.respect_pc:
 699             self.fake_pc += 4
 700
 701         print("execute one, CIA NIA", self.pc.CIA.value, self.pc.NIA.value)
 702
 703     def get_assembly_name(self):
 704         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 705         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 706         dec_insn = yield self.dec2.e.do.insn
 707         asmcode = yield self.dec2.dec.op.asmcode
 708         print("get assembly name asmcode", asmcode, hex(dec_insn))
 709         asmop = insns.get(asmcode, None)
 710         int_op = yield self.dec2.dec.op.internal_op
 711
 712         # sigh reconstruct the assembly instruction name
 713         if hasattr(self.dec2.e.do, "oe"):
 714             ov_en = yield self.dec2.e.do.oe.oe
 715             ov_ok = yield self.dec2.e.do.oe.ok
 716         else:
 717             ov_en = False
 718             ov_ok = False
 719         if hasattr(self.dec2.e.do, "rc"):
 720             rc_en = yield self.dec2.e.do.rc.rc
 721             rc_ok = yield self.dec2.e.do.rc.ok
 722         else:
 723             rc_en = False
 724             rc_ok = False
 725         # grrrr have to special-case MUL op (see DecodeOE)
 726         print("ov %d en %d rc %d en %d op %d" %
 727               (ov_ok, ov_en, rc_ok, rc_en, int_op))
 728         if int_op in [MicrOp.OP_MUL_H64.value, MicrOp.OP_MUL_H32.value]:
 729             print("mul op")
 730             if rc_en & rc_ok:
 731                 asmop += "."
 732         else:
 733             if not asmop.endswith("."):  # don't add "." to "andis."
 734                 if rc_en & rc_ok:
 735                     asmop += "."
 736         if hasattr(self.dec2.e.do, "lk"):
 737             lk = yield self.dec2.e.do.lk
 738             if lk:
 739                 asmop += "l"
 740         print("int_op", int_op)
 741         if int_op in [MicrOp.OP_B.value, MicrOp.OP_BC.value]:
 742             AA = yield self.dec2.dec.fields.FormI.AA[0:-1]
 743             print("AA", AA)
 744             if AA:
 745                 asmop += "a"
 746         spr_msb = yield from self.get_spr_msb()
 747         if int_op == MicrOp.OP_MFCR.value:
 748             if spr_msb:
 749                 asmop = 'mfocrf'
 750             else:
 751                 asmop = 'mfcr'
 752         # XXX TODO: for whatever weird reason this doesn't work
 753         # https://bugs.libre-soc.org/show_bug.cgi?id=390
 754         if int_op == MicrOp.OP_MTCRF.value:
 755             if spr_msb:
 756                 asmop = 'mtocrf'
 757             else:
 758                 asmop = 'mtcrf'
 759         return asmop
 760
 761     def get_spr_msb(self):
 762         dec_insn = yield self.dec2.e.do.insn
 763         return dec_insn & (1 << 20) != 0  # sigh - XFF.spr[-1]?
 764
 765     def call(self, name):
 766         """call(opcode) - the primary execution point for instructions
 767         """
 768         name = name.strip()  # remove spaces if not already done so
 769         if self.halted:
 770             print("halted - not executing", name)
 771             return
 772
 773         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 774         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 775         asmop = yield from self.get_assembly_name()
 776         print("call", name, asmop)
 777
 778         # check privileged
 779         int_op = yield self.dec2.dec.op.internal_op
 780         spr_msb = yield from self.get_spr_msb()
 781
 782         instr_is_privileged = False
 783         if int_op in [MicrOp.OP_ATTN.value,
 784                       MicrOp.OP_MFMSR.value,
 785                       MicrOp.OP_MTMSR.value,
 786                       MicrOp.OP_MTMSRD.value,
 787                       # TODO: OP_TLBIE
 788                       MicrOp.OP_RFID.value]:
 789             instr_is_privileged = True
 790         if int_op in [MicrOp.OP_MFSPR.value,
 791                       MicrOp.OP_MTSPR.value] and spr_msb:
 792             instr_is_privileged = True
 793
 794         print("is priv", instr_is_privileged, hex(self.msr.value),
 795               self.msr[MSRb.PR])
 796         # check MSR priv bit and whether op is privileged: if so, throw trap
 797         if instr_is_privileged and self.msr[MSRb.PR] == 1:
 798             self.TRAP(0x700, PIb.PRIV)
 799             self.namespace['NIA'] = self.trap_nia
 800             self.pc.update(self.namespace, self.is_svp64_mode)
 801             return
 802
 803         # check halted condition
 804         if name == 'attn':
 805             self.halted = True
 806             return
 807
 808         # check illegal instruction
 809         illegal = False
 810         if name not in ['mtcrf', 'mtocrf']:
 811             illegal = name != asmop
 812
 813         if illegal:
 814             print("illegal", name, asmop)
 815             self.TRAP(0x700, PIb.ILLEG)
 816             self.namespace['NIA'] = self.trap_nia
 817             self.pc.update(self.namespace, self.is_svp64_mode)
 818             print("name %s != %s - calling ILLEGAL trap, PC: %x" %
 819                   (name, asmop, self.pc.CIA.value))
 820             return
 821
 822         info = self.instrs[name]
 823         yield from self.prep_namespace(info.form, info.op_fields)
 824
 825         # preserve order of register names
 826         input_names = create_args(list(info.read_regs) +
 827                                   list(info.uninit_regs))
 828         print(input_names)
 829
 830         # get SVP64 entry for the current instruction
 831         sv_rm = self.svp64rm.instrs.get(name)
 832         if sv_rm is not None:
 833             dest_cr, src_cr, src_byname, dest_byname = decode_extra(sv_rm)
 834         else:
 835             dest_cr, src_cr, src_byname, dest_byname = False, False, {}, {}
 836         print ("sv rm", sv_rm, dest_cr, src_cr, src_byname, dest_byname)
 837
 838         # get SVSTATE srcstep.  TODO: dststep (twin predication)
 839         srcstep = self.svstate.srcstep.asint(msb0=True)
 840         vl = self.svstate.vl.asint(msb0=True)
 841         mvl = self.svstate.maxvl.asint(msb0=True)
 842
 843         # VL=0 in SVP64 mode means "do nothing: skip instruction"
 844         if self.is_svp64_mode and vl == 0:
 845             self.pc.update(self.namespace, self.is_svp64_mode)
 846             print("end of call", self.namespace['CIA'], self.namespace['NIA'])
 847             return
 848
 849         # main input registers (RT, RA ...)
 850         inputs = []
 851         for name in input_names:
 852             # using PowerDecoder2, first, find the decoder index.
 853             # (mapping name RA RB RC RS to in1, in2, in3)
 854             regnum, is_vec = yield from get_pdecode_idx_in(self.dec2, name)
 855             if regnum is None:
 856                 # doing this is not part of svp64, it's because output
 857                 # registers, to be modified, need to be in the namespace.
 858                 regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, name)
 859             # here's where we go "vector".  TODO: zero-testing (RA_IS_ZERO)
 860             # XXX already done by PowerDecoder2, now
 861             #if is_vec:
 862             #   regnum += srcstep # TODO, elwidth overrides
 863
 864             # in case getting the register number is needed, _RA, _RB
 865             regname = "_" + name
 866             self.namespace[regname] = regnum
 867             print('reading reg %s %d' % (name, regnum), is_vec)
 868             reg_val = self.gpr(regnum)
 869             inputs.append(reg_val)
 870
 871         # "special" registers
 872         for special in info.special_regs:
 873             if special in special_sprs:
 874                 inputs.append(self.spr[special])
 875             else:
 876                 inputs.append(self.namespace[special])
 877
 878         # clear trap (trap) NIA
 879         self.trap_nia = None
 880
 881         print("inputs", inputs)
 882         results = info.func(self, *inputs)
 883         print("results", results)
 884
 885         # "inject" decorator takes namespace from function locals: we need to
 886         # overwrite NIA being overwritten (sigh)
 887         if self.trap_nia is not None:
 888             self.namespace['NIA'] = self.trap_nia
 889
 890         print("after func", self.namespace['CIA'], self.namespace['NIA'])
 891
 892         # detect if CA/CA32 already in outputs (sra*, basically)
 893         already_done = 0
 894         if info.write_regs:
 895             output_names = create_args(info.write_regs)
 896             for name in output_names:
 897                 if name == 'CA':
 898                     already_done |= 1
 899                 if name == 'CA32':
 900                     already_done |= 2
 901
 902         print("carry already done?", bin(already_done))
 903         if hasattr(self.dec2.e.do, "output_carry"):
 904             carry_en = yield self.dec2.e.do.output_carry
 905         else:
 906             carry_en = False
 907         if carry_en:
 908             yield from self.handle_carry_(inputs, results, already_done)
 909
 910         # detect if overflow was in return result
 911         overflow = None
 912         if info.write_regs:
 913             for name, output in zip(output_names, results):
 914                 if name == 'overflow':
 915                     overflow = output
 916
 917         if hasattr(self.dec2.e.do, "oe"):
 918             ov_en = yield self.dec2.e.do.oe.oe
 919             ov_ok = yield self.dec2.e.do.oe.ok
 920         else:
 921             ov_en = False
 922             ov_ok = False
 923         print("internal overflow", overflow, ov_en, ov_ok)
 924         if ov_en & ov_ok:
 925             yield from self.handle_overflow(inputs, results, overflow)
 926
 927         if hasattr(self.dec2.e.do, "rc"):
 928             rc_en = yield self.dec2.e.do.rc.rc
 929         else:
 930             rc_en = False
 931         if rc_en:
 932             regnum, is_vec = yield from get_pdecode_cr_out(self.dec2, "CR0")
 933             self.handle_comparison(results, regnum)
 934
 935         # any modified return results?
 936         if info.write_regs:
 937             for name, output in zip(output_names, results):
 938                 if name == 'overflow':  # ignore, done already (above)
 939                     continue
 940                 if isinstance(output, int):
 941                     output = SelectableInt(output, 256)
 942                 if name in ['CA', 'CA32']:
 943                     if carry_en:
 944                         print("writing %s to XER" % name, output)
 945                         self.spr['XER'][XER_bits[name]] = output.value
 946                     else:
 947                         print("NOT writing %s to XER" % name, output)
 948                 elif name in info.special_regs:
 949                     print('writing special %s' % name, output, special_sprs)
 950                     if name in special_sprs:
 951                         self.spr[name] = output
 952                     else:
 953                         self.namespace[name].eq(output)
 954                     if name == 'MSR':
 955                         print('msr written', hex(self.msr.value))
 956                 else:
 957                     regnum, is_vec = yield from get_pdecode_idx_out(self.dec2,
 958                                                 name)
 959                     if regnum is None:
 960                         # temporary hack for not having 2nd output
 961                         regnum = yield getattr(self.decoder, name)
 962                         is_vec = False
 963                     print('writing reg %d %s' % (regnum, str(output)), is_vec)
 964                     if output.bits > 64:
 965                         output = SelectableInt(output.value, 64)
 966                     self.gpr[regnum] = output
 967
 968         # check if it is the SVSTATE.src/dest step that needs incrementing
 969         # this is our Sub-Program-Counter loop from 0 to VL-1
 970         if self.is_svp64_mode:
 971             # XXX twin predication TODO
 972             vl = self.svstate.vl.asint(msb0=True)
 973             mvl = self.svstate.maxvl.asint(msb0=True)
 974             srcstep = self.svstate.srcstep.asint(msb0=True)
 975             print ("    svstate.vl", vl)
 976             print ("    svstate.mvl", mvl)
 977             print ("    svstate.srcstep", srcstep)
 978             # check if srcstep needs incrementing by one, stop PC advancing
 979             # svp64 loop can end early if the dest is scalar
 980             svp64_dest_vector = not (yield self.dec2.no_out_vec)
 981             if svp64_dest_vector and srcstep != vl-1:
 982                 self.svstate.srcstep += SelectableInt(1, 7)
 983                 self.pc.NIA.value = self.pc.CIA.value
 984                 self.namespace['NIA'] = self.pc.NIA
 985                 print("end of sub-pc call", self.namespace['CIA'],
 986                                      self.namespace['NIA'])
 987                 return # DO NOT allow PC to update whilst Sub-PC loop running
 988             # reset to zero
 989             self.svstate.srcstep[0:7] = 0
 990             print ("    svstate.srcstep loop end (PC to update)")
 991             self.pc.update_nia(self.is_svp64_mode)
 992             self.namespace['NIA'] = self.pc.NIA
 993
 994         # UPDATE program counter
 995         self.pc.update(self.namespace, self.is_svp64_mode)
 996         print("end of call", self.namespace['CIA'], self.namespace['NIA'])
 997
 998
 999 def inject():
1000     """Decorator factory.
1001
1002     this decorator will "inject" variables into the function's namespace,
1003     from the *dictionary* in self.namespace.  it therefore becomes possible
1004     to make it look like a whole stack of variables which would otherwise
1005     need "self." inserted in front of them (*and* for those variables to be
1006     added to the instance) "appear" in the function.
1007
1008     "self.namespace['SI']" for example becomes accessible as just "SI" but
1009     *only* inside the function, when decorated.
1010     """
1011     def variable_injector(func):
1012         @wraps(func)
1013         def decorator(*args, **kwargs):
1014             try:
1015                 func_globals = func.__globals__  # Python 2.6+
1016             except AttributeError:
1017                 func_globals = func.func_globals  # Earlier versions.
1018
1019             context = args[0].namespace  # variables to be injected
1020             saved_values = func_globals.copy()  # Shallow copy of dict.
1021             func_globals.update(context)
1022             result = func(*args, **kwargs)
1023             print("globals after", func_globals['CIA'], func_globals['NIA'])
1024             print("args[0]", args[0].namespace['CIA'],
1025                   args[0].namespace['NIA'])
1026             args[0].namespace = func_globals
1027             #exec (func.__code__, func_globals)
1028
1029             # finally:
1030             #    func_globals = saved_values  # Undo changes.
1031
1032             return result
1033
1034         return decorator
1035
1036     return variable_injector
1037
1038