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 from soc.decoder.helpers import exts, gtu, ltu, undefined
  26 from soc.consts import PIb, MSRb  # big-endian (PowerISA versions)
  27 from soc.decoder.power_svp64 import SVP64RM, decode_extra
  28
  29 from collections import namedtuple
  30 import math
  31 import sys
  32
  33 instruction_info = namedtuple('instruction_info',
  34                               'func read_regs uninit_regs write_regs ' +
  35                               'special_regs op_fields form asmregs')
  36
  37 special_sprs = {
  38     'LR': 8,
  39     'CTR': 9,
  40     'TAR': 815,
  41     'XER': 1,
  42     'VRSAVE': 256}
  43
  44
  45 def swap_order(x, nbytes):
  46     x = x.to_bytes(nbytes, byteorder='little')
  47     x = int.from_bytes(x, byteorder='big', signed=False)
  48     return x
  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 class Mem:
  79
  80     def __init__(self, row_bytes=8, initial_mem=None):
  81         self.mem = {}
  82         self.bytes_per_word = row_bytes
  83         self.word_log2 = math.ceil(math.log2(row_bytes))
  84         print("Sim-Mem", initial_mem, self.bytes_per_word, self.word_log2)
  85         if not initial_mem:
  86             return
  87
  88         # different types of memory data structures recognised (for convenience)
  89         if isinstance(initial_mem, list):
  90             initial_mem = (0, initial_mem)
  91         if isinstance(initial_mem, tuple):
  92             startaddr, mem = initial_mem
  93             initial_mem = {}
  94             for i, val in enumerate(mem):
  95                 initial_mem[startaddr + row_bytes*i] = (val, row_bytes)
  96
  97         for addr, (val, width) in initial_mem.items():
  98             #val = swap_order(val, width)
  99             self.st(addr, val, width, swap=False)
 100
 101     def _get_shifter_mask(self, wid, remainder):
 102         shifter = ((self.bytes_per_word - wid) - remainder) * \
 103             8  # bits per byte
 104         # XXX https://bugs.libre-soc.org/show_bug.cgi?id=377
 105         # BE/LE mode?
 106         shifter = remainder * 8
 107         mask = (1 << (wid * 8)) - 1
 108         print("width,rem,shift,mask", wid, remainder, hex(shifter), hex(mask))
 109         return shifter, mask
 110
 111     # TODO: Implement ld/st of lesser width
 112     def ld(self, address, width=8, swap=True, check_in_mem=False):
 113         print("ld from addr 0x{:x} width {:d}".format(address, width))
 114         remainder = address & (self.bytes_per_word - 1)
 115         address = address >> self.word_log2
 116         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 117         if address in self.mem:
 118             val = self.mem[address]
 119         elif check_in_mem:
 120             return None
 121         else:
 122             val = 0
 123         print("mem @ 0x{:x} rem {:d} : 0x{:x}".format(address, remainder, val))
 124
 125         if width != self.bytes_per_word:
 126             shifter, mask = self._get_shifter_mask(width, remainder)
 127             print("masking", hex(val), hex(mask << shifter), shifter)
 128             val = val & (mask << shifter)
 129             val >>= shifter
 130         if swap:
 131             val = swap_order(val, width)
 132         print("Read 0x{:x} from addr 0x{:x}".format(val, address))
 133         return val
 134
 135     def st(self, addr, v, width=8, swap=True):
 136         staddr = addr
 137         remainder = addr & (self.bytes_per_word - 1)
 138         addr = addr >> self.word_log2
 139         print("Writing 0x{:x} to ST 0x{:x} "
 140               "memaddr 0x{:x}/{:x}".format(v, staddr, addr, remainder, swap))
 141         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 142         if swap:
 143             v = swap_order(v, width)
 144         if width != self.bytes_per_word:
 145             if addr in self.mem:
 146                 val = self.mem[addr]
 147             else:
 148                 val = 0
 149             shifter, mask = self._get_shifter_mask(width, remainder)
 150             val &= ~(mask << shifter)
 151             val |= v << shifter
 152             self.mem[addr] = val
 153         else:
 154             self.mem[addr] = v
 155         print("mem @ 0x{:x}: 0x{:x}".format(addr, self.mem[addr]))
 156
 157     def __call__(self, addr, sz):
 158         val = self.ld(addr.value, sz, swap=False)
 159         print("memread", addr, sz, val)
 160         return SelectableInt(val, sz*8)
 161
 162     def memassign(self, addr, sz, val):
 163         print("memassign", addr, sz, val)
 164         self.st(addr.value, val.value, sz, swap=False)
 165
 166
 167 class GPR(dict):
 168     def __init__(self, decoder, isacaller, svstate, regfile):
 169         dict.__init__(self)
 170         self.sd = decoder
 171         self.isacaller = isacaller
 172         self.svstate = svstate
 173         for i in range(32):
 174             self[i] = SelectableInt(regfile[i], 64)
 175
 176     def __call__(self, ridx):
 177         return self[ridx]
 178
 179     def set_form(self, form):
 180         self.form = form
 181
 182     def getz(self, rnum):
 183         # rnum = rnum.value # only SelectableInt allowed
 184         print("GPR getzero", rnum)
 185         if rnum == 0:
 186             return SelectableInt(0, 64)
 187         return self[rnum]
 188
 189     def _get_regnum(self, attr):
 190         getform = self.sd.sigforms[self.form]
 191         rnum = getattr(getform, attr)
 192         return rnum
 193
 194     def ___getitem__(self, attr):
 195         """ XXX currently not used
 196         """
 197         rnum = self._get_regnum(attr)
 198         offs = self.svstate.srcstep
 199         print("GPR getitem", attr, rnum, "srcoffs", offs)
 200         return self.regfile[rnum]
 201
 202     def dump(self):
 203         for i in range(0, len(self), 8):
 204             s = []
 205             for j in range(8):
 206                 s.append("%08x" % self[i+j].value)
 207             s = ' '.join(s)
 208             print("reg", "%2d" % i, s)
 209
 210
 211 class PC:
 212     def __init__(self, pc_init=0):
 213         self.CIA = SelectableInt(pc_init, 64)
 214         self.NIA = self.CIA + SelectableInt(4, 64) # only true for v3.0B!
 215
 216     def update_nia(self, is_svp64):
 217         increment = 8 if is_svp64 else 4
 218         self.NIA = self.CIA + SelectableInt(increment, 64)
 219
 220     def update(self, namespace, is_svp64):
 221         """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
 222         """
 223         self.CIA = namespace['NIA'].narrow(64)
 224         self.update_nia(is_svp64)
 225         namespace['CIA'] = self.CIA
 226         namespace['NIA'] = self.NIA
 227
 228
 229 # Simple-V: see https://libre-soc.org/openpower/sv
 230 class SVP64State:
 231     def __init__(self, init=0):
 232         self.spr = SelectableInt(init, 32)
 233         # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
 234         self.maxvl = FieldSelectableInt(self.spr, tuple(range(0,7)))
 235         self.vl = FieldSelectableInt(self.spr, tuple(range(7,14)))
 236         self.srcstep = FieldSelectableInt(self.spr, tuple(range(14,21)))
 237         self.dststep = FieldSelectableInt(self.spr, tuple(range(21,28)))
 238         self.subvl = FieldSelectableInt(self.spr, tuple(range(28,30)))
 239         self.svstep = FieldSelectableInt(self.spr, tuple(range(30,32)))
 240
 241
 242 # SVP64 ReMap field
 243 class SVP64RMFields:
 244     def __init__(self, init=0):
 245         self.spr = SelectableInt(init, 24)
 246         # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
 247         self.mmode = FieldSelectableInt(self.spr, [0])
 248         self.mask = FieldSelectableInt(self.spr, tuple(range(1,4)))
 249         self.elwidth = FieldSelectableInt(self.spr, tuple(range(4,6)))
 250         self.ewsrc = FieldSelectableInt(self.spr, tuple(range(6,8)))
 251         self.subvl = FieldSelectableInt(self.spr, tuple(range(8,10)))
 252         self.extra = FieldSelectableInt(self.spr, tuple(range(10,19)))
 253         self.mode = FieldSelectableInt(self.spr, tuple(range(19,24)))
 254
 255
 256 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
 257 class SVP64PrefixFields:
 258     def __init__(self):
 259         self.insn = SelectableInt(0, 32)
 260         # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
 261         self.major = FieldSelectableInt(self.insn, tuple(range(0,6)))
 262         self.pid = FieldSelectableInt(self.insn, (7, 9)) # must be 0b11
 263         rmfields = [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
 264         self.rm = FieldSelectableInt(self.insn, rmfields)
 265
 266
 267 class SPR(dict):
 268     def __init__(self, dec2, initial_sprs={}):
 269         self.sd = dec2
 270         dict.__init__(self)
 271         for key, v in initial_sprs.items():
 272             if isinstance(key, SelectableInt):
 273                 key = key.value
 274             key = special_sprs.get(key, key)
 275             if isinstance(key, int):
 276                 info = spr_dict[key]
 277             else:
 278                 info = spr_byname[key]
 279             if not isinstance(v, SelectableInt):
 280                 v = SelectableInt(v, info.length)
 281             self[key] = v
 282
 283     def __getitem__(self, key):
 284         print("get spr", key)
 285         print("dict", self.items())
 286         # if key in special_sprs get the special spr, otherwise return key
 287         if isinstance(key, SelectableInt):
 288             key = key.value
 289         if isinstance(key, int):
 290             key = spr_dict[key].SPR
 291         key = special_sprs.get(key, key)
 292         if key == 'HSRR0':  # HACK!
 293             key = 'SRR0'
 294         if key == 'HSRR1':  # HACK!
 295             key = 'SRR1'
 296         if key in self:
 297             res = dict.__getitem__(self, key)
 298         else:
 299             if isinstance(key, int):
 300                 info = spr_dict[key]
 301             else:
 302                 info = spr_byname[key]
 303             dict.__setitem__(self, key, SelectableInt(0, info.length))
 304             res = dict.__getitem__(self, key)
 305         print("spr returning", key, res)
 306         return res
 307
 308     def __setitem__(self, key, value):
 309         if isinstance(key, SelectableInt):
 310             key = key.value
 311         if isinstance(key, int):
 312             key = spr_dict[key].SPR
 313             print("spr key", key)
 314         key = special_sprs.get(key, key)
 315         if key == 'HSRR0':  # HACK!
 316             self.__setitem__('SRR0', value)
 317         if key == 'HSRR1':  # HACK!
 318             self.__setitem__('SRR1', value)
 319         print("setting spr", key, value)
 320         dict.__setitem__(self, key, value)
 321
 322     def __call__(self, ridx):
 323         return self[ridx]
 324
 325 def get_pdecode_idx_in(dec2, name):
 326     op = dec2.dec.op
 327     in1_sel = yield op.in1_sel
 328     in2_sel = yield op.in2_sel
 329     in3_sel = yield op.in3_sel
 330     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 331     in1 = yield dec2.e.read_reg1.data
 332     in2 = yield dec2.e.read_reg2.data
 333     in3 = yield dec2.e.read_reg3.data
 334     in1_isvec = yield dec2.in1_isvec
 335     in2_isvec = yield dec2.in2_isvec
 336     in3_isvec = yield dec2.in3_isvec
 337     print ("get_pdecode_idx", in1_sel, In1Sel.RA.value, in1, in1_isvec)
 338     # identify which regnames map to in1/2/3
 339     if name == 'RA':
 340         if (in1_sel == In1Sel.RA.value or
 341             (in1_sel == In1Sel.RA_OR_ZERO.value and in1 != 0)):
 342             return in1, in1_isvec
 343         if in1_sel == In1Sel.RA_OR_ZERO.value:
 344             return in1, in1_isvec
 345     elif name == 'RB':
 346         if in2_sel == In2Sel.RB.value:
 347             return in2, in2_isvec
 348         if in3_sel == In3Sel.RB.value:
 349             return in3, in3_isvec
 350     # XXX TODO, RC doesn't exist yet!
 351     elif name == 'RC':
 352         assert False, "RC does not exist yet"
 353     elif name == 'RS':
 354         if in1_sel == In1Sel.RS.value:
 355             return in1, in1_isvec
 356         if in2_sel == In2Sel.RS.value:
 357             return in2, in2_isvec
 358         if in3_sel == In3Sel.RS.value:
 359             return in3, in3_isvec
 360     return None, False
 361
 362
 363 def get_pdecode_cr_out(dec2, name):
 364     op = dec2.dec.op
 365     out_sel = yield op.cr_out
 366     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 367     out = yield dec2.e.write_cr.data
 368     o_isvec = yield dec2.o_isvec
 369     print ("get_pdecode_cr_out", out_sel, CROutSel.CR0.value, out, o_isvec)
 370     # identify which regnames map to out / o2
 371     if name == 'CR0':
 372         if out_sel == CROutSel.CR0.value:
 373             return out, o_isvec
 374     print ("get_pdecode_idx_out not found", name)
 375     return None, False
 376
 377
 378 def get_pdecode_idx_out(dec2, name):
 379     op = dec2.dec.op
 380     out_sel = yield op.out_sel
 381     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 382     out = yield dec2.e.write_reg.data
 383     o_isvec = yield dec2.o_isvec
 384     print ("get_pdecode_idx_out", out_sel, OutSel.RA.value, out, o_isvec)
 385     # identify which regnames map to out / o2
 386     if name == 'RA':
 387         if out_sel == OutSel.RA.value:
 388             return out, o_isvec
 389     elif name == 'RT':
 390         if out_sel == OutSel.RT.value:
 391             return out, o_isvec
 392     print ("get_pdecode_idx_out not found", name)
 393     return None, False
 394
 395
 396 # XXX TODO
 397 def get_pdecode_idx_out2(dec2, name):
 398     op = dec2.dec.op
 399     print ("TODO: get_pdecode_idx_out2", name)
 400     return None, False
 401
 402
 403 class ISACaller:
 404     # decoder2 - an instance of power_decoder2
 405     # regfile - a list of initial values for the registers
 406     # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
 407     # respect_pc - tracks the program counter.  requires initial_insns
 408     def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
 409                  initial_mem=None, initial_msr=0,
 410                  initial_svstate=0,
 411                  initial_insns=None, respect_pc=False,
 412                  disassembly=None,
 413                  initial_pc=0,
 414                  bigendian=False):
 415
 416         self.bigendian = bigendian
 417         self.halted = False
 418         self.is_svp64_mode = False
 419         self.respect_pc = respect_pc
 420         if initial_sprs is None:
 421             initial_sprs = {}
 422         if initial_mem is None:
 423             initial_mem = {}
 424         if initial_insns is None:
 425             initial_insns = {}
 426             assert self.respect_pc == False, "instructions required to honor pc"
 427
 428         print("ISACaller insns", respect_pc, initial_insns, disassembly)
 429         print("ISACaller initial_msr", initial_msr)
 430
 431         # "fake program counter" mode (for unit testing)
 432         self.fake_pc = 0
 433         disasm_start = 0
 434         if not respect_pc:
 435             if isinstance(initial_mem, tuple):
 436                 self.fake_pc = initial_mem[0]
 437                 disasm_start = self.fake_pc
 438         else:
 439             disasm_start = initial_pc
 440
 441         # disassembly: we need this for now (not given from the decoder)
 442         self.disassembly = {}
 443         if disassembly:
 444             for i, code in enumerate(disassembly):
 445                 self.disassembly[i*4 + disasm_start] = code
 446
 447         # set up registers, instruction memory, data memory, PC, SPRs, MSR
 448         self.svp64rm = SVP64RM()
 449         if isinstance(initial_svstate, int):
 450             initial_svstate = SVP64State(initial_svstate)
 451         self.svstate = initial_svstate
 452         self.gpr = GPR(decoder2, self, self.svstate, regfile)
 453         self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
 454         self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
 455         self.pc = PC()
 456         self.spr = SPR(decoder2, initial_sprs)
 457         self.msr = SelectableInt(initial_msr, 64)  # underlying reg
 458
 459         # TODO, needed here:
 460         # FPR (same as GPR except for FP nums)
 461         # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
 462         #            note that mffs, mcrfs, mtfsf "manage" this FPSCR
 463         # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
 464         #         note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
 465         #         -- Done
 466         # 2.3.2 LR   (actually SPR #8) -- Done
 467         # 2.3.3 CTR  (actually SPR #9) -- Done
 468         # 2.3.4 TAR  (actually SPR #815)
 469         # 3.2.2 p45 XER  (actually SPR #1) -- Done
 470         # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
 471
 472         # create CR then allow portions of it to be "selectable" (below)
 473         #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
 474         self.cr = SelectableInt(initial_cr, 64)  # underlying reg
 475         #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
 476
 477         # "undefined", just set to variable-bit-width int (use exts "max")
 478         #self.undefined = SelectableInt(0, 256)  # TODO, not hard-code 256!
 479
 480         self.namespace = {}
 481         self.namespace.update(self.spr)
 482         self.namespace.update({'GPR': self.gpr,
 483                                'MEM': self.mem,
 484                                'SPR': self.spr,
 485                                'memassign': self.memassign,
 486                                'NIA': self.pc.NIA,
 487                                'CIA': self.pc.CIA,
 488                                'CR': self.cr,
 489                                'MSR': self.msr,
 490                                'undefined': undefined,
 491                                'mode_is_64bit': True,
 492                                'SO': XER_bits['SO']
 493                                })
 494
 495         # update pc to requested start point
 496         self.set_pc(initial_pc)
 497
 498         # field-selectable versions of Condition Register TODO check bitranges?
 499         self.crl = []
 500         for i in range(8):
 501             bits = tuple(range(i*4+32, (i+1)*4+32))  # errr... maybe?
 502             _cr = FieldSelectableInt(self.cr, bits)
 503             self.crl.append(_cr)
 504             self.namespace["CR%d" % i] = _cr
 505
 506         self.decoder = decoder2.dec
 507         self.dec2 = decoder2
 508
 509     def TRAP(self, trap_addr=0x700, trap_bit=PIb.TRAP):
 510         print("TRAP:", hex(trap_addr), hex(self.namespace['MSR'].value))
 511         # store CIA(+4?) in SRR0, set NIA to 0x700
 512         # store MSR in SRR1, set MSR to um errr something, have to check spec
 513         self.spr['SRR0'].value = self.pc.CIA.value
 514         self.spr['SRR1'].value = self.namespace['MSR'].value
 515         self.trap_nia = SelectableInt(trap_addr, 64)
 516         self.spr['SRR1'][trap_bit] = 1  # change *copy* of MSR in SRR1
 517
 518         # set exception bits.  TODO: this should, based on the address
 519         # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
 520         # bits appropriately.  however it turns out that *for now* in all
 521         # cases (all trap_addrs) the exact same thing is needed.
 522         self.msr[MSRb.IR] = 0
 523         self.msr[MSRb.DR] = 0
 524         self.msr[MSRb.FE0] = 0
 525         self.msr[MSRb.FE1] = 0
 526         self.msr[MSRb.EE] = 0
 527         self.msr[MSRb.RI] = 0
 528         self.msr[MSRb.SF] = 1
 529         self.msr[MSRb.TM] = 0
 530         self.msr[MSRb.VEC] = 0
 531         self.msr[MSRb.VSX] = 0
 532         self.msr[MSRb.PR] = 0
 533         self.msr[MSRb.FP] = 0
 534         self.msr[MSRb.PMM] = 0
 535         self.msr[MSRb.TEs] = 0
 536         self.msr[MSRb.TEe] = 0
 537         self.msr[MSRb.UND] = 0
 538         self.msr[MSRb.LE] = 1
 539
 540     def memassign(self, ea, sz, val):
 541         self.mem.memassign(ea, sz, val)
 542
 543     def prep_namespace(self, formname, op_fields):
 544         # TODO: get field names from form in decoder*1* (not decoder2)
 545         # decoder2 is hand-created, and decoder1.sigform is auto-generated
 546         # from spec
 547         # then "yield" fields only from op_fields rather than hard-coded
 548         # list, here.
 549         fields = self.decoder.sigforms[formname]
 550         for name in op_fields:
 551             if name == 'spr':
 552                 sig = getattr(fields, name.upper())
 553             else:
 554                 sig = getattr(fields, name)
 555             val = yield sig
 556             # these are all opcode fields involved in index-selection of CR,
 557             # and need to do "standard" arithmetic.  CR[BA+32] for example
 558             # would, if using SelectableInt, only be 5-bit.
 559             if name in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
 560                 self.namespace[name] = val
 561             else:
 562                 self.namespace[name] = SelectableInt(val, sig.width)
 563
 564         self.namespace['XER'] = self.spr['XER']
 565         self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
 566         self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
 567
 568     def handle_carry_(self, inputs, outputs, already_done):
 569         inv_a = yield self.dec2.e.do.invert_in
 570         if inv_a:
 571             inputs[0] = ~inputs[0]
 572
 573         imm_ok = yield self.dec2.e.do.imm_data.ok
 574         if imm_ok:
 575             imm = yield self.dec2.e.do.imm_data.data
 576             inputs.append(SelectableInt(imm, 64))
 577         assert len(outputs) >= 1
 578         print("outputs", repr(outputs))
 579         if isinstance(outputs, list) or isinstance(outputs, tuple):
 580             output = outputs[0]
 581         else:
 582             output = outputs
 583         gts = []
 584         for x in inputs:
 585             print("gt input", x, output)
 586             gt = (gtu(x, output))
 587             gts.append(gt)
 588         print(gts)
 589         cy = 1 if any(gts) else 0
 590         print("CA", cy, gts)
 591         if not (1 & already_done):
 592             self.spr['XER'][XER_bits['CA']] = cy
 593
 594         print("inputs", already_done, inputs)
 595         # 32 bit carry
 596         # ARGH... different for OP_ADD... *sigh*...
 597         op = yield self.dec2.e.do.insn_type
 598         if op == MicrOp.OP_ADD.value:
 599             res32 = (output.value & (1 << 32)) != 0
 600             a32 = (inputs[0].value & (1 << 32)) != 0
 601             if len(inputs) >= 2:
 602                 b32 = (inputs[1].value & (1 << 32)) != 0
 603             else:
 604                 b32 = False
 605             cy32 = res32 ^ a32 ^ b32
 606             print("CA32 ADD", cy32)
 607         else:
 608             gts = []
 609             for x in inputs:
 610                 print("input", x, output)
 611                 print("     x[32:64]", x, x[32:64])
 612                 print("     o[32:64]", output, output[32:64])
 613                 gt = (gtu(x[32:64], output[32:64])) == SelectableInt(1, 1)
 614                 gts.append(gt)
 615             cy32 = 1 if any(gts) else 0
 616             print("CA32", cy32, gts)
 617         if not (2 & already_done):
 618             self.spr['XER'][XER_bits['CA32']] = cy32
 619
 620     def handle_overflow(self, inputs, outputs, div_overflow):
 621         if hasattr(self.dec2.e.do, "invert_in"):
 622             inv_a = yield self.dec2.e.do.invert_in
 623             if inv_a:
 624                 inputs[0] = ~inputs[0]
 625
 626         imm_ok = yield self.dec2.e.do.imm_data.ok
 627         if imm_ok:
 628             imm = yield self.dec2.e.do.imm_data.data
 629             inputs.append(SelectableInt(imm, 64))
 630         assert len(outputs) >= 1
 631         print("handle_overflow", inputs, outputs, div_overflow)
 632         if len(inputs) < 2 and div_overflow is None:
 633             return
 634
 635         # div overflow is different: it's returned by the pseudo-code
 636         # because it's more complex than can be done by analysing the output
 637         if div_overflow is not None:
 638             ov, ov32 = div_overflow, div_overflow
 639         # arithmetic overflow can be done by analysing the input and output
 640         elif len(inputs) >= 2:
 641             output = outputs[0]
 642
 643             # OV (64-bit)
 644             input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
 645             output_sgn = exts(output.value, output.bits) < 0
 646             ov = 1 if input_sgn[0] == input_sgn[1] and \
 647                 output_sgn != input_sgn[0] else 0
 648
 649             # OV (32-bit)
 650             input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
 651             output32_sgn = exts(output.value, 32) < 0
 652             ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
 653                 output32_sgn != input32_sgn[0] else 0
 654
 655         self.spr['XER'][XER_bits['OV']] = ov
 656         self.spr['XER'][XER_bits['OV32']] = ov32
 657         so = self.spr['XER'][XER_bits['SO']]
 658         so = so | ov
 659         self.spr['XER'][XER_bits['SO']] = so
 660
 661     def handle_comparison(self, outputs, cr_idx=0):
 662         out = outputs[0]
 663         assert isinstance(out, SelectableInt), \
 664             "out zero not a SelectableInt %s" % repr(outputs)
 665         print("handle_comparison", out.bits, hex(out.value))
 666         # TODO - XXX *processor* in 32-bit mode
 667         # https://bugs.libre-soc.org/show_bug.cgi?id=424
 668         # if is_32bit:
 669         #    o32 = exts(out.value, 32)
 670         #    print ("handle_comparison exts 32 bit", hex(o32))
 671         out = exts(out.value, out.bits)
 672         print("handle_comparison exts", hex(out))
 673         zero = SelectableInt(out == 0, 1)
 674         positive = SelectableInt(out > 0, 1)
 675         negative = SelectableInt(out < 0, 1)
 676         SO = self.spr['XER'][XER_bits['SO']]
 677         print("handle_comparison SO", SO)
 678         cr_field = selectconcat(negative, positive, zero, SO)
 679         self.crl[cr_idx].eq(cr_field)
 680
 681     def set_pc(self, pc_val):
 682         self.namespace['NIA'] = SelectableInt(pc_val, 64)
 683         self.pc.update(self.namespace, self.is_svp64_mode)
 684
 685     def setup_one(self):
 686         """set up one instruction
 687         """
 688         if self.respect_pc:
 689             pc = self.pc.CIA.value
 690         else:
 691             pc = self.fake_pc
 692         self._pc = pc
 693         ins = self.imem.ld(pc, 4, False, True)
 694         if ins is None:
 695             raise KeyError("no instruction at 0x%x" % pc)
 696         print("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
 697         print("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
 698
 699         yield self.dec2.sv_rm.eq(0)
 700         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
 701         yield self.dec2.dec.bigendian.eq(self.bigendian)
 702         yield self.dec2.state.msr.eq(self.msr.value)
 703         yield self.dec2.state.pc.eq(pc)
 704         yield self.dec2.state.svstate.eq(self.svstate.spr.value)
 705
 706         # SVP64.  first, check if the opcode is EXT001, and SVP64 id bits set
 707         yield Settle()
 708         opcode = yield self.dec2.dec.opcode_in
 709         pfx = SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
 710         pfx.insn.value = opcode
 711         major = pfx.major.asint(msb0=True) # MSB0 inversion
 712         print ("prefix test: opcode:", major, bin(major),
 713                 pfx.insn[7] == 0b1, pfx.insn[9] == 0b1)
 714         self.is_svp64_mode = ((major == 0b000001) and
 715                               pfx.insn[7].value == 0b1 and
 716                               pfx.insn[9].value == 0b1)
 717         self.pc.update_nia(self.is_svp64_mode)
 718         if not self.is_svp64_mode:
 719             return
 720
 721         # in SVP64 mode.  decode/print out svp64 prefix, get v3.0B instruction
 722         print ("svp64.rm", bin(pfx.rm.asint(msb0=True)))
 723         print ("    svstate.vl", self.svstate.vl.asint(msb0=True))
 724         print ("    svstate.mvl", self.svstate.maxvl.asint(msb0=True))
 725         sv_rm = pfx.rm.asint(msb0=True)
 726         ins = self.imem.ld(pc+4, 4, False, True)
 727         print("     svsetup: 0x%x 0x%x %s" % (pc+4, ins & 0xffffffff, bin(ins)))
 728         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff) # v3.0B suffix
 729         yield self.dec2.sv_rm.eq(sv_rm)                        # svp64 prefix
 730         yield Settle()
 731
 732     def execute_one(self):
 733         """execute one instruction
 734         """
 735         # get the disassembly code for this instruction
 736         if self.is_svp64_mode:
 737             code = self.disassembly[self._pc+4]
 738             print("    svp64 sim-execute", hex(self._pc), code)
 739         else:
 740             code = self.disassembly[self._pc]
 741             print("sim-execute", hex(self._pc), code)
 742         opname = code.split(' ')[0]
 743         yield from self.call(opname)
 744
 745         # don't use this except in special circumstances
 746         if not self.respect_pc:
 747             self.fake_pc += 4
 748
 749         print("execute one, CIA NIA", self.pc.CIA.value, self.pc.NIA.value)
 750
 751     def get_assembly_name(self):
 752         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 753         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 754         dec_insn = yield self.dec2.e.do.insn
 755         asmcode = yield self.dec2.dec.op.asmcode
 756         print("get assembly name asmcode", asmcode, hex(dec_insn))
 757         asmop = insns.get(asmcode, None)
 758         int_op = yield self.dec2.dec.op.internal_op
 759
 760         # sigh reconstruct the assembly instruction name
 761         if hasattr(self.dec2.e.do, "oe"):
 762             ov_en = yield self.dec2.e.do.oe.oe
 763             ov_ok = yield self.dec2.e.do.oe.ok
 764         else:
 765             ov_en = False
 766             ov_ok = False
 767         if hasattr(self.dec2.e.do, "rc"):
 768             rc_en = yield self.dec2.e.do.rc.rc
 769             rc_ok = yield self.dec2.e.do.rc.ok
 770         else:
 771             rc_en = False
 772             rc_ok = False
 773         # grrrr have to special-case MUL op (see DecodeOE)
 774         print("ov %d en %d rc %d en %d op %d" %
 775               (ov_ok, ov_en, rc_ok, rc_en, int_op))
 776         if int_op in [MicrOp.OP_MUL_H64.value, MicrOp.OP_MUL_H32.value]:
 777             print("mul op")
 778             if rc_en & rc_ok:
 779                 asmop += "."
 780         else:
 781             if not asmop.endswith("."):  # don't add "." to "andis."
 782                 if rc_en & rc_ok:
 783                     asmop += "."
 784         if hasattr(self.dec2.e.do, "lk"):
 785             lk = yield self.dec2.e.do.lk
 786             if lk:
 787                 asmop += "l"
 788         print("int_op", int_op)
 789         if int_op in [MicrOp.OP_B.value, MicrOp.OP_BC.value]:
 790             AA = yield self.dec2.dec.fields.FormI.AA[0:-1]
 791             print("AA", AA)
 792             if AA:
 793                 asmop += "a"
 794         spr_msb = yield from self.get_spr_msb()
 795         if int_op == MicrOp.OP_MFCR.value:
 796             if spr_msb:
 797                 asmop = 'mfocrf'
 798             else:
 799                 asmop = 'mfcr'
 800         # XXX TODO: for whatever weird reason this doesn't work
 801         # https://bugs.libre-soc.org/show_bug.cgi?id=390
 802         if int_op == MicrOp.OP_MTCRF.value:
 803             if spr_msb:
 804                 asmop = 'mtocrf'
 805             else:
 806                 asmop = 'mtcrf'
 807         return asmop
 808
 809     def get_spr_msb(self):
 810         dec_insn = yield self.dec2.e.do.insn
 811         return dec_insn & (1 << 20) != 0  # sigh - XFF.spr[-1]?
 812
 813     def call(self, name):
 814         """call(opcode) - the primary execution point for instructions
 815         """
 816         name = name.strip()  # remove spaces if not already done so
 817         if self.halted:
 818             print("halted - not executing", name)
 819             return
 820
 821         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 822         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 823         asmop = yield from self.get_assembly_name()
 824         print("call", name, asmop)
 825
 826         # check privileged
 827         int_op = yield self.dec2.dec.op.internal_op
 828         spr_msb = yield from self.get_spr_msb()
 829
 830         instr_is_privileged = False
 831         if int_op in [MicrOp.OP_ATTN.value,
 832                       MicrOp.OP_MFMSR.value,
 833                       MicrOp.OP_MTMSR.value,
 834                       MicrOp.OP_MTMSRD.value,
 835                       # TODO: OP_TLBIE
 836                       MicrOp.OP_RFID.value]:
 837             instr_is_privileged = True
 838         if int_op in [MicrOp.OP_MFSPR.value,
 839                       MicrOp.OP_MTSPR.value] and spr_msb:
 840             instr_is_privileged = True
 841
 842         print("is priv", instr_is_privileged, hex(self.msr.value),
 843               self.msr[MSRb.PR])
 844         # check MSR priv bit and whether op is privileged: if so, throw trap
 845         if instr_is_privileged and self.msr[MSRb.PR] == 1:
 846             self.TRAP(0x700, PIb.PRIV)
 847             self.namespace['NIA'] = self.trap_nia
 848             self.pc.update(self.namespace, self.is_svp64_mode)
 849             return
 850
 851         # check halted condition
 852         if name == 'attn':
 853             self.halted = True
 854             return
 855
 856         # check illegal instruction
 857         illegal = False
 858         if name not in ['mtcrf', 'mtocrf']:
 859             illegal = name != asmop
 860
 861         if illegal:
 862             print("illegal", name, asmop)
 863             self.TRAP(0x700, PIb.ILLEG)
 864             self.namespace['NIA'] = self.trap_nia
 865             self.pc.update(self.namespace, self.is_svp64_mode)
 866             print("name %s != %s - calling ILLEGAL trap, PC: %x" %
 867                   (name, asmop, self.pc.CIA.value))
 868             return
 869
 870         info = self.instrs[name]
 871         yield from self.prep_namespace(info.form, info.op_fields)
 872
 873         # preserve order of register names
 874         input_names = create_args(list(info.read_regs) +
 875                                   list(info.uninit_regs))
 876         print(input_names)
 877
 878         # get SVP64 entry for the current instruction
 879         sv_rm = self.svp64rm.instrs.get(name)
 880         if sv_rm is not None:
 881             dest_cr, src_cr, src_byname, dest_byname = decode_extra(sv_rm)
 882         else:
 883             dest_cr, src_cr, src_byname, dest_byname = False, False, {}, {}
 884         print ("sv rm", sv_rm, dest_cr, src_cr, src_byname, dest_byname)
 885
 886         # get SVSTATE srcstep.  TODO: dststep (twin predication)
 887         srcstep = self.svstate.srcstep.asint(msb0=True)
 888         vl = self.svstate.vl.asint(msb0=True)
 889         mvl = self.svstate.maxvl.asint(msb0=True)
 890
 891         # VL=0 in SVP64 mode means "do nothing: skip instruction"
 892         if self.is_svp64_mode and vl == 0:
 893             self.pc.update(self.namespace, self.is_svp64_mode)
 894             print("end of call", self.namespace['CIA'], self.namespace['NIA'])
 895             return
 896
 897         # main input registers (RT, RA ...)
 898         inputs = []
 899         for name in input_names:
 900             # using PowerDecoder2, first, find the decoder index.
 901             # (mapping name RA RB RC RS to in1, in2, in3)
 902             regnum, is_vec = yield from get_pdecode_idx_in(self.dec2, name)
 903             if regnum is None:
 904                 # doing this is not part of svp64, it's because output
 905                 # registers, to be modified, need to be in the namespace.
 906                 regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, name)
 907             # here's where we go "vector".  TODO: zero-testing (RA_IS_ZERO)
 908             # XXX already done by PowerDecoder2, now
 909             #if is_vec:
 910             #   regnum += srcstep # TODO, elwidth overrides
 911
 912             # in case getting the register number is needed, _RA, _RB
 913             regname = "_" + name
 914             self.namespace[regname] = regnum
 915             print('reading reg %s %d' % (name, regnum), is_vec)
 916             reg_val = self.gpr(regnum)
 917             inputs.append(reg_val)
 918
 919         # "special" registers
 920         for special in info.special_regs:
 921             if special in special_sprs:
 922                 inputs.append(self.spr[special])
 923             else:
 924                 inputs.append(self.namespace[special])
 925
 926         # clear trap (trap) NIA
 927         self.trap_nia = None
 928
 929         print(inputs)
 930         results = info.func(self, *inputs)
 931         print(results)
 932
 933         # "inject" decorator takes namespace from function locals: we need to
 934         # overwrite NIA being overwritten (sigh)
 935         if self.trap_nia is not None:
 936             self.namespace['NIA'] = self.trap_nia
 937
 938         print("after func", self.namespace['CIA'], self.namespace['NIA'])
 939
 940         # detect if CA/CA32 already in outputs (sra*, basically)
 941         already_done = 0
 942         if info.write_regs:
 943             output_names = create_args(info.write_regs)
 944             for name in output_names:
 945                 if name == 'CA':
 946                     already_done |= 1
 947                 if name == 'CA32':
 948                     already_done |= 2
 949
 950         print("carry already done?", bin(already_done))
 951         if hasattr(self.dec2.e.do, "output_carry"):
 952             carry_en = yield self.dec2.e.do.output_carry
 953         else:
 954             carry_en = False
 955         if carry_en:
 956             yield from self.handle_carry_(inputs, results, already_done)
 957
 958         # detect if overflow was in return result
 959         overflow = None
 960         if info.write_regs:
 961             for name, output in zip(output_names, results):
 962                 if name == 'overflow':
 963                     overflow = output
 964
 965         if hasattr(self.dec2.e.do, "oe"):
 966             ov_en = yield self.dec2.e.do.oe.oe
 967             ov_ok = yield self.dec2.e.do.oe.ok
 968         else:
 969             ov_en = False
 970             ov_ok = False
 971         print("internal overflow", overflow, ov_en, ov_ok)
 972         if ov_en & ov_ok:
 973             yield from self.handle_overflow(inputs, results, overflow)
 974
 975         if hasattr(self.dec2.e.do, "rc"):
 976             rc_en = yield self.dec2.e.do.rc.rc
 977         else:
 978             rc_en = False
 979         if rc_en:
 980             regnum, is_vec = yield from get_pdecode_cr_out(self.dec2, "CR0")
 981             regnum = 0 # TODO fix
 982             self.handle_comparison(results, regnum)
 983
 984         # svp64 loop can end early if the dest is scalar
 985         svp64_dest_vector = False
 986
 987         # any modified return results?
 988         if info.write_regs:
 989             for name, output in zip(output_names, results):
 990                 if name == 'overflow':  # ignore, done already (above)
 991                     continue
 992                 if isinstance(output, int):
 993                     output = SelectableInt(output, 256)
 994                 if name in ['CA', 'CA32']:
 995                     if carry_en:
 996                         print("writing %s to XER" % name, output)
 997                         self.spr['XER'][XER_bits[name]] = output.value
 998                     else:
 999                         print("NOT writing %s to XER" % name, output)
1000                 elif name in info.special_regs:
1001                     print('writing special %s' % name, output, special_sprs)
1002                     if name in special_sprs:
1003                         self.spr[name] = output
1004                     else:
1005                         self.namespace[name].eq(output)
1006                     if name == 'MSR':
1007                         print('msr written', hex(self.msr.value))
1008                 else:
1009                     regnum, is_vec = yield from get_pdecode_idx_out(self.dec2,
1010                                                 name)
1011                     if regnum is None:
1012                         # temporary hack for not having 2nd output
1013                         regnum = yield getattr(self.decoder, name)
1014                         is_vec = False
1015                     # here's where we go "vector".
1016                     if is_vec:
1017                         # XXX already done by PowerDecoder2
1018                         # regnum += srcstep # TODO, elwidth overrides
1019                         svp64_dest_vector = True
1020                     print('writing reg %d %s' % (regnum, str(output)), is_vec)
1021                     if output.bits > 64:
1022                         output = SelectableInt(output.value, 64)
1023                     self.gpr[regnum] = output
1024
1025         # check if it is the SVSTATE.src/dest step that needs incrementing
1026         # this is our Sub-Program-Counter loop from 0 to VL-1
1027         if self.is_svp64_mode:
1028             # XXX twin predication TODO
1029             vl = self.svstate.vl.asint(msb0=True)
1030             mvl = self.svstate.maxvl.asint(msb0=True)
1031             srcstep = self.svstate.srcstep.asint(msb0=True)
1032             print ("    svstate.vl", vl)
1033             print ("    svstate.mvl", mvl)
1034             print ("    svstate.srcstep", srcstep)
1035             # check if srcstep needs incrementing by one, stop PC advancing
1036             if svp64_dest_vector and srcstep != vl-1:
1037                 self.svstate.srcstep += SelectableInt(1, 7)
1038                 self.pc.NIA.value = self.pc.CIA.value
1039                 self.namespace['NIA'] = self.pc.NIA
1040                 print("end of sub-pc call", self.namespace['CIA'],
1041                                      self.namespace['NIA'])
1042                 return # DO NOT allow PC to update whilst Sub-PC loop running
1043             # reset to zero
1044             self.svstate.srcstep[0:7] = 0
1045             print ("    svstate.srcstep loop end (PC to update)")
1046             self.pc.update_nia(self.is_svp64_mode)
1047             self.namespace['NIA'] = self.pc.NIA
1048
1049         # UPDATE program counter
1050         self.pc.update(self.namespace, self.is_svp64_mode)
1051         print("end of call", self.namespace['CIA'], self.namespace['NIA'])
1052
1053
1054 def inject():
1055     """Decorator factory.
1056
1057     this decorator will "inject" variables into the function's namespace,
1058     from the *dictionary* in self.namespace.  it therefore becomes possible
1059     to make it look like a whole stack of variables which would otherwise
1060     need "self." inserted in front of them (*and* for those variables to be
1061     added to the instance) "appear" in the function.
1062
1063     "self.namespace['SI']" for example becomes accessible as just "SI" but
1064     *only* inside the function, when decorated.
1065     """
1066     def variable_injector(func):
1067         @wraps(func)
1068         def decorator(*args, **kwargs):
1069             try:
1070                 func_globals = func.__globals__  # Python 2.6+
1071             except AttributeError:
1072                 func_globals = func.func_globals  # Earlier versions.
1073
1074             context = args[0].namespace  # variables to be injected
1075             saved_values = func_globals.copy()  # Shallow copy of dict.
1076             func_globals.update(context)
1077             result = func(*args, **kwargs)
1078             print("globals after", func_globals['CIA'], func_globals['NIA'])
1079             print("args[0]", args[0].namespace['CIA'],
1080                   args[0].namespace['NIA'])
1081             args[0].namespace = func_globals
1082             #exec (func.__code__, func_globals)
1083
1084             # finally:
1085             #    func_globals = saved_values  # Undo changes.
1086
1087             return result
1088
1089         return decorator
1090
1091     return variable_injector