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