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 soc
.decoder
.orderedset
import OrderedSet
20 from soc
.decoder
.selectable_int
import (FieldSelectableInt
, SelectableInt
,
22 from soc
.decoder
.power_enums
import (spr_dict
, spr_byname
, XER_bits
,
23 insns
, MicrOp
, In1Sel
, In2Sel
, In3Sel
,
25 SVP64RMMode
, SVP64PredMode
,
26 SVP64PredInt
, SVP64PredCR
)
28 from soc
.decoder
.power_enums
import SVPtype
30 from soc
.decoder
.helpers
import exts
, gtu
, ltu
, undefined
31 from soc
.consts
import PIb
, MSRb
# big-endian (PowerISA versions)
32 from soc
.consts
import SVP64CROffs
33 from soc
.decoder
.power_svp64
import SVP64RM
, decode_extra
35 from soc
.decoder
.isa
.radixmmu
import RADIX
36 from soc
.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
, pc_init
=0):
129 self
.CIA
= SelectableInt(pc_init
, 64)
130 self
.NIA
= self
.CIA
+ SelectableInt(4, 64) # only true for v3.0B!
132 def update_nia(self
, is_svp64
):
133 increment
= 8 if is_svp64
else 4
134 self
.NIA
= self
.CIA
+ SelectableInt(increment
, 64)
136 def update(self
, namespace
, is_svp64
):
137 """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
139 self
.CIA
= namespace
['NIA'].narrow(64)
140 self
.update_nia(is_svp64
)
141 namespace
['CIA'] = self
.CIA
142 namespace
['NIA'] = self
.NIA
145 # Simple-V: see https://libre-soc.org/openpower/sv
147 def __init__(self
, init
=0):
148 self
.spr
= SelectableInt(init
, 32)
149 # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
150 self
.maxvl
= FieldSelectableInt(self
.spr
, tuple(range(0,7)))
151 self
.vl
= FieldSelectableInt(self
.spr
, tuple(range(7,14)))
152 self
.srcstep
= FieldSelectableInt(self
.spr
, tuple(range(14,21)))
153 self
.dststep
= FieldSelectableInt(self
.spr
, tuple(range(21,28)))
154 self
.subvl
= FieldSelectableInt(self
.spr
, tuple(range(28,30)))
155 self
.svstep
= FieldSelectableInt(self
.spr
, tuple(range(30,32)))
160 def __init__(self
, init
=0):
161 self
.spr
= SelectableInt(init
, 24)
162 # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
163 self
.mmode
= FieldSelectableInt(self
.spr
, [0])
164 self
.mask
= FieldSelectableInt(self
.spr
, tuple(range(1,4)))
165 self
.elwidth
= FieldSelectableInt(self
.spr
, tuple(range(4,6)))
166 self
.ewsrc
= FieldSelectableInt(self
.spr
, tuple(range(6,8)))
167 self
.subvl
= FieldSelectableInt(self
.spr
, tuple(range(8,10)))
168 self
.extra
= FieldSelectableInt(self
.spr
, tuple(range(10,19)))
169 self
.mode
= FieldSelectableInt(self
.spr
, tuple(range(19,24)))
170 # these cover the same extra field, split into parts as EXTRA2
171 self
.extra2
= list(range(4))
172 self
.extra2
[0] = FieldSelectableInt(self
.spr
, tuple(range(10,12)))
173 self
.extra2
[1] = FieldSelectableInt(self
.spr
, tuple(range(12,14)))
174 self
.extra2
[2] = FieldSelectableInt(self
.spr
, tuple(range(14,16)))
175 self
.extra2
[3] = FieldSelectableInt(self
.spr
, tuple(range(16,18)))
176 self
.smask
= FieldSelectableInt(self
.spr
, tuple(range(16,19)))
177 # and here as well, but EXTRA3
178 self
.extra3
= list(range(3))
179 self
.extra3
[0] = FieldSelectableInt(self
.spr
, tuple(range(10,13)))
180 self
.extra3
[1] = FieldSelectableInt(self
.spr
, tuple(range(13,16)))
181 self
.extra3
[2] = FieldSelectableInt(self
.spr
, tuple(range(16,19)))
184 SVP64RM_MMODE_SIZE
= len(SVP64RMFields().mmode
.br
)
185 SVP64RM_MASK_SIZE
= len(SVP64RMFields().mask
.br
)
186 SVP64RM_ELWIDTH_SIZE
= len(SVP64RMFields().elwidth
.br
)
187 SVP64RM_EWSRC_SIZE
= len(SVP64RMFields().ewsrc
.br
)
188 SVP64RM_SUBVL_SIZE
= len(SVP64RMFields().subvl
.br
)
189 SVP64RM_EXTRA2_SPEC_SIZE
= len(SVP64RMFields().extra2
[0].br
)
190 SVP64RM_EXTRA3_SPEC_SIZE
= len(SVP64RMFields().extra3
[0].br
)
191 SVP64RM_SMASK_SIZE
= len(SVP64RMFields().smask
.br
)
192 SVP64RM_MODE_SIZE
= len(SVP64RMFields().mode
.br
)
195 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
196 class SVP64PrefixFields
:
198 self
.insn
= SelectableInt(0, 32)
199 # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
200 self
.major
= FieldSelectableInt(self
.insn
, tuple(range(0,6)))
201 self
.pid
= FieldSelectableInt(self
.insn
, (7, 9)) # must be 0b11
202 rmfields
= [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
203 self
.rm
= FieldSelectableInt(self
.insn
, rmfields
)
206 SV64P_MAJOR_SIZE
= len(SVP64PrefixFields().major
.br
)
207 SV64P_PID_SIZE
= len(SVP64PrefixFields().pid
.br
)
208 SV64P_RM_SIZE
= len(SVP64PrefixFields().rm
.br
)
210 # decode SVP64 predicate integer to reg number and invert
211 def get_predint(gpr
, mask
):
214 if mask
== SVP64PredInt
.ALWAYS
.value
:
215 return 0xffff_ffff_ffff_ffff
216 if mask
== SVP64PredInt
.R3_UNARY
.value
:
217 return 1 << (gpr(3).value
& 0b111111)
218 if mask
== SVP64PredInt
.R3
.value
:
220 if mask
== SVP64PredInt
.R3_N
.value
:
222 if mask
== SVP64PredInt
.R10
.value
:
224 if mask
== SVP64PredInt
.R10_N
.value
:
225 return ~
gpr(10).value
226 if mask
== SVP64PredInt
.R30
.value
:
228 if mask
== SVP64PredInt
.R30_N
.value
:
229 return ~
gpr(30).value
231 # decode SVP64 predicate CR to reg number and invert
232 def _get_predcr(mask
):
233 if mask
== SVP64PredCR
.LT
.value
:
235 if mask
== SVP64PredCR
.GE
.value
:
237 if mask
== SVP64PredCR
.GT
.value
:
239 if mask
== SVP64PredCR
.LE
.value
:
241 if mask
== SVP64PredCR
.EQ
.value
:
243 if mask
== SVP64PredCR
.NE
.value
:
245 if mask
== SVP64PredCR
.SO
.value
:
247 if mask
== SVP64PredCR
.NS
.value
:
250 def get_predcr(crl
, mask
, vl
):
251 idx
, noninv
= _get_predcr(mask
)
254 cr
= crl
[i
+SVP64CROffs
.CRPred
]
255 if cr
[idx
].value
== noninv
:
261 def __init__(self
, dec2
, initial_sprs
={}):
264 for key
, v
in initial_sprs
.items():
265 if isinstance(key
, SelectableInt
):
267 key
= special_sprs
.get(key
, key
)
268 if isinstance(key
, int):
271 info
= spr_byname
[key
]
272 if not isinstance(v
, SelectableInt
):
273 v
= SelectableInt(v
, info
.length
)
276 def __getitem__(self
, key
):
277 print("get spr", key
)
278 print("dict", self
.items())
279 # if key in special_sprs get the special spr, otherwise return key
280 if isinstance(key
, SelectableInt
):
282 if isinstance(key
, int):
283 key
= spr_dict
[key
].SPR
284 key
= special_sprs
.get(key
, key
)
285 if key
== 'HSRR0': # HACK!
287 if key
== 'HSRR1': # HACK!
290 res
= dict.__getitem
__(self
, key
)
292 if isinstance(key
, int):
295 info
= spr_byname
[key
]
296 dict.__setitem
__(self
, key
, SelectableInt(0, info
.length
))
297 res
= dict.__getitem
__(self
, key
)
298 print("spr returning", key
, res
)
301 def __setitem__(self
, key
, value
):
302 if isinstance(key
, SelectableInt
):
304 if isinstance(key
, int):
305 key
= spr_dict
[key
].SPR
306 print("spr key", key
)
307 key
= special_sprs
.get(key
, key
)
308 if key
== 'HSRR0': # HACK!
309 self
.__setitem
__('SRR0', value
)
310 if key
== 'HSRR1': # HACK!
311 self
.__setitem
__('SRR1', value
)
312 print("setting spr", key
, value
)
313 dict.__setitem
__(self
, key
, value
)
315 def __call__(self
, ridx
):
318 def get_pdecode_idx_in(dec2
, name
):
320 in1_sel
= yield op
.in1_sel
321 in2_sel
= yield op
.in2_sel
322 in3_sel
= yield op
.in3_sel
323 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
324 in1
= yield dec2
.e
.read_reg1
.data
325 in2
= yield dec2
.e
.read_reg2
.data
326 in3
= yield dec2
.e
.read_reg3
.data
327 in1_isvec
= yield dec2
.in1_isvec
328 in2_isvec
= yield dec2
.in2_isvec
329 in3_isvec
= yield dec2
.in3_isvec
330 print ("get_pdecode_idx_in in1", name
, in1_sel
, In1Sel
.RA
.value
,
332 print ("get_pdecode_idx_in in2", name
, in2_sel
, In2Sel
.RB
.value
,
334 print ("get_pdecode_idx_in in3", name
, in3_sel
, In3Sel
.RS
.value
,
336 # identify which regnames map to in1/2/3
338 if (in1_sel
== In1Sel
.RA
.value
or
339 (in1_sel
== In1Sel
.RA_OR_ZERO
.value
and in1
!= 0)):
340 return in1
, in1_isvec
341 if in1_sel
== In1Sel
.RA_OR_ZERO
.value
:
342 return in1
, in1_isvec
344 if in2_sel
== In2Sel
.RB
.value
:
345 return in2
, in2_isvec
346 if in3_sel
== In3Sel
.RB
.value
:
347 return in3
, in3_isvec
348 # XXX TODO, RC doesn't exist yet!
350 assert False, "RC does not exist yet"
352 if in1_sel
== In1Sel
.RS
.value
:
353 return in1
, in1_isvec
354 if in2_sel
== In2Sel
.RS
.value
:
355 return in2
, in2_isvec
356 if in3_sel
== In3Sel
.RS
.value
:
357 return in3
, in3_isvec
361 def get_pdecode_cr_out(dec2
, name
):
363 out_sel
= yield op
.cr_out
364 out_bitfield
= yield dec2
.dec_cr_out
.cr_bitfield
.data
365 sv_cr_out
= yield op
.sv_cr_out
366 spec
= yield dec2
.crout_svdec
.spec
367 sv_override
= yield dec2
.dec_cr_out
.sv_override
368 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
369 out
= yield dec2
.e
.write_cr
.data
370 o_isvec
= yield dec2
.o_isvec
371 print ("get_pdecode_cr_out", out_sel
, CROutSel
.CR0
.value
, out
, o_isvec
)
372 print (" sv_cr_out", sv_cr_out
)
373 print (" cr_bf", out_bitfield
)
374 print (" spec", spec
)
375 print (" override", sv_override
)
376 # identify which regnames map to out / o2
378 if out_sel
== CROutSel
.CR0
.value
:
380 print ("get_pdecode_idx_out not found", name
)
384 def get_pdecode_idx_out(dec2
, name
):
386 out_sel
= yield op
.out_sel
387 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
388 out
= yield dec2
.e
.write_reg
.data
389 o_isvec
= yield dec2
.o_isvec
390 # identify which regnames map to out / o2
392 print ("get_pdecode_idx_out", out_sel
, OutSel
.RA
.value
, out
, o_isvec
)
393 if out_sel
== OutSel
.RA
.value
:
396 print ("get_pdecode_idx_out", out_sel
, OutSel
.RT
.value
,
397 OutSel
.RT_OR_ZERO
.value
, out
, o_isvec
)
398 if out_sel
== OutSel
.RT
.value
:
400 print ("get_pdecode_idx_out not found", name
)
405 def get_pdecode_idx_out2(dec2
, name
):
407 print ("TODO: get_pdecode_idx_out2", name
)
412 # decoder2 - an instance of power_decoder2
413 # regfile - a list of initial values for the registers
414 # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
415 # respect_pc - tracks the program counter. requires initial_insns
416 def __init__(self
, decoder2
, regfile
, initial_sprs
=None, initial_cr
=0,
417 initial_mem
=None, initial_msr
=0,
419 initial_insns
=None, respect_pc
=False,
425 self
.bigendian
= bigendian
427 self
.is_svp64_mode
= False
428 self
.respect_pc
= respect_pc
429 if initial_sprs
is None:
431 if initial_mem
is None:
433 if initial_insns
is None:
435 assert self
.respect_pc
== False, "instructions required to honor pc"
437 print("ISACaller insns", respect_pc
, initial_insns
, disassembly
)
438 print("ISACaller initial_msr", initial_msr
)
440 # "fake program counter" mode (for unit testing)
444 if isinstance(initial_mem
, tuple):
445 self
.fake_pc
= initial_mem
[0]
446 disasm_start
= self
.fake_pc
448 disasm_start
= initial_pc
450 # disassembly: we need this for now (not given from the decoder)
451 self
.disassembly
= {}
453 for i
, code
in enumerate(disassembly
):
454 self
.disassembly
[i
*4 + disasm_start
] = code
456 # set up registers, instruction memory, data memory, PC, SPRs, MSR
457 self
.svp64rm
= SVP64RM()
458 if initial_svstate
is None:
460 if isinstance(initial_svstate
, int):
461 initial_svstate
= SVP64State(initial_svstate
)
462 self
.svstate
= initial_svstate
463 self
.gpr
= GPR(decoder2
, self
, self
.svstate
, regfile
)
464 self
.spr
= SPR(decoder2
, initial_sprs
) # initialise SPRs before MMU
465 self
.mem
= Mem(row_bytes
=8, initial_mem
=initial_mem
)
466 self
.imem
= Mem(row_bytes
=4, initial_mem
=initial_insns
)
467 # MMU mode, redirect underlying Mem through RADIX
469 self
.mem
= RADIX(self
.mem
, self
)
470 self
.imem
= RADIX(self
.imem
, self
)
472 self
.msr
= SelectableInt(initial_msr
, 64) # underlying reg
475 # FPR (same as GPR except for FP nums)
476 # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
477 # note that mffs, mcrfs, mtfsf "manage" this FPSCR
478 # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
479 # note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
481 # 2.3.2 LR (actually SPR #8) -- Done
482 # 2.3.3 CTR (actually SPR #9) -- Done
483 # 2.3.4 TAR (actually SPR #815)
484 # 3.2.2 p45 XER (actually SPR #1) -- Done
485 # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
487 # create CR then allow portions of it to be "selectable" (below)
488 #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
489 self
.cr
= SelectableInt(initial_cr
, 64) # underlying reg
490 #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
492 # "undefined", just set to variable-bit-width int (use exts "max")
493 #self.undefined = SelectableInt(0, 256) # TODO, not hard-code 256!
496 self
.namespace
.update(self
.spr
)
497 self
.namespace
.update({'GPR': self
.gpr
,
500 'memassign': self
.memassign
,
503 'SVSTATE': self
.svstate
.spr
,
506 'undefined': undefined
,
507 'mode_is_64bit': True,
511 # update pc to requested start point
512 self
.set_pc(initial_pc
)
514 # field-selectable versions of Condition Register TODO check bitranges?
517 bits
= tuple(range(i
*4+32, (i
+1)*4+32)) # errr... maybe?
518 _cr
= FieldSelectableInt(self
.cr
, bits
)
520 self
.namespace
["CR%d" % i
] = _cr
522 self
.decoder
= decoder2
.dec
525 def TRAP(self
, trap_addr
=0x700, trap_bit
=PIb
.TRAP
):
526 print("TRAP:", hex(trap_addr
), hex(self
.namespace
['MSR'].value
))
527 # store CIA(+4?) in SRR0, set NIA to 0x700
528 # store MSR in SRR1, set MSR to um errr something, have to check spec
529 self
.spr
['SRR0'].value
= self
.pc
.CIA
.value
530 self
.spr
['SRR1'].value
= self
.namespace
['MSR'].value
531 self
.trap_nia
= SelectableInt(trap_addr
, 64)
532 self
.spr
['SRR1'][trap_bit
] = 1 # change *copy* of MSR in SRR1
534 # set exception bits. TODO: this should, based on the address
535 # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
536 # bits appropriately. however it turns out that *for now* in all
537 # cases (all trap_addrs) the exact same thing is needed.
538 self
.msr
[MSRb
.IR
] = 0
539 self
.msr
[MSRb
.DR
] = 0
540 self
.msr
[MSRb
.FE0
] = 0
541 self
.msr
[MSRb
.FE1
] = 0
542 self
.msr
[MSRb
.EE
] = 0
543 self
.msr
[MSRb
.RI
] = 0
544 self
.msr
[MSRb
.SF
] = 1
545 self
.msr
[MSRb
.TM
] = 0
546 self
.msr
[MSRb
.VEC
] = 0
547 self
.msr
[MSRb
.VSX
] = 0
548 self
.msr
[MSRb
.PR
] = 0
549 self
.msr
[MSRb
.FP
] = 0
550 self
.msr
[MSRb
.PMM
] = 0
551 self
.msr
[MSRb
.TEs
] = 0
552 self
.msr
[MSRb
.TEe
] = 0
553 self
.msr
[MSRb
.UND
] = 0
554 self
.msr
[MSRb
.LE
] = 1
556 def memassign(self
, ea
, sz
, val
):
557 self
.mem
.memassign(ea
, sz
, val
)
559 def prep_namespace(self
, formname
, op_fields
):
560 # TODO: get field names from form in decoder*1* (not decoder2)
561 # decoder2 is hand-created, and decoder1.sigform is auto-generated
563 # then "yield" fields only from op_fields rather than hard-coded
565 fields
= self
.decoder
.sigforms
[formname
]
566 for name
in op_fields
:
568 sig
= getattr(fields
, name
.upper())
570 sig
= getattr(fields
, name
)
572 # these are all opcode fields involved in index-selection of CR,
573 # and need to do "standard" arithmetic. CR[BA+32] for example
574 # would, if using SelectableInt, only be 5-bit.
575 if name
in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
576 self
.namespace
[name
] = val
578 self
.namespace
[name
] = SelectableInt(val
, sig
.width
)
580 self
.namespace
['XER'] = self
.spr
['XER']
581 self
.namespace
['CA'] = self
.spr
['XER'][XER_bits
['CA']].value
582 self
.namespace
['CA32'] = self
.spr
['XER'][XER_bits
['CA32']].value
584 def handle_carry_(self
, inputs
, outputs
, already_done
):
585 inv_a
= yield self
.dec2
.e
.do
.invert_in
587 inputs
[0] = ~inputs
[0]
589 imm_ok
= yield self
.dec2
.e
.do
.imm_data
.ok
591 imm
= yield self
.dec2
.e
.do
.imm_data
.data
592 inputs
.append(SelectableInt(imm
, 64))
593 assert len(outputs
) >= 1
594 print("outputs", repr(outputs
))
595 if isinstance(outputs
, list) or isinstance(outputs
, tuple):
601 print("gt input", x
, output
)
602 gt
= (gtu(x
, output
))
605 cy
= 1 if any(gts
) else 0
607 if not (1 & already_done
):
608 self
.spr
['XER'][XER_bits
['CA']] = cy
610 print("inputs", already_done
, inputs
)
612 # ARGH... different for OP_ADD... *sigh*...
613 op
= yield self
.dec2
.e
.do
.insn_type
614 if op
== MicrOp
.OP_ADD
.value
:
615 res32
= (output
.value
& (1 << 32)) != 0
616 a32
= (inputs
[0].value
& (1 << 32)) != 0
618 b32
= (inputs
[1].value
& (1 << 32)) != 0
621 cy32
= res32 ^ a32 ^ b32
622 print("CA32 ADD", cy32
)
626 print("input", x
, output
)
627 print(" x[32:64]", x
, x
[32:64])
628 print(" o[32:64]", output
, output
[32:64])
629 gt
= (gtu(x
[32:64], output
[32:64])) == SelectableInt(1, 1)
631 cy32
= 1 if any(gts
) else 0
632 print("CA32", cy32
, gts
)
633 if not (2 & already_done
):
634 self
.spr
['XER'][XER_bits
['CA32']] = cy32
636 def handle_overflow(self
, inputs
, outputs
, div_overflow
):
637 if hasattr(self
.dec2
.e
.do
, "invert_in"):
638 inv_a
= yield self
.dec2
.e
.do
.invert_in
640 inputs
[0] = ~inputs
[0]
642 imm_ok
= yield self
.dec2
.e
.do
.imm_data
.ok
644 imm
= yield self
.dec2
.e
.do
.imm_data
.data
645 inputs
.append(SelectableInt(imm
, 64))
646 assert len(outputs
) >= 1
647 print("handle_overflow", inputs
, outputs
, div_overflow
)
648 if len(inputs
) < 2 and div_overflow
is None:
651 # div overflow is different: it's returned by the pseudo-code
652 # because it's more complex than can be done by analysing the output
653 if div_overflow
is not None:
654 ov
, ov32
= div_overflow
, div_overflow
655 # arithmetic overflow can be done by analysing the input and output
656 elif len(inputs
) >= 2:
660 input_sgn
= [exts(x
.value
, x
.bits
) < 0 for x
in inputs
]
661 output_sgn
= exts(output
.value
, output
.bits
) < 0
662 ov
= 1 if input_sgn
[0] == input_sgn
[1] and \
663 output_sgn
!= input_sgn
[0] else 0
666 input32_sgn
= [exts(x
.value
, 32) < 0 for x
in inputs
]
667 output32_sgn
= exts(output
.value
, 32) < 0
668 ov32
= 1 if input32_sgn
[0] == input32_sgn
[1] and \
669 output32_sgn
!= input32_sgn
[0] else 0
671 self
.spr
['XER'][XER_bits
['OV']] = ov
672 self
.spr
['XER'][XER_bits
['OV32']] = ov32
673 so
= self
.spr
['XER'][XER_bits
['SO']]
675 self
.spr
['XER'][XER_bits
['SO']] = so
677 def handle_comparison(self
, outputs
, cr_idx
=0):
679 assert isinstance(out
, SelectableInt
), \
680 "out zero not a SelectableInt %s" % repr(outputs
)
681 print("handle_comparison", out
.bits
, hex(out
.value
))
682 # TODO - XXX *processor* in 32-bit mode
683 # https://bugs.libre-soc.org/show_bug.cgi?id=424
685 # o32 = exts(out.value, 32)
686 # print ("handle_comparison exts 32 bit", hex(o32))
687 out
= exts(out
.value
, out
.bits
)
688 print("handle_comparison exts", hex(out
))
689 zero
= SelectableInt(out
== 0, 1)
690 positive
= SelectableInt(out
> 0, 1)
691 negative
= SelectableInt(out
< 0, 1)
692 SO
= self
.spr
['XER'][XER_bits
['SO']]
693 print("handle_comparison SO", SO
)
694 cr_field
= selectconcat(negative
, positive
, zero
, SO
)
695 self
.crl
[cr_idx
].eq(cr_field
)
697 def set_pc(self
, pc_val
):
698 self
.namespace
['NIA'] = SelectableInt(pc_val
, 64)
699 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
702 """set up one instruction
705 pc
= self
.pc
.CIA
.value
709 ins
= self
.imem
.ld(pc
, 4, False, True, instr_fetch
=True)
711 raise KeyError("no instruction at 0x%x" % pc
)
712 print("setup: 0x%x 0x%x %s" % (pc
, ins
& 0xffffffff, bin(ins
)))
713 print("CIA NIA", self
.respect_pc
, self
.pc
.CIA
.value
, self
.pc
.NIA
.value
)
715 yield self
.dec2
.sv_rm
.eq(0)
716 yield self
.dec2
.dec
.raw_opcode_in
.eq(ins
& 0xffffffff)
717 yield self
.dec2
.dec
.bigendian
.eq(self
.bigendian
)
718 yield self
.dec2
.state
.msr
.eq(self
.msr
.value
)
719 yield self
.dec2
.state
.pc
.eq(pc
)
720 if self
.svstate
is not None:
721 yield self
.dec2
.state
.svstate
.eq(self
.svstate
.spr
.value
)
723 # SVP64. first, check if the opcode is EXT001, and SVP64 id bits set
725 opcode
= yield self
.dec2
.dec
.opcode_in
726 pfx
= SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
727 pfx
.insn
.value
= opcode
728 major
= pfx
.major
.asint(msb0
=True) # MSB0 inversion
729 print ("prefix test: opcode:", major
, bin(major
),
730 pfx
.insn
[7] == 0b1, pfx
.insn
[9] == 0b1)
731 self
.is_svp64_mode
= ((major
== 0b000001) and
732 pfx
.insn
[7].value
== 0b1 and
733 pfx
.insn
[9].value
== 0b1)
734 self
.pc
.update_nia(self
.is_svp64_mode
)
735 self
.namespace
['NIA'] = self
.pc
.NIA
736 self
.namespace
['SVSTATE'] = self
.svstate
.spr
737 if not self
.is_svp64_mode
:
740 # in SVP64 mode. decode/print out svp64 prefix, get v3.0B instruction
741 print ("svp64.rm", bin(pfx
.rm
.asint(msb0
=True)))
742 print (" svstate.vl", self
.svstate
.vl
.asint(msb0
=True))
743 print (" svstate.mvl", self
.svstate
.maxvl
.asint(msb0
=True))
744 sv_rm
= pfx
.rm
.asint(msb0
=True)
745 ins
= self
.imem
.ld(pc
+4, 4, False, True, instr_fetch
=True)
746 print(" svsetup: 0x%x 0x%x %s" % (pc
+4, ins
& 0xffffffff, bin(ins
)))
747 yield self
.dec2
.dec
.raw_opcode_in
.eq(ins
& 0xffffffff) # v3.0B suffix
748 yield self
.dec2
.sv_rm
.eq(sv_rm
) # svp64 prefix
751 def execute_one(self
):
752 """execute one instruction
754 # get the disassembly code for this instruction
755 if self
.is_svp64_mode
:
756 code
= self
.disassembly
[self
._pc
+4]
757 print(" svp64 sim-execute", hex(self
._pc
), code
)
759 code
= self
.disassembly
[self
._pc
]
760 print("sim-execute", hex(self
._pc
), code
)
761 opname
= code
.split(' ')[0]
762 yield from self
.call(opname
)
764 # don't use this except in special circumstances
765 if not self
.respect_pc
:
768 print("execute one, CIA NIA", self
.pc
.CIA
.value
, self
.pc
.NIA
.value
)
770 def get_assembly_name(self
):
771 # TODO, asmregs is from the spec, e.g. add RT,RA,RB
772 # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
773 dec_insn
= yield self
.dec2
.e
.do
.insn
774 asmcode
= yield self
.dec2
.dec
.op
.asmcode
775 print("get assembly name asmcode", asmcode
, hex(dec_insn
))
776 asmop
= insns
.get(asmcode
, None)
777 int_op
= yield self
.dec2
.dec
.op
.internal_op
779 # sigh reconstruct the assembly instruction name
780 if hasattr(self
.dec2
.e
.do
, "oe"):
781 ov_en
= yield self
.dec2
.e
.do
.oe
.oe
782 ov_ok
= yield self
.dec2
.e
.do
.oe
.ok
786 if hasattr(self
.dec2
.e
.do
, "rc"):
787 rc_en
= yield self
.dec2
.e
.do
.rc
.rc
788 rc_ok
= yield self
.dec2
.e
.do
.rc
.ok
792 # grrrr have to special-case MUL op (see DecodeOE)
793 print("ov %d en %d rc %d en %d op %d" %
794 (ov_ok
, ov_en
, rc_ok
, rc_en
, int_op
))
795 if int_op
in [MicrOp
.OP_MUL_H64
.value
, MicrOp
.OP_MUL_H32
.value
]:
800 if not asmop
.endswith("."): # don't add "." to "andis."
803 if hasattr(self
.dec2
.e
.do
, "lk"):
804 lk
= yield self
.dec2
.e
.do
.lk
807 print("int_op", int_op
)
808 if int_op
in [MicrOp
.OP_B
.value
, MicrOp
.OP_BC
.value
]:
809 AA
= yield self
.dec2
.dec
.fields
.FormI
.AA
[0:-1]
813 spr_msb
= yield from self
.get_spr_msb()
814 if int_op
== MicrOp
.OP_MFCR
.value
:
819 # XXX TODO: for whatever weird reason this doesn't work
820 # https://bugs.libre-soc.org/show_bug.cgi?id=390
821 if int_op
== MicrOp
.OP_MTCRF
.value
:
828 def get_spr_msb(self
):
829 dec_insn
= yield self
.dec2
.e
.do
.insn
830 return dec_insn
& (1 << 20) != 0 # sigh - XFF.spr[-1]?
832 def call(self
, name
):
833 """call(opcode) - the primary execution point for instructions
835 name
= name
.strip() # remove spaces if not already done so
837 print("halted - not executing", name
)
840 # TODO, asmregs is from the spec, e.g. add RT,RA,RB
841 # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
842 asmop
= yield from self
.get_assembly_name()
843 print("call", name
, asmop
)
846 int_op
= yield self
.dec2
.dec
.op
.internal_op
847 spr_msb
= yield from self
.get_spr_msb()
849 instr_is_privileged
= False
850 if int_op
in [MicrOp
.OP_ATTN
.value
,
851 MicrOp
.OP_MFMSR
.value
,
852 MicrOp
.OP_MTMSR
.value
,
853 MicrOp
.OP_MTMSRD
.value
,
855 MicrOp
.OP_RFID
.value
]:
856 instr_is_privileged
= True
857 if int_op
in [MicrOp
.OP_MFSPR
.value
,
858 MicrOp
.OP_MTSPR
.value
] and spr_msb
:
859 instr_is_privileged
= True
861 print("is priv", instr_is_privileged
, hex(self
.msr
.value
),
863 # check MSR priv bit and whether op is privileged: if so, throw trap
864 if instr_is_privileged
and self
.msr
[MSRb
.PR
] == 1:
865 self
.TRAP(0x700, PIb
.PRIV
)
866 self
.namespace
['NIA'] = self
.trap_nia
867 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
870 # check halted condition
875 # check illegal instruction
877 if name
not in ['mtcrf', 'mtocrf']:
878 illegal
= name
!= asmop
880 # sigh deal with setvl not being supported by binutils (.long)
881 if asmop
.startswith('setvl'):
886 print("illegal", name
, asmop
)
887 self
.TRAP(0x700, PIb
.ILLEG
)
888 self
.namespace
['NIA'] = self
.trap_nia
889 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
890 print("name %s != %s - calling ILLEGAL trap, PC: %x" %
891 (name
, asmop
, self
.pc
.CIA
.value
))
894 info
= self
.instrs
[name
]
895 yield from self
.prep_namespace(info
.form
, info
.op_fields
)
897 # preserve order of register names
898 input_names
= create_args(list(info
.read_regs
) +
899 list(info
.uninit_regs
))
902 # get SVP64 entry for the current instruction
903 sv_rm
= self
.svp64rm
.instrs
.get(name
)
904 if sv_rm
is not None:
905 dest_cr
, src_cr
, src_byname
, dest_byname
= decode_extra(sv_rm
)
907 dest_cr
, src_cr
, src_byname
, dest_byname
= False, False, {}, {}
908 print ("sv rm", sv_rm
, dest_cr
, src_cr
, src_byname
, dest_byname
)
910 # get SVSTATE VL (oh and print out some debug stuff)
911 if self
.is_svp64_mode
:
912 vl
= self
.svstate
.vl
.asint(msb0
=True)
913 srcstep
= self
.svstate
.srcstep
.asint(msb0
=True)
914 dststep
= self
.svstate
.srcstep
.asint(msb0
=True)
915 sv_a_nz
= yield self
.dec2
.sv_a_nz
916 in1
= yield self
.dec2
.e
.read_reg1
.data
917 print ("SVP64: VL, srcstep, dststep, sv_a_nz, in1",
918 vl
, srcstep
, dststep
, sv_a_nz
, in1
)
921 srcmask
= dstmask
= 0xffff_ffff_ffff_ffff
922 if self
.is_svp64_mode
:
923 pmode
= yield self
.dec2
.rm_dec
.predmode
924 sv_ptype
= yield self
.dec2
.dec
.op
.SV_Ptype
925 srcpred
= yield self
.dec2
.rm_dec
.srcpred
926 dstpred
= yield self
.dec2
.rm_dec
.dstpred
927 if pmode
== SVP64PredMode
.INT
.value
:
928 srcmask
= dstmask
= get_predint(self
.gpr
, dstpred
)
929 if sv_ptype
== SVPtype
.P2
.value
:
930 srcmask
= get_predint(self
.gpr
, srcpred
)
931 elif pmode
== SVP64PredMode
.CR
.value
:
932 srcmask
= dstmask
= get_predcr(self
.crl
, dstpred
, vl
)
933 if sv_ptype
== SVPtype
.P2
.value
:
934 srcmask
= get_predcr(self
.crl
, srcpred
, vl
)
935 print (" pmode", pmode
)
936 print (" ptype", sv_ptype
)
937 print (" srcmask", bin(srcmask
))
938 print (" dstmask", bin(dstmask
))
940 # okaaay, so here we simply advance srcstep (TODO dststep)
941 # until the predicate mask has a "1" bit... or we run out of VL
942 # let srcstep==VL be the indicator to move to next instruction
943 while (((1<<srcstep
) & srcmask
) == 0) and (srcstep
!= vl
):
944 print (" skip", bin(1<<srcstep
))
947 while (((1<<dststep
) & dstmask
) == 0) and (dststep
!= vl
):
948 print (" skip", bin(1<<dststep
))
951 # update SVSTATE with new srcstep
952 self
.svstate
.srcstep
[0:7] = srcstep
953 self
.svstate
.dststep
[0:7] = dststep
954 self
.namespace
['SVSTATE'] = self
.svstate
.spr
955 yield self
.dec2
.state
.svstate
.eq(self
.svstate
.spr
.value
)
956 yield Settle() # let decoder update
957 srcstep
= self
.svstate
.srcstep
.asint(msb0
=True)
958 dststep
= self
.svstate
.dststep
.asint(msb0
=True)
959 print (" srcstep", srcstep
)
960 print (" dststep", dststep
)
962 # check if end reached (we let srcstep overrun, above)
963 # nothing needs doing (TODO zeroing): just do next instruction
965 self
.svp64_reset_loop()
966 self
.update_pc_next()
969 # VL=0 in SVP64 mode means "do nothing: skip instruction"
970 if self
.is_svp64_mode
and vl
== 0:
971 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
972 print("SVP64: VL=0, end of call", self
.namespace
['CIA'],
973 self
.namespace
['NIA'])
976 # main input registers (RT, RA ...)
978 for name
in input_names
:
979 # using PowerDecoder2, first, find the decoder index.
980 # (mapping name RA RB RC RS to in1, in2, in3)
981 regnum
, is_vec
= yield from get_pdecode_idx_in(self
.dec2
, name
)
983 # doing this is not part of svp64, it's because output
984 # registers, to be modified, need to be in the namespace.
985 regnum
, is_vec
= yield from get_pdecode_idx_out(self
.dec2
, name
)
987 # in case getting the register number is needed, _RA, _RB
989 self
.namespace
[regname
] = regnum
990 print('reading reg %s %s' % (name
, str(regnum
)), is_vec
)
991 reg_val
= self
.gpr(regnum
)
992 inputs
.append(reg_val
)
994 # "special" registers
995 for special
in info
.special_regs
:
996 if special
in special_sprs
:
997 inputs
.append(self
.spr
[special
])
999 inputs
.append(self
.namespace
[special
])
1001 # clear trap (trap) NIA
1002 self
.trap_nia
= None
1004 # execute actual instruction here
1005 print("inputs", inputs
)
1006 results
= info
.func(self
, *inputs
)
1007 print("results", results
)
1009 # "inject" decorator takes namespace from function locals: we need to
1010 # overwrite NIA being overwritten (sigh)
1011 if self
.trap_nia
is not None:
1012 self
.namespace
['NIA'] = self
.trap_nia
1014 print("after func", self
.namespace
['CIA'], self
.namespace
['NIA'])
1016 # detect if CA/CA32 already in outputs (sra*, basically)
1019 output_names
= create_args(info
.write_regs
)
1020 for name
in output_names
:
1026 print("carry already done?", bin(already_done
))
1027 if hasattr(self
.dec2
.e
.do
, "output_carry"):
1028 carry_en
= yield self
.dec2
.e
.do
.output_carry
1032 yield from self
.handle_carry_(inputs
, results
, already_done
)
1034 # detect if overflow was in return result
1037 for name
, output
in zip(output_names
, results
):
1038 if name
== 'overflow':
1041 if hasattr(self
.dec2
.e
.do
, "oe"):
1042 ov_en
= yield self
.dec2
.e
.do
.oe
.oe
1043 ov_ok
= yield self
.dec2
.e
.do
.oe
.ok
1047 print("internal overflow", overflow
, ov_en
, ov_ok
)
1049 yield from self
.handle_overflow(inputs
, results
, overflow
)
1051 if hasattr(self
.dec2
.e
.do
, "rc"):
1052 rc_en
= yield self
.dec2
.e
.do
.rc
.rc
1056 regnum
, is_vec
= yield from get_pdecode_cr_out(self
.dec2
, "CR0")
1057 self
.handle_comparison(results
, regnum
)
1059 # any modified return results?
1061 for name
, output
in zip(output_names
, results
):
1062 if name
== 'overflow': # ignore, done already (above)
1064 if isinstance(output
, int):
1065 output
= SelectableInt(output
, 256)
1066 if name
in ['CA', 'CA32']:
1068 print("writing %s to XER" % name
, output
)
1069 self
.spr
['XER'][XER_bits
[name
]] = output
.value
1071 print("NOT writing %s to XER" % name
, output
)
1072 elif name
in info
.special_regs
:
1073 print('writing special %s' % name
, output
, special_sprs
)
1074 if name
in special_sprs
:
1075 self
.spr
[name
] = output
1077 self
.namespace
[name
].eq(output
)
1079 print('msr written', hex(self
.msr
.value
))
1081 regnum
, is_vec
= yield from get_pdecode_idx_out(self
.dec2
,
1084 # temporary hack for not having 2nd output
1085 regnum
= yield getattr(self
.decoder
, name
)
1087 print('writing reg %d %s' % (regnum
, str(output
)), is_vec
)
1088 if output
.bits
> 64:
1089 output
= SelectableInt(output
.value
, 64)
1090 self
.gpr
[regnum
] = output
1092 # check if it is the SVSTATE.src/dest step that needs incrementing
1093 # this is our Sub-Program-Counter loop from 0 to VL-1
1094 if self
.is_svp64_mode
:
1095 # XXX twin predication TODO
1096 vl
= self
.svstate
.vl
.asint(msb0
=True)
1097 mvl
= self
.svstate
.maxvl
.asint(msb0
=True)
1098 srcstep
= self
.svstate
.srcstep
.asint(msb0
=True)
1099 dststep
= self
.svstate
.srcstep
.asint(msb0
=True)
1100 sv_ptype
= yield self
.dec2
.dec
.op
.SV_Ptype
1101 no_out_vec
= not (yield self
.dec2
.no_out_vec
)
1102 no_in_vec
= not (yield self
.dec2
.no_in_vec
)
1103 print (" svstate.vl", vl
)
1104 print (" svstate.mvl", mvl
)
1105 print (" svstate.srcstep", srcstep
)
1106 print (" svstate.dststep", dststep
)
1107 print (" no_out_vec", no_out_vec
)
1108 print (" no_in_vec", no_in_vec
)
1109 print (" sv_ptype", sv_ptype
, sv_ptype
== SVPtype
.P2
.value
)
1110 # check if srcstep needs incrementing by one, stop PC advancing
1111 # svp64 loop can end early if the dest is scalar for single-pred
1112 # but for 2-pred both src/dest have to be checked.
1113 # XXX this might not be true! it may just be LD/ST
1114 if sv_ptype
== SVPtype
.P2
.value
:
1115 svp64_is_vector
= (no_out_vec
or no_in_vec
)
1117 svp64_is_vector
= no_out_vec
1118 if svp64_is_vector
and srcstep
!= vl
-1:
1119 self
.svstate
.srcstep
+= SelectableInt(1, 7)
1120 self
.svstate
.dststep
+= SelectableInt(1, 7)
1121 self
.pc
.NIA
.value
= self
.pc
.CIA
.value
1122 self
.namespace
['NIA'] = self
.pc
.NIA
1123 self
.namespace
['SVSTATE'] = self
.svstate
.spr
1124 print("end of sub-pc call", self
.namespace
['CIA'],
1125 self
.namespace
['NIA'])
1126 return # DO NOT allow PC to update whilst Sub-PC loop running
1127 # reset loop to zero
1128 self
.svp64_reset_loop()
1130 self
.update_pc_next()
1132 def update_pc_next(self
):
1133 # UPDATE program counter
1134 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1135 self
.svstate
.spr
= self
.namespace
['SVSTATE']
1136 print("end of call", self
.namespace
['CIA'],
1137 self
.namespace
['NIA'],
1138 self
.namespace
['SVSTATE'])
1140 def svp64_reset_loop(self
):
1141 self
.svstate
.srcstep
[0:7] = 0
1142 self
.svstate
.dststep
[0:7] = 0
1143 print (" svstate.srcstep loop end (PC to update)")
1144 self
.pc
.update_nia(self
.is_svp64_mode
)
1145 self
.namespace
['NIA'] = self
.pc
.NIA
1146 self
.namespace
['SVSTATE'] = self
.svstate
.spr
1149 """Decorator factory.
1151 this decorator will "inject" variables into the function's namespace,
1152 from the *dictionary* in self.namespace. it therefore becomes possible
1153 to make it look like a whole stack of variables which would otherwise
1154 need "self." inserted in front of them (*and* for those variables to be
1155 added to the instance) "appear" in the function.
1157 "self.namespace['SI']" for example becomes accessible as just "SI" but
1158 *only* inside the function, when decorated.
1160 def variable_injector(func
):
1162 def decorator(*args
, **kwargs
):
1164 func_globals
= func
.__globals
__ # Python 2.6+
1165 except AttributeError:
1166 func_globals
= func
.func_globals
# Earlier versions.
1168 context
= args
[0].namespace
# variables to be injected
1169 saved_values
= func_globals
.copy() # Shallow copy of dict.
1170 func_globals
.update(context
)
1171 result
= func(*args
, **kwargs
)
1172 print("globals after", func_globals
['CIA'], func_globals
['NIA'])
1173 print("args[0]", args
[0].namespace
['CIA'],
1174 args
[0].namespace
['NIA'],
1175 args
[0].namespace
['SVSTATE'])
1176 args
[0].namespace
= func_globals
1177 #exec (func.__code__, func_globals)
1180 # func_globals = saved_values # Undo changes.
1186 return variable_injector