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 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
29 from collections
import namedtuple
33 instruction_info
= namedtuple('instruction_info',
34 'func read_regs uninit_regs write_regs ' +
35 'special_regs op_fields form asmregs')
45 def swap_order(x
, nbytes
):
46 x
= x
.to_bytes(nbytes
, byteorder
='little')
47 x
= int.from_bytes(x
, byteorder
='big', signed
=False)
52 # TODO (lkcl): adjust other registers that should be in a particular order
53 # probably CA, CA32, and CR
70 def create_args(reglist
, extra
=None):
71 retval
= list(OrderedSet(reglist
))
72 retval
.sort(key
=lambda reg
: REG_SORT_ORDER
[reg
])
74 return [extra
] + retval
78 # very quick, TODO move to SelectableInt utils later
79 def genmask(shift
, size
):
80 res
= SelectableInt(0, size
)
83 res
[size
-1-i
] = SelectableInt(1, 1)
89 //Accessing 2nd double word of partition table (pate1)
90 //Ref: Power ISA Manual v3.0B, Book-III, section 5.7.6.1
92 // ====================================================
93 // -----------------------------------------------
94 // | /// | PATB | /// | PATS |
95 // -----------------------------------------------
97 // PATB[4:51] holds the base address of the Partition Table,
98 // right shifted by 12 bits.
99 // This is because the address of the Partition base is
100 // 4k aligned. Hence, the lower 12bits, which are always
101 // 0 are ommitted from the PTCR.
103 // Thus, The Partition Table Base is obtained by (PATB << 12)
105 // PATS represents the partition table size right-shifted by 12 bits.
106 // The minimal size of the partition table is 4k.
107 // Thus partition table size = (1 << PATS + 12).
110 // ====================================================
111 // 0 PATE0 63 PATE1 127
112 // |----------------------|----------------------|
114 // |----------------------|----------------------|
116 // |----------------------|----------------------|
118 // |----------------------|----------------------|
122 // |----------------------|----------------------|
124 // |----------------------|----------------------|
126 // The effective LPID forms the index into the Partition Table.
128 // Each entry in the partition table contains 2 double words, PATE0, PATE1,
129 // corresponding to that partition.
131 // In case of Radix, The structure of PATE0 and PATE1 is as follows.
134 // -----------------------------------------------
135 // |1|RTS1|/| RPDB | RTS2 | RPDS |
136 // -----------------------------------------------
137 // 0 1 2 3 4 55 56 58 59 63
139 // HR[0] : For Radix Page table, first bit should be 1.
140 // RTS1[1:2] : Gives one fragment of the Radix treesize
141 // RTS2[56:58] : Gives the second fragment of the Radix Tree size.
142 // RTS = (RTS1 << 3 + RTS2) + 31.
144 // RPDB[4:55] = Root Page Directory Base.
145 // RPDS = Logarithm of Root Page Directory Size right shifted by 3.
146 // Thus, Root page directory size = 1 << (RPDS + 3).
150 // -----------------------------------------------
151 // |///| PRTB | // | PRTS |
152 // -----------------------------------------------
153 // 0 3 4 51 52 58 59 63
155 // PRTB[4:51] = Process Table Base. This is aligned to size.
156 // PRTS[59: 63] = Process Table Size right shifted by 12.
157 // Minimal size of the process table is 4k.
158 // Process Table Size = (1 << PRTS + 12).
161 // Computing the size aligned Process Table Base:
162 // table_base = (PRTB & ~((1 << PRTS) - 1)) << 12
163 // Thus, the lower 12+PRTS bits of table_base will
167 //Ref: Power ISA Manual v3.0B, Book-III, section 5.7.6.2
170 // ==========================
171 // 0 PRTE0 63 PRTE1 127
172 // |----------------------|----------------------|
174 // |----------------------|----------------------|
176 // |----------------------|----------------------|
178 // |----------------------|----------------------|
182 // |----------------------|----------------------|
184 // |----------------------|----------------------|
186 // The effective Process id (PID) forms the index into the Process Table.
188 // Each entry in the partition table contains 2 double words, PRTE0, PRTE1,
189 // corresponding to that process
191 // In case of Radix, The structure of PRTE0 and PRTE1 is as follows.
194 // -----------------------------------------------
195 // |/|RTS1|/| RPDB | RTS2 | RPDS |
196 // -----------------------------------------------
197 // 0 1 2 3 4 55 56 58 59 63
199 // RTS1[1:2] : Gives one fragment of the Radix treesize
200 // RTS2[56:58] : Gives the second fragment of the Radix Tree size.
201 // RTS = (RTS1 << 3 + RTS2) << 31,
202 // since minimal Radix Tree size is 4G.
204 // RPDB = Root Page Directory Base.
205 // RPDS = Root Page Directory Size right shifted by 3.
206 // Thus, Root page directory size = RPDS << 3.
210 // -----------------------------------------------
212 // -----------------------------------------------
214 // All bits are reserved.
219 # see qemu/target/ppc/mmu-radix64.c for reference
221 def __init__(self
, mem
, caller
):
225 # cached page table stuff
227 self
.pt0_valid
= False
229 self
.pt3_valid
= False
231 def __call__(self
,*args
, **kwargs
):
232 print("TODO: implement RADIX.__call__()")
237 def ld(self
, address
, width
=8, swap
=True, check_in_mem
=False):
238 print("RADIX: ld from addr 0x%x width %d" % (address
, width
))
240 pte
= self
._walk
_tree
()
241 # use pte to caclculate phys address
242 return self
.mem
.ld(address
, width
, swap
, check_in_mem
)
245 def st(self
, addr
, v
, width
=8, swap
=True):
246 print("RADIX: st to addr 0x%x width %d data %x" % (addr
, width
, v
))
247 # use pte to caclculate phys address (addr)
248 return self
.mem
.st(addr
, v
, width
, swap
)
250 def memassign(self
, addr
, sz
, val
):
251 print("memassign", addr
, sz
, val
)
252 self
.st(addr
.value
, val
.value
, sz
, swap
=False)
254 def _next_level(self
):
259 ## Prepare for next iteration
261 def _walk_tree(self
):
265 // vaddr |-----------------------------------------------------|
267 // |-----------|-----------------------------------------|
268 // | 0000000 | usefulBits = X bits (typically 52) |
269 // |-----------|-----------------------------------------|
270 // | |<--Cursize---->| |
274 // |-----------------------------------------------------|
277 // PDE |---------------------------| |
278 // |V|L|//| NLB |///|NLS| |
279 // |---------------------------| |
280 // PDE = Page Directory Entry |
281 // [0] = V = Valid Bit |
282 // [1] = L = Leaf bit. If 0, then |
283 // [4:55] = NLB = Next Level Base |
284 // right shifted by 8 |
285 // [59:63] = NLS = Next Level Size |
288 // | |--------------------------|
289 // | | usfulBits = X-Cursize |
290 // | |--------------------------|
291 // |---------------------><--NLS-->| |
295 // |--------------------------|
297 // If the next PDE obtained by |
298 // (NLB << 8 + 8 * index) is a |
299 // nonleaf, then repeat the above. |
301 // If the next PDE is a leaf, |
302 // then Leaf PDE structure is as |
307 // |------------------------------| |----------------|
308 // |V|L|sw|//|RPN|sw|R|C|/|ATT|EAA| | usefulBits |
309 // |------------------------------| |----------------|
310 // [0] = V = Valid Bit |
311 // [1] = L = Leaf Bit = 1 if leaf |
313 // [2] = Sw = Sw bit 0. |
314 // [7:51] = RPN = Real Page Number, V
315 // real_page = RPN << 12 -------------> Logical OR
316 // [52:54] = Sw Bits 1:3 |
317 // [55] = R = Reference |
318 // [56] = C = Change V
319 // [58:59] = Att = Physical Address
320 // 0b00 = Normal Memory
322 // 0b10 = Non Idenmpotent
323 // 0b11 = Tolerant I/O
324 // [60:63] = Encoded Access
328 # walk tree starts on prtbl
330 ret
= self
._next
_level
()
333 def _decode_prte(self
, data
):
335 -----------------------------------------------
336 |/|RTS1|/| RPDB | RTS2 | RPDS |
337 -----------------------------------------------
338 0 1 2 3 4 55 56 58 59 63
340 # note that SelectableInt does big-endian! so the indices
341 # below *directly* match the spec, unlike microwatt which
342 # has to turn them around (to LE)
343 zero
= SelectableInt(0, 1)
344 rts
= selectconcat(zero
,
348 masksize
= data
[59:64] # RPDS
349 mbits
= selectconcat(zero
, masksize
)
350 pgbase
= selectconcat(data
[8:56], # part of RPDB
351 SelectableInt(0, 16),)
353 return (rts
, mbits
, pgbase
)
355 def _segment_check(self
, addr
, mbits
, shift
):
356 """checks segment valid
357 mbits := '0' & r.mask_size;
358 v.shift := r.shift + (31 - 12) - mbits;
359 nonzero := or(r.addr(61 downto 31) and not finalmask(30 downto 0));
360 if r.addr(63) /= r.addr(62) or nonzero = '1' then
361 v.state := RADIX_FINISH;
363 elsif mbits < 5 or mbits > 16 or mbits > (r.shift + (31 - 12)) then
364 v.state := RADIX_FINISH;
367 v.state := RADIX_LOOKUP;
369 # note that SelectableInt does big-endian! so the indices
370 # below *directly* match the spec, unlike microwatt which
371 # has to turn them around (to LE)
372 mask
= genmask(shift
, 43)
373 nonzero
= addr
[1:32] & mask
[12:43] # mask 31 LSBs (BE numbered 12:43)
374 print ("RADIX _segment_check nonzero", bin(nonzero
.value
))
375 print ("RADIX _segment_check addr[0-1]", addr
[0].value
, addr
[1].value
)
376 if addr
[0] != addr
[1] or nonzero
== 1:
378 limit
= shift
+ (31 - 12)
379 if mbits
< 5 or mbits
> 16 or mbits
> limit
:
381 new_shift
= shift
+ (31 - 12) - mbits
384 def _check_perms(self
):
385 """check page permissions
387 if data(62) = '1' then
388 -- check permissions and RC bits
390 if r.priv = '1' or data(3) = '0' then
391 if r.iside = '0' then
392 perm_ok := data(1) or (data(2) and not r.store);
394 -- no IAMR, so no KUEP support for now
395 -- deny execute permission if cache inhibited
396 perm_ok := data(0) and not data(5);
399 rc_ok := data(8) and (data(7) or not r.store);
400 if perm_ok = '1' and rc_ok = '1' then
401 v.state := RADIX_LOAD_TLB;
403 v.state := RADIX_FINISH;
404 v.perm_err := not perm_ok;
405 -- permission error takes precedence over RC error
406 v.rc_error := perm_ok;
410 def _get_prtable_addr(self
, prtbl
, addr
):
412 if r.addr(63) = '1' then
413 effpid := x"00000000";
417 prtable_addr := x"00" & r.prtbl(55 downto 36) &
418 ((r.prtbl(35 downto 12) and not finalmask(23 downto 0)) or
419 (effpid(31 downto 8) and finalmask(23 downto 0))) &
420 effpid(7 downto 0) & "0000";
423 def _get_pgtable_addr(self
):
425 pgtable_addr := x"00" & r.pgbase(55 downto 19) &
426 ((r.pgbase(18 downto 3) and not mask) or (addrsh and mask)) &
433 ((r.pde(55 downto 12) and not finalmask) or (r.addr(55 downto 12) and finalmask))
434 & r.pde(11 downto 0);
440 def __init__(self
, row_bytes
=8, initial_mem
=None):
442 self
.bytes_per_word
= row_bytes
443 self
.word_log2
= math
.ceil(math
.log2(row_bytes
))
444 print("Sim-Mem", initial_mem
, self
.bytes_per_word
, self
.word_log2
)
448 # different types of memory data structures recognised (for convenience)
449 if isinstance(initial_mem
, list):
450 initial_mem
= (0, initial_mem
)
451 if isinstance(initial_mem
, tuple):
452 startaddr
, mem
= initial_mem
454 for i
, val
in enumerate(mem
):
455 initial_mem
[startaddr
+ row_bytes
*i
] = (val
, row_bytes
)
457 for addr
, (val
, width
) in initial_mem
.items():
458 #val = swap_order(val, width)
459 self
.st(addr
, val
, width
, swap
=False)
461 def _get_shifter_mask(self
, wid
, remainder
):
462 shifter
= ((self
.bytes_per_word
- wid
) - remainder
) * \
464 # XXX https://bugs.libre-soc.org/show_bug.cgi?id=377
466 shifter
= remainder
* 8
467 mask
= (1 << (wid
* 8)) - 1
468 print("width,rem,shift,mask", wid
, remainder
, hex(shifter
), hex(mask
))
471 # TODO: Implement ld/st of lesser width
472 def ld(self
, address
, width
=8, swap
=True, check_in_mem
=False):
473 print("ld from addr 0x{:x} width {:d}".format(address
, width
))
474 remainder
= address
& (self
.bytes_per_word
- 1)
475 address
= address
>> self
.word_log2
476 assert remainder
& (width
- 1) == 0, "Unaligned access unsupported!"
477 if address
in self
.mem
:
478 val
= self
.mem
[address
]
483 print("mem @ 0x{:x} rem {:d} : 0x{:x}".format(address
, remainder
, val
))
485 if width
!= self
.bytes_per_word
:
486 shifter
, mask
= self
._get
_shifter
_mask
(width
, remainder
)
487 print("masking", hex(val
), hex(mask
<< shifter
), shifter
)
488 val
= val
& (mask
<< shifter
)
491 val
= swap_order(val
, width
)
492 print("Read 0x{:x} from addr 0x{:x}".format(val
, address
))
495 def st(self
, addr
, v
, width
=8, swap
=True):
497 remainder
= addr
& (self
.bytes_per_word
- 1)
498 addr
= addr
>> self
.word_log2
499 print("Writing 0x{:x} to ST 0x{:x} "
500 "memaddr 0x{:x}/{:x}".format(v
, staddr
, addr
, remainder
, swap
))
501 assert remainder
& (width
- 1) == 0, "Unaligned access unsupported!"
503 v
= swap_order(v
, width
)
504 if width
!= self
.bytes_per_word
:
509 shifter
, mask
= self
._get
_shifter
_mask
(width
, remainder
)
510 val
&= ~
(mask
<< shifter
)
515 print("mem @ 0x{:x}: 0x{:x}".format(addr
, self
.mem
[addr
]))
517 def __call__(self
, addr
, sz
):
518 val
= self
.ld(addr
.value
, sz
, swap
=False)
519 print("memread", addr
, sz
, val
)
520 return SelectableInt(val
, sz
*8)
522 def memassign(self
, addr
, sz
, val
):
523 print("memassign", addr
, sz
, val
)
524 self
.st(addr
.value
, val
.value
, sz
, swap
=False)
528 def __init__(self
, decoder
, isacaller
, svstate
, regfile
):
531 self
.isacaller
= isacaller
532 self
.svstate
= svstate
534 self
[i
] = SelectableInt(regfile
[i
], 64)
536 def __call__(self
, ridx
):
539 def set_form(self
, form
):
542 def getz(self
, rnum
):
543 # rnum = rnum.value # only SelectableInt allowed
544 print("GPR getzero", rnum
)
546 return SelectableInt(0, 64)
549 def _get_regnum(self
, attr
):
550 getform
= self
.sd
.sigforms
[self
.form
]
551 rnum
= getattr(getform
, attr
)
554 def ___getitem__(self
, attr
):
555 """ XXX currently not used
557 rnum
= self
._get
_regnum
(attr
)
558 offs
= self
.svstate
.srcstep
559 print("GPR getitem", attr
, rnum
, "srcoffs", offs
)
560 return self
.regfile
[rnum
]
563 for i
in range(0, len(self
), 8):
566 s
.append("%08x" % self
[i
+j
].value
)
568 print("reg", "%2d" % i
, s
)
572 def __init__(self
, pc_init
=0):
573 self
.CIA
= SelectableInt(pc_init
, 64)
574 self
.NIA
= self
.CIA
+ SelectableInt(4, 64) # only true for v3.0B!
576 def update_nia(self
, is_svp64
):
577 increment
= 8 if is_svp64
else 4
578 self
.NIA
= self
.CIA
+ SelectableInt(increment
, 64)
580 def update(self
, namespace
, is_svp64
):
581 """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
583 self
.CIA
= namespace
['NIA'].narrow(64)
584 self
.update_nia(is_svp64
)
585 namespace
['CIA'] = self
.CIA
586 namespace
['NIA'] = self
.NIA
589 # Simple-V: see https://libre-soc.org/openpower/sv
591 def __init__(self
, init
=0):
592 self
.spr
= SelectableInt(init
, 32)
593 # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
594 self
.maxvl
= FieldSelectableInt(self
.spr
, tuple(range(0,7)))
595 self
.vl
= FieldSelectableInt(self
.spr
, tuple(range(7,14)))
596 self
.srcstep
= FieldSelectableInt(self
.spr
, tuple(range(14,21)))
597 self
.dststep
= FieldSelectableInt(self
.spr
, tuple(range(21,28)))
598 self
.subvl
= FieldSelectableInt(self
.spr
, tuple(range(28,30)))
599 self
.svstep
= FieldSelectableInt(self
.spr
, tuple(range(30,32)))
604 def __init__(self
, init
=0):
605 self
.spr
= SelectableInt(init
, 24)
606 # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
607 self
.mmode
= FieldSelectableInt(self
.spr
, [0])
608 self
.mask
= FieldSelectableInt(self
.spr
, tuple(range(1,4)))
609 self
.elwidth
= FieldSelectableInt(self
.spr
, tuple(range(4,6)))
610 self
.ewsrc
= FieldSelectableInt(self
.spr
, tuple(range(6,8)))
611 self
.subvl
= FieldSelectableInt(self
.spr
, tuple(range(8,10)))
612 self
.extra
= FieldSelectableInt(self
.spr
, tuple(range(10,19)))
613 self
.mode
= FieldSelectableInt(self
.spr
, tuple(range(19,24)))
614 # these cover the same extra field, split into parts as EXTRA2
615 self
.extra2
= list(range(4))
616 self
.extra2
[0] = FieldSelectableInt(self
.spr
, tuple(range(10,12)))
617 self
.extra2
[1] = FieldSelectableInt(self
.spr
, tuple(range(12,14)))
618 self
.extra2
[2] = FieldSelectableInt(self
.spr
, tuple(range(14,16)))
619 self
.extra2
[3] = FieldSelectableInt(self
.spr
, tuple(range(16,18)))
620 self
.smask
= FieldSelectableInt(self
.spr
, tuple(range(16,19)))
621 # and here as well, but EXTRA3
622 self
.extra3
= list(range(3))
623 self
.extra3
[0] = FieldSelectableInt(self
.spr
, tuple(range(10,13)))
624 self
.extra3
[1] = FieldSelectableInt(self
.spr
, tuple(range(13,16)))
625 self
.extra3
[2] = FieldSelectableInt(self
.spr
, tuple(range(16,19)))
628 SVP64RM_MMODE_SIZE
= len(SVP64RMFields().mmode
.br
)
629 SVP64RM_MASK_SIZE
= len(SVP64RMFields().mask
.br
)
630 SVP64RM_ELWIDTH_SIZE
= len(SVP64RMFields().elwidth
.br
)
631 SVP64RM_EWSRC_SIZE
= len(SVP64RMFields().ewsrc
.br
)
632 SVP64RM_SUBVL_SIZE
= len(SVP64RMFields().subvl
.br
)
633 SVP64RM_EXTRA2_SPEC_SIZE
= len(SVP64RMFields().extra2
[0].br
)
634 SVP64RM_EXTRA3_SPEC_SIZE
= len(SVP64RMFields().extra3
[0].br
)
635 SVP64RM_SMASK_SIZE
= len(SVP64RMFields().smask
.br
)
636 SVP64RM_MODE_SIZE
= len(SVP64RMFields().mode
.br
)
639 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
640 class SVP64PrefixFields
:
642 self
.insn
= SelectableInt(0, 32)
643 # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
644 self
.major
= FieldSelectableInt(self
.insn
, tuple(range(0,6)))
645 self
.pid
= FieldSelectableInt(self
.insn
, (7, 9)) # must be 0b11
646 rmfields
= [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
647 self
.rm
= FieldSelectableInt(self
.insn
, rmfields
)
650 SV64P_MAJOR_SIZE
= len(SVP64PrefixFields().major
.br
)
651 SV64P_PID_SIZE
= len(SVP64PrefixFields().pid
.br
)
652 SV64P_RM_SIZE
= len(SVP64PrefixFields().rm
.br
)
656 def __init__(self
, dec2
, initial_sprs
={}):
659 for key
, v
in initial_sprs
.items():
660 if isinstance(key
, SelectableInt
):
662 key
= special_sprs
.get(key
, key
)
663 if isinstance(key
, int):
666 info
= spr_byname
[key
]
667 if not isinstance(v
, SelectableInt
):
668 v
= SelectableInt(v
, info
.length
)
671 def __getitem__(self
, key
):
672 print("get spr", key
)
673 print("dict", self
.items())
674 # if key in special_sprs get the special spr, otherwise return key
675 if isinstance(key
, SelectableInt
):
677 if isinstance(key
, int):
678 key
= spr_dict
[key
].SPR
679 key
= special_sprs
.get(key
, key
)
680 if key
== 'HSRR0': # HACK!
682 if key
== 'HSRR1': # HACK!
685 res
= dict.__getitem
__(self
, key
)
687 if isinstance(key
, int):
690 info
= spr_byname
[key
]
691 dict.__setitem
__(self
, key
, SelectableInt(0, info
.length
))
692 res
= dict.__getitem
__(self
, key
)
693 print("spr returning", key
, res
)
696 def __setitem__(self
, key
, value
):
697 if isinstance(key
, SelectableInt
):
699 if isinstance(key
, int):
700 key
= spr_dict
[key
].SPR
701 print("spr key", key
)
702 key
= special_sprs
.get(key
, key
)
703 if key
== 'HSRR0': # HACK!
704 self
.__setitem
__('SRR0', value
)
705 if key
== 'HSRR1': # HACK!
706 self
.__setitem
__('SRR1', value
)
707 print("setting spr", key
, value
)
708 dict.__setitem
__(self
, key
, value
)
710 def __call__(self
, ridx
):
713 def get_pdecode_idx_in(dec2
, name
):
715 in1_sel
= yield op
.in1_sel
716 in2_sel
= yield op
.in2_sel
717 in3_sel
= yield op
.in3_sel
718 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
719 in1
= yield dec2
.e
.read_reg1
.data
720 in2
= yield dec2
.e
.read_reg2
.data
721 in3
= yield dec2
.e
.read_reg3
.data
722 in1_isvec
= yield dec2
.in1_isvec
723 in2_isvec
= yield dec2
.in2_isvec
724 in3_isvec
= yield dec2
.in3_isvec
725 print ("get_pdecode_idx", in1_sel
, In1Sel
.RA
.value
, in1
, in1_isvec
)
726 # identify which regnames map to in1/2/3
728 if (in1_sel
== In1Sel
.RA
.value
or
729 (in1_sel
== In1Sel
.RA_OR_ZERO
.value
and in1
!= 0)):
730 return in1
, in1_isvec
731 if in1_sel
== In1Sel
.RA_OR_ZERO
.value
:
732 return in1
, in1_isvec
734 if in2_sel
== In2Sel
.RB
.value
:
735 return in2
, in2_isvec
736 if in3_sel
== In3Sel
.RB
.value
:
737 return in3
, in3_isvec
738 # XXX TODO, RC doesn't exist yet!
740 assert False, "RC does not exist yet"
742 if in1_sel
== In1Sel
.RS
.value
:
743 return in1
, in1_isvec
744 if in2_sel
== In2Sel
.RS
.value
:
745 return in2
, in2_isvec
746 if in3_sel
== In3Sel
.RS
.value
:
747 return in3
, in3_isvec
751 def get_pdecode_cr_out(dec2
, name
):
753 out_sel
= yield op
.cr_out
754 out_bitfield
= yield dec2
.dec_cr_out
.cr_bitfield
.data
755 sv_cr_out
= yield op
.sv_cr_out
756 spec
= yield dec2
.crout_svdec
.spec
757 sv_override
= yield dec2
.dec_cr_out
.sv_override
758 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
759 out
= yield dec2
.e
.write_cr
.data
760 o_isvec
= yield dec2
.o_isvec
761 print ("get_pdecode_cr_out", out_sel
, CROutSel
.CR0
.value
, out
, o_isvec
)
762 print (" sv_cr_out", sv_cr_out
)
763 print (" cr_bf", out_bitfield
)
764 print (" spec", spec
)
765 print (" override", sv_override
)
766 # identify which regnames map to out / o2
768 if out_sel
== CROutSel
.CR0
.value
:
770 print ("get_pdecode_idx_out not found", name
)
774 def get_pdecode_idx_out(dec2
, name
):
776 out_sel
= yield op
.out_sel
777 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
778 out
= yield dec2
.e
.write_reg
.data
779 o_isvec
= yield dec2
.o_isvec
780 print ("get_pdecode_idx_out", out_sel
, OutSel
.RA
.value
, out
, o_isvec
)
781 # identify which regnames map to out / o2
783 if out_sel
== OutSel
.RA
.value
:
786 if out_sel
== OutSel
.RT
.value
:
788 print ("get_pdecode_idx_out not found", name
)
793 def get_pdecode_idx_out2(dec2
, name
):
795 print ("TODO: get_pdecode_idx_out2", name
)
800 # decoder2 - an instance of power_decoder2
801 # regfile - a list of initial values for the registers
802 # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
803 # respect_pc - tracks the program counter. requires initial_insns
804 def __init__(self
, decoder2
, regfile
, initial_sprs
=None, initial_cr
=0,
805 initial_mem
=None, initial_msr
=0,
807 initial_insns
=None, respect_pc
=False,
813 self
.bigendian
= bigendian
815 self
.is_svp64_mode
= False
816 self
.respect_pc
= respect_pc
817 if initial_sprs
is None:
819 if initial_mem
is None:
821 if initial_insns
is None:
823 assert self
.respect_pc
== False, "instructions required to honor pc"
825 print("ISACaller insns", respect_pc
, initial_insns
, disassembly
)
826 print("ISACaller initial_msr", initial_msr
)
828 # "fake program counter" mode (for unit testing)
832 if isinstance(initial_mem
, tuple):
833 self
.fake_pc
= initial_mem
[0]
834 disasm_start
= self
.fake_pc
836 disasm_start
= initial_pc
838 # disassembly: we need this for now (not given from the decoder)
839 self
.disassembly
= {}
841 for i
, code
in enumerate(disassembly
):
842 self
.disassembly
[i
*4 + disasm_start
] = code
844 # set up registers, instruction memory, data memory, PC, SPRs, MSR
845 self
.svp64rm
= SVP64RM()
846 if isinstance(initial_svstate
, int):
847 initial_svstate
= SVP64State(initial_svstate
)
848 self
.svstate
= initial_svstate
849 self
.gpr
= GPR(decoder2
, self
, self
.svstate
, regfile
)
850 self
.mem
= Mem(row_bytes
=8, initial_mem
=initial_mem
)
852 self
.mem
= RADIX(self
.mem
, self
)
853 self
.imem
= Mem(row_bytes
=4, initial_mem
=initial_insns
)
855 self
.spr
= SPR(decoder2
, initial_sprs
)
856 self
.msr
= SelectableInt(initial_msr
, 64) # underlying reg
859 # FPR (same as GPR except for FP nums)
860 # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
861 # note that mffs, mcrfs, mtfsf "manage" this FPSCR
862 # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
863 # note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
865 # 2.3.2 LR (actually SPR #8) -- Done
866 # 2.3.3 CTR (actually SPR #9) -- Done
867 # 2.3.4 TAR (actually SPR #815)
868 # 3.2.2 p45 XER (actually SPR #1) -- Done
869 # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
871 # create CR then allow portions of it to be "selectable" (below)
872 #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
873 self
.cr
= SelectableInt(initial_cr
, 64) # underlying reg
874 #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
876 # "undefined", just set to variable-bit-width int (use exts "max")
877 #self.undefined = SelectableInt(0, 256) # TODO, not hard-code 256!
880 self
.namespace
.update(self
.spr
)
881 self
.namespace
.update({'GPR': self
.gpr
,
884 'memassign': self
.memassign
,
889 'undefined': undefined
,
890 'mode_is_64bit': True,
894 # update pc to requested start point
895 self
.set_pc(initial_pc
)
897 # field-selectable versions of Condition Register TODO check bitranges?
900 bits
= tuple(range(i
*4+32, (i
+1)*4+32)) # errr... maybe?
901 _cr
= FieldSelectableInt(self
.cr
, bits
)
903 self
.namespace
["CR%d" % i
] = _cr
905 self
.decoder
= decoder2
.dec
908 def TRAP(self
, trap_addr
=0x700, trap_bit
=PIb
.TRAP
):
909 print("TRAP:", hex(trap_addr
), hex(self
.namespace
['MSR'].value
))
910 # store CIA(+4?) in SRR0, set NIA to 0x700
911 # store MSR in SRR1, set MSR to um errr something, have to check spec
912 self
.spr
['SRR0'].value
= self
.pc
.CIA
.value
913 self
.spr
['SRR1'].value
= self
.namespace
['MSR'].value
914 self
.trap_nia
= SelectableInt(trap_addr
, 64)
915 self
.spr
['SRR1'][trap_bit
] = 1 # change *copy* of MSR in SRR1
917 # set exception bits. TODO: this should, based on the address
918 # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
919 # bits appropriately. however it turns out that *for now* in all
920 # cases (all trap_addrs) the exact same thing is needed.
921 self
.msr
[MSRb
.IR
] = 0
922 self
.msr
[MSRb
.DR
] = 0
923 self
.msr
[MSRb
.FE0
] = 0
924 self
.msr
[MSRb
.FE1
] = 0
925 self
.msr
[MSRb
.EE
] = 0
926 self
.msr
[MSRb
.RI
] = 0
927 self
.msr
[MSRb
.SF
] = 1
928 self
.msr
[MSRb
.TM
] = 0
929 self
.msr
[MSRb
.VEC
] = 0
930 self
.msr
[MSRb
.VSX
] = 0
931 self
.msr
[MSRb
.PR
] = 0
932 self
.msr
[MSRb
.FP
] = 0
933 self
.msr
[MSRb
.PMM
] = 0
934 self
.msr
[MSRb
.TEs
] = 0
935 self
.msr
[MSRb
.TEe
] = 0
936 self
.msr
[MSRb
.UND
] = 0
937 self
.msr
[MSRb
.LE
] = 1
939 def memassign(self
, ea
, sz
, val
):
940 self
.mem
.memassign(ea
, sz
, val
)
942 def prep_namespace(self
, formname
, op_fields
):
943 # TODO: get field names from form in decoder*1* (not decoder2)
944 # decoder2 is hand-created, and decoder1.sigform is auto-generated
946 # then "yield" fields only from op_fields rather than hard-coded
948 fields
= self
.decoder
.sigforms
[formname
]
949 for name
in op_fields
:
951 sig
= getattr(fields
, name
.upper())
953 sig
= getattr(fields
, name
)
955 # these are all opcode fields involved in index-selection of CR,
956 # and need to do "standard" arithmetic. CR[BA+32] for example
957 # would, if using SelectableInt, only be 5-bit.
958 if name
in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
959 self
.namespace
[name
] = val
961 self
.namespace
[name
] = SelectableInt(val
, sig
.width
)
963 self
.namespace
['XER'] = self
.spr
['XER']
964 self
.namespace
['CA'] = self
.spr
['XER'][XER_bits
['CA']].value
965 self
.namespace
['CA32'] = self
.spr
['XER'][XER_bits
['CA32']].value
967 def handle_carry_(self
, inputs
, outputs
, already_done
):
968 inv_a
= yield self
.dec2
.e
.do
.invert_in
970 inputs
[0] = ~inputs
[0]
972 imm_ok
= yield self
.dec2
.e
.do
.imm_data
.ok
974 imm
= yield self
.dec2
.e
.do
.imm_data
.data
975 inputs
.append(SelectableInt(imm
, 64))
976 assert len(outputs
) >= 1
977 print("outputs", repr(outputs
))
978 if isinstance(outputs
, list) or isinstance(outputs
, tuple):
984 print("gt input", x
, output
)
985 gt
= (gtu(x
, output
))
988 cy
= 1 if any(gts
) else 0
990 if not (1 & already_done
):
991 self
.spr
['XER'][XER_bits
['CA']] = cy
993 print("inputs", already_done
, inputs
)
995 # ARGH... different for OP_ADD... *sigh*...
996 op
= yield self
.dec2
.e
.do
.insn_type
997 if op
== MicrOp
.OP_ADD
.value
:
998 res32
= (output
.value
& (1 << 32)) != 0
999 a32
= (inputs
[0].value
& (1 << 32)) != 0
1000 if len(inputs
) >= 2:
1001 b32
= (inputs
[1].value
& (1 << 32)) != 0
1004 cy32
= res32 ^ a32 ^ b32
1005 print("CA32 ADD", cy32
)
1009 print("input", x
, output
)
1010 print(" x[32:64]", x
, x
[32:64])
1011 print(" o[32:64]", output
, output
[32:64])
1012 gt
= (gtu(x
[32:64], output
[32:64])) == SelectableInt(1, 1)
1014 cy32
= 1 if any(gts
) else 0
1015 print("CA32", cy32
, gts
)
1016 if not (2 & already_done
):
1017 self
.spr
['XER'][XER_bits
['CA32']] = cy32
1019 def handle_overflow(self
, inputs
, outputs
, div_overflow
):
1020 if hasattr(self
.dec2
.e
.do
, "invert_in"):
1021 inv_a
= yield self
.dec2
.e
.do
.invert_in
1023 inputs
[0] = ~inputs
[0]
1025 imm_ok
= yield self
.dec2
.e
.do
.imm_data
.ok
1027 imm
= yield self
.dec2
.e
.do
.imm_data
.data
1028 inputs
.append(SelectableInt(imm
, 64))
1029 assert len(outputs
) >= 1
1030 print("handle_overflow", inputs
, outputs
, div_overflow
)
1031 if len(inputs
) < 2 and div_overflow
is None:
1034 # div overflow is different: it's returned by the pseudo-code
1035 # because it's more complex than can be done by analysing the output
1036 if div_overflow
is not None:
1037 ov
, ov32
= div_overflow
, div_overflow
1038 # arithmetic overflow can be done by analysing the input and output
1039 elif len(inputs
) >= 2:
1043 input_sgn
= [exts(x
.value
, x
.bits
) < 0 for x
in inputs
]
1044 output_sgn
= exts(output
.value
, output
.bits
) < 0
1045 ov
= 1 if input_sgn
[0] == input_sgn
[1] and \
1046 output_sgn
!= input_sgn
[0] else 0
1049 input32_sgn
= [exts(x
.value
, 32) < 0 for x
in inputs
]
1050 output32_sgn
= exts(output
.value
, 32) < 0
1051 ov32
= 1 if input32_sgn
[0] == input32_sgn
[1] and \
1052 output32_sgn
!= input32_sgn
[0] else 0
1054 self
.spr
['XER'][XER_bits
['OV']] = ov
1055 self
.spr
['XER'][XER_bits
['OV32']] = ov32
1056 so
= self
.spr
['XER'][XER_bits
['SO']]
1058 self
.spr
['XER'][XER_bits
['SO']] = so
1060 def handle_comparison(self
, outputs
, cr_idx
=0):
1062 assert isinstance(out
, SelectableInt
), \
1063 "out zero not a SelectableInt %s" % repr(outputs
)
1064 print("handle_comparison", out
.bits
, hex(out
.value
))
1065 # TODO - XXX *processor* in 32-bit mode
1066 # https://bugs.libre-soc.org/show_bug.cgi?id=424
1068 # o32 = exts(out.value, 32)
1069 # print ("handle_comparison exts 32 bit", hex(o32))
1070 out
= exts(out
.value
, out
.bits
)
1071 print("handle_comparison exts", hex(out
))
1072 zero
= SelectableInt(out
== 0, 1)
1073 positive
= SelectableInt(out
> 0, 1)
1074 negative
= SelectableInt(out
< 0, 1)
1075 SO
= self
.spr
['XER'][XER_bits
['SO']]
1076 print("handle_comparison SO", SO
)
1077 cr_field
= selectconcat(negative
, positive
, zero
, SO
)
1078 self
.crl
[cr_idx
].eq(cr_field
)
1080 def set_pc(self
, pc_val
):
1081 self
.namespace
['NIA'] = SelectableInt(pc_val
, 64)
1082 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1084 def setup_one(self
):
1085 """set up one instruction
1088 pc
= self
.pc
.CIA
.value
1092 ins
= self
.imem
.ld(pc
, 4, False, True)
1094 raise KeyError("no instruction at 0x%x" % pc
)
1095 print("setup: 0x%x 0x%x %s" % (pc
, ins
& 0xffffffff, bin(ins
)))
1096 print("CIA NIA", self
.respect_pc
, self
.pc
.CIA
.value
, self
.pc
.NIA
.value
)
1098 yield self
.dec2
.sv_rm
.eq(0)
1099 yield self
.dec2
.dec
.raw_opcode_in
.eq(ins
& 0xffffffff)
1100 yield self
.dec2
.dec
.bigendian
.eq(self
.bigendian
)
1101 yield self
.dec2
.state
.msr
.eq(self
.msr
.value
)
1102 yield self
.dec2
.state
.pc
.eq(pc
)
1103 yield self
.dec2
.state
.svstate
.eq(self
.svstate
.spr
.value
)
1105 # SVP64. first, check if the opcode is EXT001, and SVP64 id bits set
1107 opcode
= yield self
.dec2
.dec
.opcode_in
1108 pfx
= SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
1109 pfx
.insn
.value
= opcode
1110 major
= pfx
.major
.asint(msb0
=True) # MSB0 inversion
1111 print ("prefix test: opcode:", major
, bin(major
),
1112 pfx
.insn
[7] == 0b1, pfx
.insn
[9] == 0b1)
1113 self
.is_svp64_mode
= ((major
== 0b000001) and
1114 pfx
.insn
[7].value
== 0b1 and
1115 pfx
.insn
[9].value
== 0b1)
1116 self
.pc
.update_nia(self
.is_svp64_mode
)
1117 if not self
.is_svp64_mode
:
1120 # in SVP64 mode. decode/print out svp64 prefix, get v3.0B instruction
1121 print ("svp64.rm", bin(pfx
.rm
.asint(msb0
=True)))
1122 print (" svstate.vl", self
.svstate
.vl
.asint(msb0
=True))
1123 print (" svstate.mvl", self
.svstate
.maxvl
.asint(msb0
=True))
1124 sv_rm
= pfx
.rm
.asint(msb0
=True)
1125 ins
= self
.imem
.ld(pc
+4, 4, False, True)
1126 print(" svsetup: 0x%x 0x%x %s" % (pc
+4, ins
& 0xffffffff, bin(ins
)))
1127 yield self
.dec2
.dec
.raw_opcode_in
.eq(ins
& 0xffffffff) # v3.0B suffix
1128 yield self
.dec2
.sv_rm
.eq(sv_rm
) # svp64 prefix
1131 def execute_one(self
):
1132 """execute one instruction
1134 # get the disassembly code for this instruction
1135 if self
.is_svp64_mode
:
1136 code
= self
.disassembly
[self
._pc
+4]
1137 print(" svp64 sim-execute", hex(self
._pc
), code
)
1139 code
= self
.disassembly
[self
._pc
]
1140 print("sim-execute", hex(self
._pc
), code
)
1141 opname
= code
.split(' ')[0]
1142 yield from self
.call(opname
)
1144 # don't use this except in special circumstances
1145 if not self
.respect_pc
:
1148 print("execute one, CIA NIA", self
.pc
.CIA
.value
, self
.pc
.NIA
.value
)
1150 def get_assembly_name(self
):
1151 # TODO, asmregs is from the spec, e.g. add RT,RA,RB
1152 # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
1153 dec_insn
= yield self
.dec2
.e
.do
.insn
1154 asmcode
= yield self
.dec2
.dec
.op
.asmcode
1155 print("get assembly name asmcode", asmcode
, hex(dec_insn
))
1156 asmop
= insns
.get(asmcode
, None)
1157 int_op
= yield self
.dec2
.dec
.op
.internal_op
1159 # sigh reconstruct the assembly instruction name
1160 if hasattr(self
.dec2
.e
.do
, "oe"):
1161 ov_en
= yield self
.dec2
.e
.do
.oe
.oe
1162 ov_ok
= yield self
.dec2
.e
.do
.oe
.ok
1166 if hasattr(self
.dec2
.e
.do
, "rc"):
1167 rc_en
= yield self
.dec2
.e
.do
.rc
.rc
1168 rc_ok
= yield self
.dec2
.e
.do
.rc
.ok
1172 # grrrr have to special-case MUL op (see DecodeOE)
1173 print("ov %d en %d rc %d en %d op %d" %
1174 (ov_ok
, ov_en
, rc_ok
, rc_en
, int_op
))
1175 if int_op
in [MicrOp
.OP_MUL_H64
.value
, MicrOp
.OP_MUL_H32
.value
]:
1180 if not asmop
.endswith("."): # don't add "." to "andis."
1183 if hasattr(self
.dec2
.e
.do
, "lk"):
1184 lk
= yield self
.dec2
.e
.do
.lk
1187 print("int_op", int_op
)
1188 if int_op
in [MicrOp
.OP_B
.value
, MicrOp
.OP_BC
.value
]:
1189 AA
= yield self
.dec2
.dec
.fields
.FormI
.AA
[0:-1]
1193 spr_msb
= yield from self
.get_spr_msb()
1194 if int_op
== MicrOp
.OP_MFCR
.value
:
1199 # XXX TODO: for whatever weird reason this doesn't work
1200 # https://bugs.libre-soc.org/show_bug.cgi?id=390
1201 if int_op
== MicrOp
.OP_MTCRF
.value
:
1208 def get_spr_msb(self
):
1209 dec_insn
= yield self
.dec2
.e
.do
.insn
1210 return dec_insn
& (1 << 20) != 0 # sigh - XFF.spr[-1]?
1212 def call(self
, name
):
1213 """call(opcode) - the primary execution point for instructions
1215 name
= name
.strip() # remove spaces if not already done so
1217 print("halted - not executing", name
)
1220 # TODO, asmregs is from the spec, e.g. add RT,RA,RB
1221 # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
1222 asmop
= yield from self
.get_assembly_name()
1223 print("call", name
, asmop
)
1226 int_op
= yield self
.dec2
.dec
.op
.internal_op
1227 spr_msb
= yield from self
.get_spr_msb()
1229 instr_is_privileged
= False
1230 if int_op
in [MicrOp
.OP_ATTN
.value
,
1231 MicrOp
.OP_MFMSR
.value
,
1232 MicrOp
.OP_MTMSR
.value
,
1233 MicrOp
.OP_MTMSRD
.value
,
1235 MicrOp
.OP_RFID
.value
]:
1236 instr_is_privileged
= True
1237 if int_op
in [MicrOp
.OP_MFSPR
.value
,
1238 MicrOp
.OP_MTSPR
.value
] and spr_msb
:
1239 instr_is_privileged
= True
1241 print("is priv", instr_is_privileged
, hex(self
.msr
.value
),
1243 # check MSR priv bit and whether op is privileged: if so, throw trap
1244 if instr_is_privileged
and self
.msr
[MSRb
.PR
] == 1:
1245 self
.TRAP(0x700, PIb
.PRIV
)
1246 self
.namespace
['NIA'] = self
.trap_nia
1247 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1250 # check halted condition
1255 # check illegal instruction
1257 if name
not in ['mtcrf', 'mtocrf']:
1258 illegal
= name
!= asmop
1261 print("illegal", name
, asmop
)
1262 self
.TRAP(0x700, PIb
.ILLEG
)
1263 self
.namespace
['NIA'] = self
.trap_nia
1264 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1265 print("name %s != %s - calling ILLEGAL trap, PC: %x" %
1266 (name
, asmop
, self
.pc
.CIA
.value
))
1269 info
= self
.instrs
[name
]
1270 yield from self
.prep_namespace(info
.form
, info
.op_fields
)
1272 # preserve order of register names
1273 input_names
= create_args(list(info
.read_regs
) +
1274 list(info
.uninit_regs
))
1277 # get SVP64 entry for the current instruction
1278 sv_rm
= self
.svp64rm
.instrs
.get(name
)
1279 if sv_rm
is not None:
1280 dest_cr
, src_cr
, src_byname
, dest_byname
= decode_extra(sv_rm
)
1282 dest_cr
, src_cr
, src_byname
, dest_byname
= False, False, {}, {}
1283 print ("sv rm", sv_rm
, dest_cr
, src_cr
, src_byname
, dest_byname
)
1285 # get SVSTATE srcstep. TODO: dststep (twin predication)
1286 srcstep
= self
.svstate
.srcstep
.asint(msb0
=True)
1287 vl
= self
.svstate
.vl
.asint(msb0
=True)
1288 mvl
= self
.svstate
.maxvl
.asint(msb0
=True)
1290 # VL=0 in SVP64 mode means "do nothing: skip instruction"
1291 if self
.is_svp64_mode
and vl
== 0:
1292 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1293 print("end of call", self
.namespace
['CIA'], self
.namespace
['NIA'])
1296 # main input registers (RT, RA ...)
1298 for name
in input_names
:
1299 # using PowerDecoder2, first, find the decoder index.
1300 # (mapping name RA RB RC RS to in1, in2, in3)
1301 regnum
, is_vec
= yield from get_pdecode_idx_in(self
.dec2
, name
)
1303 # doing this is not part of svp64, it's because output
1304 # registers, to be modified, need to be in the namespace.
1305 regnum
, is_vec
= yield from get_pdecode_idx_out(self
.dec2
, name
)
1306 # here's where we go "vector". TODO: zero-testing (RA_IS_ZERO)
1307 # XXX already done by PowerDecoder2, now
1309 # regnum += srcstep # TODO, elwidth overrides
1311 # in case getting the register number is needed, _RA, _RB
1312 regname
= "_" + name
1313 self
.namespace
[regname
] = regnum
1314 print('reading reg %s %d' % (name
, regnum
), is_vec
)
1315 reg_val
= self
.gpr(regnum
)
1316 inputs
.append(reg_val
)
1318 # "special" registers
1319 for special
in info
.special_regs
:
1320 if special
in special_sprs
:
1321 inputs
.append(self
.spr
[special
])
1323 inputs
.append(self
.namespace
[special
])
1325 # clear trap (trap) NIA
1326 self
.trap_nia
= None
1328 print("inputs", inputs
)
1329 results
= info
.func(self
, *inputs
)
1330 print("results", results
)
1332 # "inject" decorator takes namespace from function locals: we need to
1333 # overwrite NIA being overwritten (sigh)
1334 if self
.trap_nia
is not None:
1335 self
.namespace
['NIA'] = self
.trap_nia
1337 print("after func", self
.namespace
['CIA'], self
.namespace
['NIA'])
1339 # detect if CA/CA32 already in outputs (sra*, basically)
1342 output_names
= create_args(info
.write_regs
)
1343 for name
in output_names
:
1349 print("carry already done?", bin(already_done
))
1350 if hasattr(self
.dec2
.e
.do
, "output_carry"):
1351 carry_en
= yield self
.dec2
.e
.do
.output_carry
1355 yield from self
.handle_carry_(inputs
, results
, already_done
)
1357 # detect if overflow was in return result
1360 for name
, output
in zip(output_names
, results
):
1361 if name
== 'overflow':
1364 if hasattr(self
.dec2
.e
.do
, "oe"):
1365 ov_en
= yield self
.dec2
.e
.do
.oe
.oe
1366 ov_ok
= yield self
.dec2
.e
.do
.oe
.ok
1370 print("internal overflow", overflow
, ov_en
, ov_ok
)
1372 yield from self
.handle_overflow(inputs
, results
, overflow
)
1374 if hasattr(self
.dec2
.e
.do
, "rc"):
1375 rc_en
= yield self
.dec2
.e
.do
.rc
.rc
1379 regnum
, is_vec
= yield from get_pdecode_cr_out(self
.dec2
, "CR0")
1380 self
.handle_comparison(results
, regnum
)
1382 # any modified return results?
1384 for name
, output
in zip(output_names
, results
):
1385 if name
== 'overflow': # ignore, done already (above)
1387 if isinstance(output
, int):
1388 output
= SelectableInt(output
, 256)
1389 if name
in ['CA', 'CA32']:
1391 print("writing %s to XER" % name
, output
)
1392 self
.spr
['XER'][XER_bits
[name
]] = output
.value
1394 print("NOT writing %s to XER" % name
, output
)
1395 elif name
in info
.special_regs
:
1396 print('writing special %s' % name
, output
, special_sprs
)
1397 if name
in special_sprs
:
1398 self
.spr
[name
] = output
1400 self
.namespace
[name
].eq(output
)
1402 print('msr written', hex(self
.msr
.value
))
1404 regnum
, is_vec
= yield from get_pdecode_idx_out(self
.dec2
,
1407 # temporary hack for not having 2nd output
1408 regnum
= yield getattr(self
.decoder
, name
)
1410 print('writing reg %d %s' % (regnum
, str(output
)), is_vec
)
1411 if output
.bits
> 64:
1412 output
= SelectableInt(output
.value
, 64)
1413 self
.gpr
[regnum
] = output
1415 # check if it is the SVSTATE.src/dest step that needs incrementing
1416 # this is our Sub-Program-Counter loop from 0 to VL-1
1417 if self
.is_svp64_mode
:
1418 # XXX twin predication TODO
1419 vl
= self
.svstate
.vl
.asint(msb0
=True)
1420 mvl
= self
.svstate
.maxvl
.asint(msb0
=True)
1421 srcstep
= self
.svstate
.srcstep
.asint(msb0
=True)
1422 print (" svstate.vl", vl
)
1423 print (" svstate.mvl", mvl
)
1424 print (" svstate.srcstep", srcstep
)
1425 # check if srcstep needs incrementing by one, stop PC advancing
1426 # svp64 loop can end early if the dest is scalar
1427 svp64_dest_vector
= not (yield self
.dec2
.no_out_vec
)
1428 if svp64_dest_vector
and srcstep
!= vl
-1:
1429 self
.svstate
.srcstep
+= SelectableInt(1, 7)
1430 self
.pc
.NIA
.value
= self
.pc
.CIA
.value
1431 self
.namespace
['NIA'] = self
.pc
.NIA
1432 print("end of sub-pc call", self
.namespace
['CIA'],
1433 self
.namespace
['NIA'])
1434 return # DO NOT allow PC to update whilst Sub-PC loop running
1436 self
.svstate
.srcstep
[0:7] = 0
1437 print (" svstate.srcstep loop end (PC to update)")
1438 self
.pc
.update_nia(self
.is_svp64_mode
)
1439 self
.namespace
['NIA'] = self
.pc
.NIA
1441 # UPDATE program counter
1442 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1443 print("end of call", self
.namespace
['CIA'], self
.namespace
['NIA'])
1447 """Decorator factory.
1449 this decorator will "inject" variables into the function's namespace,
1450 from the *dictionary* in self.namespace. it therefore becomes possible
1451 to make it look like a whole stack of variables which would otherwise
1452 need "self." inserted in front of them (*and* for those variables to be
1453 added to the instance) "appear" in the function.
1455 "self.namespace['SI']" for example becomes accessible as just "SI" but
1456 *only* inside the function, when decorated.
1458 def variable_injector(func
):
1460 def decorator(*args
, **kwargs
):
1462 func_globals
= func
.__globals
__ # Python 2.6+
1463 except AttributeError:
1464 func_globals
= func
.func_globals
# Earlier versions.
1466 context
= args
[0].namespace
# variables to be injected
1467 saved_values
= func_globals
.copy() # Shallow copy of dict.
1468 func_globals
.update(context
)
1469 result
= func(*args
, **kwargs
)
1470 print("globals after", func_globals
['CIA'], func_globals
['NIA'])
1471 print("args[0]", args
[0].namespace
['CIA'],
1472 args
[0].namespace
['NIA'])
1473 args
[0].namespace
= func_globals
1474 #exec (func.__code__, func_globals)
1477 # func_globals = saved_values # Undo changes.
1483 return variable_injector
1486 # very quick test of maskgen function (TODO, move to util later)
1487 if __name__
== '__main__':
1488 shift
= SelectableInt(5, 6)
1489 mask
= genmask(shift
, 43)
1490 print (" mask", bin(mask
.value
))
1492 mem
= Mem(row_bytes
=8)
1493 mem
= RADIX(mem
, None)
1494 # -----------------------------------------------
1495 # |/|RTS1|/| RPDB | RTS2 | RPDS |
1496 # -----------------------------------------------
1497 # |0|1 2|3|4 55|56 58|59 63|
1498 data
= SelectableInt(0, 64)
1501 data
[59:64] = 0b01101 # mask
1503 (rts
, mbits
, pgbase
) = mem
._decode
_prte
(data
)
1504 print (" rts", bin(rts
.value
), rts
.bits
)
1505 print (" mbits", bin(mbits
.value
), mbits
.bits
)
1506 print (" pgbase", hex(pgbase
.value
), pgbase
.bits
)
1507 addr
= SelectableInt(0x1000, 64)
1508 check
= mem
._segment
_check
(addr
, mbits
, shift
)
1509 print (" segment check", check
)