0e36a4c257a59f9e18fa69caee4d2a760f8b80f8
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
,
26 from soc
.decoder
.power_enums
import SPR
as DEC_SPR
28 from soc
.decoder
.helpers
import exts
, gtu
, ltu
, undefined
29 from soc
.consts
import PIb
, MSRb
# big-endian (PowerISA versions)
30 from soc
.decoder
.power_svp64
import SVP64RM
, decode_extra
32 from collections
import namedtuple
36 instruction_info
= namedtuple('instruction_info',
37 'func read_regs uninit_regs write_regs ' +
38 'special_regs op_fields form asmregs')
48 def swap_order(x
, nbytes
):
49 x
= x
.to_bytes(nbytes
, byteorder
='little')
50 x
= int.from_bytes(x
, byteorder
='big', signed
=False)
55 # TODO (lkcl): adjust other registers that should be in a particular order
56 # probably CA, CA32, and CR
73 def create_args(reglist
, extra
=None):
74 retval
= list(OrderedSet(reglist
))
75 retval
.sort(key
=lambda reg
: REG_SORT_ORDER
[reg
])
77 return [extra
] + retval
81 # very quick, TODO move to SelectableInt utils later
82 def genmask(shift
, size
):
83 res
= SelectableInt(0, size
)
86 res
[size
-1-i
] = SelectableInt(1, 1)
92 //Accessing 2nd double word of partition table (pate1)
93 //Ref: Power ISA Manual v3.0B, Book-III, section 5.7.6.1
95 // ====================================================
96 // -----------------------------------------------
97 // | /// | PATB | /// | PATS |
98 // -----------------------------------------------
100 // PATB[4:51] holds the base address of the Partition Table,
101 // right shifted by 12 bits.
102 // This is because the address of the Partition base is
103 // 4k aligned. Hence, the lower 12bits, which are always
104 // 0 are ommitted from the PTCR.
106 // Thus, The Partition Table Base is obtained by (PATB << 12)
108 // PATS represents the partition table size right-shifted by 12 bits.
109 // The minimal size of the partition table is 4k.
110 // Thus partition table size = (1 << PATS + 12).
113 // ====================================================
114 // 0 PATE0 63 PATE1 127
115 // |----------------------|----------------------|
117 // |----------------------|----------------------|
119 // |----------------------|----------------------|
121 // |----------------------|----------------------|
125 // |----------------------|----------------------|
127 // |----------------------|----------------------|
129 // The effective LPID forms the index into the Partition Table.
131 // Each entry in the partition table contains 2 double words, PATE0, PATE1,
132 // corresponding to that partition.
134 // In case of Radix, The structure of PATE0 and PATE1 is as follows.
137 // -----------------------------------------------
138 // |1|RTS1|/| RPDB | RTS2 | RPDS |
139 // -----------------------------------------------
140 // 0 1 2 3 4 55 56 58 59 63
142 // HR[0] : For Radix Page table, first bit should be 1.
143 // RTS1[1:2] : Gives one fragment of the Radix treesize
144 // RTS2[56:58] : Gives the second fragment of the Radix Tree size.
145 // RTS = (RTS1 << 3 + RTS2) + 31.
147 // RPDB[4:55] = Root Page Directory Base.
148 // RPDS = Logarithm of Root Page Directory Size right shifted by 3.
149 // Thus, Root page directory size = 1 << (RPDS + 3).
153 // -----------------------------------------------
154 // |///| PRTB | // | PRTS |
155 // -----------------------------------------------
156 // 0 3 4 51 52 58 59 63
158 // PRTB[4:51] = Process Table Base. This is aligned to size.
159 // PRTS[59: 63] = Process Table Size right shifted by 12.
160 // Minimal size of the process table is 4k.
161 // Process Table Size = (1 << PRTS + 12).
164 // Computing the size aligned Process Table Base:
165 // table_base = (PRTB & ~((1 << PRTS) - 1)) << 12
166 // Thus, the lower 12+PRTS bits of table_base will
170 //Ref: Power ISA Manual v3.0B, Book-III, section 5.7.6.2
173 // ==========================
174 // 0 PRTE0 63 PRTE1 127
175 // |----------------------|----------------------|
177 // |----------------------|----------------------|
179 // |----------------------|----------------------|
181 // |----------------------|----------------------|
185 // |----------------------|----------------------|
187 // |----------------------|----------------------|
189 // The effective Process id (PID) forms the index into the Process Table.
191 // Each entry in the partition table contains 2 double words, PRTE0, PRTE1,
192 // corresponding to that process
194 // In case of Radix, The structure of PRTE0 and PRTE1 is as follows.
197 // -----------------------------------------------
198 // |/|RTS1|/| RPDB | RTS2 | RPDS |
199 // -----------------------------------------------
200 // 0 1 2 3 4 55 56 58 59 63
202 // RTS1[1:2] : Gives one fragment of the Radix treesize
203 // RTS2[56:58] : Gives the second fragment of the Radix Tree size.
204 // RTS = (RTS1 << 3 + RTS2) << 31,
205 // since minimal Radix Tree size is 4G.
207 // RPDB = Root Page Directory Base.
208 // RPDS = Root Page Directory Size right shifted by 3.
209 // Thus, Root page directory size = RPDS << 3.
213 // -----------------------------------------------
215 // -----------------------------------------------
217 // All bits are reserved.
222 # see qemu/target/ppc/mmu-radix64.c for reference
224 def __init__(self
, mem
, caller
):
228 # cached page table stuff
230 self
.pt0_valid
= False
232 self
.pt3_valid
= False
234 def __call__(self
,*args
, **kwargs
):
235 print("TODO: implement RADIX.__call__()")
240 def ld(self
, address
, width
=8, swap
=True, check_in_mem
=False):
241 print("RADIX: ld from addr 0x%x width %d" % (address
, width
))
242 dsisr
= self
.caller
.spr
[DEC_SPR
.DSISR
.value
]
243 dar
= self
.caller
.spr
[DEC_SPR
.DAR
.value
]
244 pidr
= self
.caller
.spr
[DEC_SPR
.PIDR
.value
]
245 prtbl
= self
.caller
.spr
[DEC_SPR
.PRTBL
.value
]
247 pte
= self
._walk
_tree
()
248 # use pte to caclculate phys address
249 return self
.mem
.ld(address
, width
, swap
, check_in_mem
)
251 # XXX set SPRs on error
254 def st(self
, addr
, v
, width
=8, swap
=True):
255 print("RADIX: st to addr 0x%x width %d data %x" % (addr
, width
, v
))
256 dsisr
= self
.caller
.spr
[DEC_SPR
.DSISR
.value
]
257 dar
= self
.caller
.spr
[DEC_SPR
.DAR
.value
]
258 pidr
= self
.caller
.spr
[DEC_SPR
.PIDR
.value
]
259 prtbl
= self
.caller
.spr
[DEC_SPR
.PRTBL
.value
]
261 # use pte to caclculate phys address (addr)
262 return self
.mem
.st(addr
, v
, width
, swap
)
264 # XXX set SPRs on error
266 def memassign(self
, addr
, sz
, val
):
267 print("memassign", addr
, sz
, val
)
268 self
.st(addr
.value
, val
.value
, sz
, swap
=False)
270 def _next_level(self
):
275 ## Prepare for next iteration
277 def _walk_tree(self
):
281 // vaddr |-----------------------------------------------------|
283 // |-----------|-----------------------------------------|
284 // | 0000000 | usefulBits = X bits (typically 52) |
285 // |-----------|-----------------------------------------|
286 // | |<--Cursize---->| |
290 // |-----------------------------------------------------|
293 // PDE |---------------------------| |
294 // |V|L|//| NLB |///|NLS| |
295 // |---------------------------| |
296 // PDE = Page Directory Entry |
297 // [0] = V = Valid Bit |
298 // [1] = L = Leaf bit. If 0, then |
299 // [4:55] = NLB = Next Level Base |
300 // right shifted by 8 |
301 // [59:63] = NLS = Next Level Size |
304 // | |--------------------------|
305 // | | usfulBits = X-Cursize |
306 // | |--------------------------|
307 // |---------------------><--NLS-->| |
311 // |--------------------------|
313 // If the next PDE obtained by |
314 // (NLB << 8 + 8 * index) is a |
315 // nonleaf, then repeat the above. |
317 // If the next PDE is a leaf, |
318 // then Leaf PDE structure is as |
323 // |------------------------------| |----------------|
324 // |V|L|sw|//|RPN|sw|R|C|/|ATT|EAA| | usefulBits |
325 // |------------------------------| |----------------|
326 // [0] = V = Valid Bit |
327 // [1] = L = Leaf Bit = 1 if leaf |
329 // [2] = Sw = Sw bit 0. |
330 // [7:51] = RPN = Real Page Number, V
331 // real_page = RPN << 12 -------------> Logical OR
332 // [52:54] = Sw Bits 1:3 |
333 // [55] = R = Reference |
334 // [56] = C = Change V
335 // [58:59] = Att = Physical Address
336 // 0b00 = Normal Memory
338 // 0b10 = Non Idenmpotent
339 // 0b11 = Tolerant I/O
340 // [60:63] = Encoded Access
344 # walk tree starts on prtbl
346 ret
= self
._next
_level
()
349 def _decode_prte(self
, data
):
351 -----------------------------------------------
352 |/|RTS1|/| RPDB | RTS2 | RPDS |
353 -----------------------------------------------
354 0 1 2 3 4 55 56 58 59 63
356 # note that SelectableInt does big-endian! so the indices
357 # below *directly* match the spec, unlike microwatt which
358 # has to turn them around (to LE)
359 zero
= SelectableInt(0, 1)
360 rts
= selectconcat(zero
,
364 masksize
= data
[59:64] # RPDS
365 mbits
= selectconcat(zero
, masksize
)
366 pgbase
= selectconcat(data
[8:56], # part of RPDB
367 SelectableInt(0, 16),)
369 return (rts
, mbits
, pgbase
)
371 def _segment_check(self
, addr
, mbits
, shift
):
372 """checks segment valid
373 mbits := '0' & r.mask_size;
374 v.shift := r.shift + (31 - 12) - mbits;
375 nonzero := or(r.addr(61 downto 31) and not finalmask(30 downto 0));
376 if r.addr(63) /= r.addr(62) or nonzero = '1' then
377 v.state := RADIX_FINISH;
379 elsif mbits < 5 or mbits > 16 or mbits > (r.shift + (31 - 12)) then
380 v.state := RADIX_FINISH;
383 v.state := RADIX_LOOKUP;
385 # note that SelectableInt does big-endian! so the indices
386 # below *directly* match the spec, unlike microwatt which
387 # has to turn them around (to LE)
388 mask
= genmask(shift
, 44)
389 nonzero
= addr
[1:32] & mask
[13:44] # mask 31 LSBs (BE numbered 13:44)
390 print ("RADIX _segment_check nonzero", bin(nonzero
.value
))
391 print ("RADIX _segment_check addr[0-1]", addr
[0].value
, addr
[1].value
)
392 if addr
[0] != addr
[1] or nonzero
== 1:
394 limit
= shift
+ (31 - 12)
395 if mbits
< 5 or mbits
> 16 or mbits
> limit
:
397 new_shift
= shift
+ (31 - 12) - mbits
400 def _check_perms(self
):
401 """check page permissions
403 if data(62) = '1' then
404 -- check permissions and RC bits
406 if r.priv = '1' or data(3) = '0' then
407 if r.iside = '0' then
408 perm_ok := data(1) or (data(2) and not r.store);
410 -- no IAMR, so no KUEP support for now
411 -- deny execute permission if cache inhibited
412 perm_ok := data(0) and not data(5);
415 rc_ok := data(8) and (data(7) or not r.store);
416 if perm_ok = '1' and rc_ok = '1' then
417 v.state := RADIX_LOAD_TLB;
419 v.state := RADIX_FINISH;
420 v.perm_err := not perm_ok;
421 -- permission error takes precedence over RC error
422 v.rc_error := perm_ok;
426 def _get_prtable_addr(self
, shift
, prtbl
, addr
, pid
):
428 if r.addr(63) = '1' then
429 effpid := x"00000000";
433 x"00" & r.prtbl(55 downto 36) &
434 ((r.prtbl(35 downto 12) and not finalmask(23 downto 0)) or
435 (effpid(31 downto 8) and finalmask(23 downto 0))) &
436 effpid(7 downto 0) & "0000";
438 finalmask
= genmask(shift
, 44)
439 finalmask24
= finalmask
[20:44]
440 if addr
[0].value
== 1:
441 effpid
= SelectableInt(0, 32)
443 effpid
= self
.pid
[32:64] # TODO, check on this
444 zero16
= SelectableInt(0, 16)
445 zero4
= SelectableInt(0, 4)
446 rts
= selectconcat(zero16
,
448 (prtbl
[28:52] & ~finalmask24
) |
#
449 (effpid
[0:24] & finalmask24
), #
453 def _get_pgtable_addr(self
):
455 x"00" & r.pgbase(55 downto 19) &
456 ((r.pgbase(18 downto 3) and not mask) or (addrsh and mask)) &
463 ((r.pde(55 downto 12) and not finalmask) or
464 (r.addr(55 downto 12) and finalmask))
465 & r.pde(11 downto 0);
471 def __init__(self
, row_bytes
=8, initial_mem
=None):
473 self
.bytes_per_word
= row_bytes
474 self
.word_log2
= math
.ceil(math
.log2(row_bytes
))
475 print("Sim-Mem", initial_mem
, self
.bytes_per_word
, self
.word_log2
)
479 # different types of memory data structures recognised (for convenience)
480 if isinstance(initial_mem
, list):
481 initial_mem
= (0, initial_mem
)
482 if isinstance(initial_mem
, tuple):
483 startaddr
, mem
= initial_mem
485 for i
, val
in enumerate(mem
):
486 initial_mem
[startaddr
+ row_bytes
*i
] = (val
, row_bytes
)
488 for addr
, (val
, width
) in initial_mem
.items():
489 #val = swap_order(val, width)
490 self
.st(addr
, val
, width
, swap
=False)
492 def _get_shifter_mask(self
, wid
, remainder
):
493 shifter
= ((self
.bytes_per_word
- wid
) - remainder
) * \
495 # XXX https://bugs.libre-soc.org/show_bug.cgi?id=377
497 shifter
= remainder
* 8
498 mask
= (1 << (wid
* 8)) - 1
499 print("width,rem,shift,mask", wid
, remainder
, hex(shifter
), hex(mask
))
502 # TODO: Implement ld/st of lesser width
503 def ld(self
, address
, width
=8, swap
=True, check_in_mem
=False):
504 print("ld from addr 0x{:x} width {:d}".format(address
, width
))
505 remainder
= address
& (self
.bytes_per_word
- 1)
506 address
= address
>> self
.word_log2
507 assert remainder
& (width
- 1) == 0, "Unaligned access unsupported!"
508 if address
in self
.mem
:
509 val
= self
.mem
[address
]
514 print("mem @ 0x{:x} rem {:d} : 0x{:x}".format(address
, remainder
, val
))
516 if width
!= self
.bytes_per_word
:
517 shifter
, mask
= self
._get
_shifter
_mask
(width
, remainder
)
518 print("masking", hex(val
), hex(mask
<< shifter
), shifter
)
519 val
= val
& (mask
<< shifter
)
522 val
= swap_order(val
, width
)
523 print("Read 0x{:x} from addr 0x{:x}".format(val
, address
))
526 def st(self
, addr
, v
, width
=8, swap
=True):
528 remainder
= addr
& (self
.bytes_per_word
- 1)
529 addr
= addr
>> self
.word_log2
530 print("Writing 0x{:x} to ST 0x{:x} "
531 "memaddr 0x{:x}/{:x}".format(v
, staddr
, addr
, remainder
, swap
))
532 assert remainder
& (width
- 1) == 0, "Unaligned access unsupported!"
534 v
= swap_order(v
, width
)
535 if width
!= self
.bytes_per_word
:
540 shifter
, mask
= self
._get
_shifter
_mask
(width
, remainder
)
541 val
&= ~
(mask
<< shifter
)
546 print("mem @ 0x{:x}: 0x{:x}".format(addr
, self
.mem
[addr
]))
548 def __call__(self
, addr
, sz
):
549 val
= self
.ld(addr
.value
, sz
, swap
=False)
550 print("memread", addr
, sz
, val
)
551 return SelectableInt(val
, sz
*8)
553 def memassign(self
, addr
, sz
, val
):
554 print("memassign", addr
, sz
, val
)
555 self
.st(addr
.value
, val
.value
, sz
, swap
=False)
559 def __init__(self
, decoder
, isacaller
, svstate
, regfile
):
562 self
.isacaller
= isacaller
563 self
.svstate
= svstate
565 self
[i
] = SelectableInt(regfile
[i
], 64)
567 def __call__(self
, ridx
):
570 def set_form(self
, form
):
573 def getz(self
, rnum
):
574 # rnum = rnum.value # only SelectableInt allowed
575 print("GPR getzero", rnum
)
577 return SelectableInt(0, 64)
580 def _get_regnum(self
, attr
):
581 getform
= self
.sd
.sigforms
[self
.form
]
582 rnum
= getattr(getform
, attr
)
585 def ___getitem__(self
, attr
):
586 """ XXX currently not used
588 rnum
= self
._get
_regnum
(attr
)
589 offs
= self
.svstate
.srcstep
590 print("GPR getitem", attr
, rnum
, "srcoffs", offs
)
591 return self
.regfile
[rnum
]
594 for i
in range(0, len(self
), 8):
597 s
.append("%08x" % self
[i
+j
].value
)
599 print("reg", "%2d" % i
, s
)
603 def __init__(self
, pc_init
=0):
604 self
.CIA
= SelectableInt(pc_init
, 64)
605 self
.NIA
= self
.CIA
+ SelectableInt(4, 64) # only true for v3.0B!
607 def update_nia(self
, is_svp64
):
608 increment
= 8 if is_svp64
else 4
609 self
.NIA
= self
.CIA
+ SelectableInt(increment
, 64)
611 def update(self
, namespace
, is_svp64
):
612 """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
614 self
.CIA
= namespace
['NIA'].narrow(64)
615 self
.update_nia(is_svp64
)
616 namespace
['CIA'] = self
.CIA
617 namespace
['NIA'] = self
.NIA
620 # Simple-V: see https://libre-soc.org/openpower/sv
622 def __init__(self
, init
=0):
623 self
.spr
= SelectableInt(init
, 32)
624 # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
625 self
.maxvl
= FieldSelectableInt(self
.spr
, tuple(range(0,7)))
626 self
.vl
= FieldSelectableInt(self
.spr
, tuple(range(7,14)))
627 self
.srcstep
= FieldSelectableInt(self
.spr
, tuple(range(14,21)))
628 self
.dststep
= FieldSelectableInt(self
.spr
, tuple(range(21,28)))
629 self
.subvl
= FieldSelectableInt(self
.spr
, tuple(range(28,30)))
630 self
.svstep
= FieldSelectableInt(self
.spr
, tuple(range(30,32)))
635 def __init__(self
, init
=0):
636 self
.spr
= SelectableInt(init
, 24)
637 # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
638 self
.mmode
= FieldSelectableInt(self
.spr
, [0])
639 self
.mask
= FieldSelectableInt(self
.spr
, tuple(range(1,4)))
640 self
.elwidth
= FieldSelectableInt(self
.spr
, tuple(range(4,6)))
641 self
.ewsrc
= FieldSelectableInt(self
.spr
, tuple(range(6,8)))
642 self
.subvl
= FieldSelectableInt(self
.spr
, tuple(range(8,10)))
643 self
.extra
= FieldSelectableInt(self
.spr
, tuple(range(10,19)))
644 self
.mode
= FieldSelectableInt(self
.spr
, tuple(range(19,24)))
645 # these cover the same extra field, split into parts as EXTRA2
646 self
.extra2
= list(range(4))
647 self
.extra2
[0] = FieldSelectableInt(self
.spr
, tuple(range(10,12)))
648 self
.extra2
[1] = FieldSelectableInt(self
.spr
, tuple(range(12,14)))
649 self
.extra2
[2] = FieldSelectableInt(self
.spr
, tuple(range(14,16)))
650 self
.extra2
[3] = FieldSelectableInt(self
.spr
, tuple(range(16,18)))
651 self
.smask
= FieldSelectableInt(self
.spr
, tuple(range(16,19)))
652 # and here as well, but EXTRA3
653 self
.extra3
= list(range(3))
654 self
.extra3
[0] = FieldSelectableInt(self
.spr
, tuple(range(10,13)))
655 self
.extra3
[1] = FieldSelectableInt(self
.spr
, tuple(range(13,16)))
656 self
.extra3
[2] = FieldSelectableInt(self
.spr
, tuple(range(16,19)))
659 SVP64RM_MMODE_SIZE
= len(SVP64RMFields().mmode
.br
)
660 SVP64RM_MASK_SIZE
= len(SVP64RMFields().mask
.br
)
661 SVP64RM_ELWIDTH_SIZE
= len(SVP64RMFields().elwidth
.br
)
662 SVP64RM_EWSRC_SIZE
= len(SVP64RMFields().ewsrc
.br
)
663 SVP64RM_SUBVL_SIZE
= len(SVP64RMFields().subvl
.br
)
664 SVP64RM_EXTRA2_SPEC_SIZE
= len(SVP64RMFields().extra2
[0].br
)
665 SVP64RM_EXTRA3_SPEC_SIZE
= len(SVP64RMFields().extra3
[0].br
)
666 SVP64RM_SMASK_SIZE
= len(SVP64RMFields().smask
.br
)
667 SVP64RM_MODE_SIZE
= len(SVP64RMFields().mode
.br
)
670 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
671 class SVP64PrefixFields
:
673 self
.insn
= SelectableInt(0, 32)
674 # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
675 self
.major
= FieldSelectableInt(self
.insn
, tuple(range(0,6)))
676 self
.pid
= FieldSelectableInt(self
.insn
, (7, 9)) # must be 0b11
677 rmfields
= [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
678 self
.rm
= FieldSelectableInt(self
.insn
, rmfields
)
681 SV64P_MAJOR_SIZE
= len(SVP64PrefixFields().major
.br
)
682 SV64P_PID_SIZE
= len(SVP64PrefixFields().pid
.br
)
683 SV64P_RM_SIZE
= len(SVP64PrefixFields().rm
.br
)
687 def __init__(self
, dec2
, initial_sprs
={}):
690 for key
, v
in initial_sprs
.items():
691 if isinstance(key
, SelectableInt
):
693 key
= special_sprs
.get(key
, key
)
694 if isinstance(key
, int):
697 info
= spr_byname
[key
]
698 if not isinstance(v
, SelectableInt
):
699 v
= SelectableInt(v
, info
.length
)
702 def __getitem__(self
, key
):
703 print("get spr", key
)
704 print("dict", self
.items())
705 # if key in special_sprs get the special spr, otherwise return key
706 if isinstance(key
, SelectableInt
):
708 if isinstance(key
, int):
709 key
= spr_dict
[key
].SPR
710 key
= special_sprs
.get(key
, key
)
711 if key
== 'HSRR0': # HACK!
713 if key
== 'HSRR1': # HACK!
716 res
= dict.__getitem
__(self
, key
)
718 if isinstance(key
, int):
721 info
= spr_byname
[key
]
722 dict.__setitem
__(self
, key
, SelectableInt(0, info
.length
))
723 res
= dict.__getitem
__(self
, key
)
724 print("spr returning", key
, res
)
727 def __setitem__(self
, key
, value
):
728 if isinstance(key
, SelectableInt
):
730 if isinstance(key
, int):
731 key
= spr_dict
[key
].SPR
732 print("spr key", key
)
733 key
= special_sprs
.get(key
, key
)
734 if key
== 'HSRR0': # HACK!
735 self
.__setitem
__('SRR0', value
)
736 if key
== 'HSRR1': # HACK!
737 self
.__setitem
__('SRR1', value
)
738 print("setting spr", key
, value
)
739 dict.__setitem
__(self
, key
, value
)
741 def __call__(self
, ridx
):
744 def get_pdecode_idx_in(dec2
, name
):
746 in1_sel
= yield op
.in1_sel
747 in2_sel
= yield op
.in2_sel
748 in3_sel
= yield op
.in3_sel
749 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
750 in1
= yield dec2
.e
.read_reg1
.data
751 in2
= yield dec2
.e
.read_reg2
.data
752 in3
= yield dec2
.e
.read_reg3
.data
753 in1_isvec
= yield dec2
.in1_isvec
754 in2_isvec
= yield dec2
.in2_isvec
755 in3_isvec
= yield dec2
.in3_isvec
756 print ("get_pdecode_idx", in1_sel
, In1Sel
.RA
.value
, in1
, in1_isvec
)
757 # identify which regnames map to in1/2/3
759 if (in1_sel
== In1Sel
.RA
.value
or
760 (in1_sel
== In1Sel
.RA_OR_ZERO
.value
and in1
!= 0)):
761 return in1
, in1_isvec
762 if in1_sel
== In1Sel
.RA_OR_ZERO
.value
:
763 return in1
, in1_isvec
765 if in2_sel
== In2Sel
.RB
.value
:
766 return in2
, in2_isvec
767 if in3_sel
== In3Sel
.RB
.value
:
768 return in3
, in3_isvec
769 # XXX TODO, RC doesn't exist yet!
771 assert False, "RC does not exist yet"
773 if in1_sel
== In1Sel
.RS
.value
:
774 return in1
, in1_isvec
775 if in2_sel
== In2Sel
.RS
.value
:
776 return in2
, in2_isvec
777 if in3_sel
== In3Sel
.RS
.value
:
778 return in3
, in3_isvec
782 def get_pdecode_cr_out(dec2
, name
):
784 out_sel
= yield op
.cr_out
785 out_bitfield
= yield dec2
.dec_cr_out
.cr_bitfield
.data
786 sv_cr_out
= yield op
.sv_cr_out
787 spec
= yield dec2
.crout_svdec
.spec
788 sv_override
= yield dec2
.dec_cr_out
.sv_override
789 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
790 out
= yield dec2
.e
.write_cr
.data
791 o_isvec
= yield dec2
.o_isvec
792 print ("get_pdecode_cr_out", out_sel
, CROutSel
.CR0
.value
, out
, o_isvec
)
793 print (" sv_cr_out", sv_cr_out
)
794 print (" cr_bf", out_bitfield
)
795 print (" spec", spec
)
796 print (" override", sv_override
)
797 # identify which regnames map to out / o2
799 if out_sel
== CROutSel
.CR0
.value
:
801 print ("get_pdecode_idx_out not found", name
)
805 def get_pdecode_idx_out(dec2
, name
):
807 out_sel
= yield op
.out_sel
808 # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
809 out
= yield dec2
.e
.write_reg
.data
810 o_isvec
= yield dec2
.o_isvec
811 print ("get_pdecode_idx_out", out_sel
, OutSel
.RA
.value
, out
, o_isvec
)
812 # identify which regnames map to out / o2
814 if out_sel
== OutSel
.RA
.value
:
817 if out_sel
== OutSel
.RT
.value
:
819 print ("get_pdecode_idx_out not found", name
)
824 def get_pdecode_idx_out2(dec2
, name
):
826 print ("TODO: get_pdecode_idx_out2", name
)
831 # decoder2 - an instance of power_decoder2
832 # regfile - a list of initial values for the registers
833 # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
834 # respect_pc - tracks the program counter. requires initial_insns
835 def __init__(self
, decoder2
, regfile
, initial_sprs
=None, initial_cr
=0,
836 initial_mem
=None, initial_msr
=0,
838 initial_insns
=None, respect_pc
=False,
844 self
.bigendian
= bigendian
846 self
.is_svp64_mode
= False
847 self
.respect_pc
= respect_pc
848 if initial_sprs
is None:
850 if initial_mem
is None:
852 if initial_insns
is None:
854 assert self
.respect_pc
== False, "instructions required to honor pc"
856 print("ISACaller insns", respect_pc
, initial_insns
, disassembly
)
857 print("ISACaller initial_msr", initial_msr
)
859 # "fake program counter" mode (for unit testing)
863 if isinstance(initial_mem
, tuple):
864 self
.fake_pc
= initial_mem
[0]
865 disasm_start
= self
.fake_pc
867 disasm_start
= initial_pc
869 # disassembly: we need this for now (not given from the decoder)
870 self
.disassembly
= {}
872 for i
, code
in enumerate(disassembly
):
873 self
.disassembly
[i
*4 + disasm_start
] = code
875 # set up registers, instruction memory, data memory, PC, SPRs, MSR
876 self
.svp64rm
= SVP64RM()
877 if isinstance(initial_svstate
, int):
878 initial_svstate
= SVP64State(initial_svstate
)
879 self
.svstate
= initial_svstate
880 self
.gpr
= GPR(decoder2
, self
, self
.svstate
, regfile
)
881 self
.mem
= Mem(row_bytes
=8, initial_mem
=initial_mem
)
883 self
.mem
= RADIX(self
.mem
, self
)
884 self
.imem
= Mem(row_bytes
=4, initial_mem
=initial_insns
)
886 self
.spr
= SPR(decoder2
, initial_sprs
)
887 self
.msr
= SelectableInt(initial_msr
, 64) # underlying reg
890 # FPR (same as GPR except for FP nums)
891 # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
892 # note that mffs, mcrfs, mtfsf "manage" this FPSCR
893 # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
894 # note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
896 # 2.3.2 LR (actually SPR #8) -- Done
897 # 2.3.3 CTR (actually SPR #9) -- Done
898 # 2.3.4 TAR (actually SPR #815)
899 # 3.2.2 p45 XER (actually SPR #1) -- Done
900 # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
902 # create CR then allow portions of it to be "selectable" (below)
903 #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
904 self
.cr
= SelectableInt(initial_cr
, 64) # underlying reg
905 #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
907 # "undefined", just set to variable-bit-width int (use exts "max")
908 #self.undefined = SelectableInt(0, 256) # TODO, not hard-code 256!
911 self
.namespace
.update(self
.spr
)
912 self
.namespace
.update({'GPR': self
.gpr
,
915 'memassign': self
.memassign
,
920 'undefined': undefined
,
921 'mode_is_64bit': True,
925 # update pc to requested start point
926 self
.set_pc(initial_pc
)
928 # field-selectable versions of Condition Register TODO check bitranges?
931 bits
= tuple(range(i
*4+32, (i
+1)*4+32)) # errr... maybe?
932 _cr
= FieldSelectableInt(self
.cr
, bits
)
934 self
.namespace
["CR%d" % i
] = _cr
936 self
.decoder
= decoder2
.dec
939 def TRAP(self
, trap_addr
=0x700, trap_bit
=PIb
.TRAP
):
940 print("TRAP:", hex(trap_addr
), hex(self
.namespace
['MSR'].value
))
941 # store CIA(+4?) in SRR0, set NIA to 0x700
942 # store MSR in SRR1, set MSR to um errr something, have to check spec
943 self
.spr
['SRR0'].value
= self
.pc
.CIA
.value
944 self
.spr
['SRR1'].value
= self
.namespace
['MSR'].value
945 self
.trap_nia
= SelectableInt(trap_addr
, 64)
946 self
.spr
['SRR1'][trap_bit
] = 1 # change *copy* of MSR in SRR1
948 # set exception bits. TODO: this should, based on the address
949 # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
950 # bits appropriately. however it turns out that *for now* in all
951 # cases (all trap_addrs) the exact same thing is needed.
952 self
.msr
[MSRb
.IR
] = 0
953 self
.msr
[MSRb
.DR
] = 0
954 self
.msr
[MSRb
.FE0
] = 0
955 self
.msr
[MSRb
.FE1
] = 0
956 self
.msr
[MSRb
.EE
] = 0
957 self
.msr
[MSRb
.RI
] = 0
958 self
.msr
[MSRb
.SF
] = 1
959 self
.msr
[MSRb
.TM
] = 0
960 self
.msr
[MSRb
.VEC
] = 0
961 self
.msr
[MSRb
.VSX
] = 0
962 self
.msr
[MSRb
.PR
] = 0
963 self
.msr
[MSRb
.FP
] = 0
964 self
.msr
[MSRb
.PMM
] = 0
965 self
.msr
[MSRb
.TEs
] = 0
966 self
.msr
[MSRb
.TEe
] = 0
967 self
.msr
[MSRb
.UND
] = 0
968 self
.msr
[MSRb
.LE
] = 1
970 def memassign(self
, ea
, sz
, val
):
971 self
.mem
.memassign(ea
, sz
, val
)
973 def prep_namespace(self
, formname
, op_fields
):
974 # TODO: get field names from form in decoder*1* (not decoder2)
975 # decoder2 is hand-created, and decoder1.sigform is auto-generated
977 # then "yield" fields only from op_fields rather than hard-coded
979 fields
= self
.decoder
.sigforms
[formname
]
980 for name
in op_fields
:
982 sig
= getattr(fields
, name
.upper())
984 sig
= getattr(fields
, name
)
986 # these are all opcode fields involved in index-selection of CR,
987 # and need to do "standard" arithmetic. CR[BA+32] for example
988 # would, if using SelectableInt, only be 5-bit.
989 if name
in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
990 self
.namespace
[name
] = val
992 self
.namespace
[name
] = SelectableInt(val
, sig
.width
)
994 self
.namespace
['XER'] = self
.spr
['XER']
995 self
.namespace
['CA'] = self
.spr
['XER'][XER_bits
['CA']].value
996 self
.namespace
['CA32'] = self
.spr
['XER'][XER_bits
['CA32']].value
998 def handle_carry_(self
, inputs
, outputs
, already_done
):
999 inv_a
= yield self
.dec2
.e
.do
.invert_in
1001 inputs
[0] = ~inputs
[0]
1003 imm_ok
= yield self
.dec2
.e
.do
.imm_data
.ok
1005 imm
= yield self
.dec2
.e
.do
.imm_data
.data
1006 inputs
.append(SelectableInt(imm
, 64))
1007 assert len(outputs
) >= 1
1008 print("outputs", repr(outputs
))
1009 if isinstance(outputs
, list) or isinstance(outputs
, tuple):
1015 print("gt input", x
, output
)
1016 gt
= (gtu(x
, output
))
1019 cy
= 1 if any(gts
) else 0
1020 print("CA", cy
, gts
)
1021 if not (1 & already_done
):
1022 self
.spr
['XER'][XER_bits
['CA']] = cy
1024 print("inputs", already_done
, inputs
)
1026 # ARGH... different for OP_ADD... *sigh*...
1027 op
= yield self
.dec2
.e
.do
.insn_type
1028 if op
== MicrOp
.OP_ADD
.value
:
1029 res32
= (output
.value
& (1 << 32)) != 0
1030 a32
= (inputs
[0].value
& (1 << 32)) != 0
1031 if len(inputs
) >= 2:
1032 b32
= (inputs
[1].value
& (1 << 32)) != 0
1035 cy32
= res32 ^ a32 ^ b32
1036 print("CA32 ADD", cy32
)
1040 print("input", x
, output
)
1041 print(" x[32:64]", x
, x
[32:64])
1042 print(" o[32:64]", output
, output
[32:64])
1043 gt
= (gtu(x
[32:64], output
[32:64])) == SelectableInt(1, 1)
1045 cy32
= 1 if any(gts
) else 0
1046 print("CA32", cy32
, gts
)
1047 if not (2 & already_done
):
1048 self
.spr
['XER'][XER_bits
['CA32']] = cy32
1050 def handle_overflow(self
, inputs
, outputs
, div_overflow
):
1051 if hasattr(self
.dec2
.e
.do
, "invert_in"):
1052 inv_a
= yield self
.dec2
.e
.do
.invert_in
1054 inputs
[0] = ~inputs
[0]
1056 imm_ok
= yield self
.dec2
.e
.do
.imm_data
.ok
1058 imm
= yield self
.dec2
.e
.do
.imm_data
.data
1059 inputs
.append(SelectableInt(imm
, 64))
1060 assert len(outputs
) >= 1
1061 print("handle_overflow", inputs
, outputs
, div_overflow
)
1062 if len(inputs
) < 2 and div_overflow
is None:
1065 # div overflow is different: it's returned by the pseudo-code
1066 # because it's more complex than can be done by analysing the output
1067 if div_overflow
is not None:
1068 ov
, ov32
= div_overflow
, div_overflow
1069 # arithmetic overflow can be done by analysing the input and output
1070 elif len(inputs
) >= 2:
1074 input_sgn
= [exts(x
.value
, x
.bits
) < 0 for x
in inputs
]
1075 output_sgn
= exts(output
.value
, output
.bits
) < 0
1076 ov
= 1 if input_sgn
[0] == input_sgn
[1] and \
1077 output_sgn
!= input_sgn
[0] else 0
1080 input32_sgn
= [exts(x
.value
, 32) < 0 for x
in inputs
]
1081 output32_sgn
= exts(output
.value
, 32) < 0
1082 ov32
= 1 if input32_sgn
[0] == input32_sgn
[1] and \
1083 output32_sgn
!= input32_sgn
[0] else 0
1085 self
.spr
['XER'][XER_bits
['OV']] = ov
1086 self
.spr
['XER'][XER_bits
['OV32']] = ov32
1087 so
= self
.spr
['XER'][XER_bits
['SO']]
1089 self
.spr
['XER'][XER_bits
['SO']] = so
1091 def handle_comparison(self
, outputs
, cr_idx
=0):
1093 assert isinstance(out
, SelectableInt
), \
1094 "out zero not a SelectableInt %s" % repr(outputs
)
1095 print("handle_comparison", out
.bits
, hex(out
.value
))
1096 # TODO - XXX *processor* in 32-bit mode
1097 # https://bugs.libre-soc.org/show_bug.cgi?id=424
1099 # o32 = exts(out.value, 32)
1100 # print ("handle_comparison exts 32 bit", hex(o32))
1101 out
= exts(out
.value
, out
.bits
)
1102 print("handle_comparison exts", hex(out
))
1103 zero
= SelectableInt(out
== 0, 1)
1104 positive
= SelectableInt(out
> 0, 1)
1105 negative
= SelectableInt(out
< 0, 1)
1106 SO
= self
.spr
['XER'][XER_bits
['SO']]
1107 print("handle_comparison SO", SO
)
1108 cr_field
= selectconcat(negative
, positive
, zero
, SO
)
1109 self
.crl
[cr_idx
].eq(cr_field
)
1111 def set_pc(self
, pc_val
):
1112 self
.namespace
['NIA'] = SelectableInt(pc_val
, 64)
1113 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1115 def setup_one(self
):
1116 """set up one instruction
1119 pc
= self
.pc
.CIA
.value
1123 ins
= self
.imem
.ld(pc
, 4, False, True)
1125 raise KeyError("no instruction at 0x%x" % pc
)
1126 print("setup: 0x%x 0x%x %s" % (pc
, ins
& 0xffffffff, bin(ins
)))
1127 print("CIA NIA", self
.respect_pc
, self
.pc
.CIA
.value
, self
.pc
.NIA
.value
)
1129 yield self
.dec2
.sv_rm
.eq(0)
1130 yield self
.dec2
.dec
.raw_opcode_in
.eq(ins
& 0xffffffff)
1131 yield self
.dec2
.dec
.bigendian
.eq(self
.bigendian
)
1132 yield self
.dec2
.state
.msr
.eq(self
.msr
.value
)
1133 yield self
.dec2
.state
.pc
.eq(pc
)
1134 yield self
.dec2
.state
.svstate
.eq(self
.svstate
.spr
.value
)
1136 # SVP64. first, check if the opcode is EXT001, and SVP64 id bits set
1138 opcode
= yield self
.dec2
.dec
.opcode_in
1139 pfx
= SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
1140 pfx
.insn
.value
= opcode
1141 major
= pfx
.major
.asint(msb0
=True) # MSB0 inversion
1142 print ("prefix test: opcode:", major
, bin(major
),
1143 pfx
.insn
[7] == 0b1, pfx
.insn
[9] == 0b1)
1144 self
.is_svp64_mode
= ((major
== 0b000001) and
1145 pfx
.insn
[7].value
== 0b1 and
1146 pfx
.insn
[9].value
== 0b1)
1147 self
.pc
.update_nia(self
.is_svp64_mode
)
1148 self
.namespace
['NIA'] = self
.pc
.NIA
1149 if not self
.is_svp64_mode
:
1152 # in SVP64 mode. decode/print out svp64 prefix, get v3.0B instruction
1153 print ("svp64.rm", bin(pfx
.rm
.asint(msb0
=True)))
1154 print (" svstate.vl", self
.svstate
.vl
.asint(msb0
=True))
1155 print (" svstate.mvl", self
.svstate
.maxvl
.asint(msb0
=True))
1156 sv_rm
= pfx
.rm
.asint(msb0
=True)
1157 ins
= self
.imem
.ld(pc
+4, 4, False, True)
1158 print(" svsetup: 0x%x 0x%x %s" % (pc
+4, ins
& 0xffffffff, bin(ins
)))
1159 yield self
.dec2
.dec
.raw_opcode_in
.eq(ins
& 0xffffffff) # v3.0B suffix
1160 yield self
.dec2
.sv_rm
.eq(sv_rm
) # svp64 prefix
1163 def execute_one(self
):
1164 """execute one instruction
1166 # get the disassembly code for this instruction
1167 if self
.is_svp64_mode
:
1168 code
= self
.disassembly
[self
._pc
+4]
1169 print(" svp64 sim-execute", hex(self
._pc
), code
)
1171 code
= self
.disassembly
[self
._pc
]
1172 print("sim-execute", hex(self
._pc
), code
)
1173 opname
= code
.split(' ')[0]
1174 yield from self
.call(opname
)
1176 # don't use this except in special circumstances
1177 if not self
.respect_pc
:
1180 print("execute one, CIA NIA", self
.pc
.CIA
.value
, self
.pc
.NIA
.value
)
1182 def get_assembly_name(self
):
1183 # TODO, asmregs is from the spec, e.g. add RT,RA,RB
1184 # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
1185 dec_insn
= yield self
.dec2
.e
.do
.insn
1186 asmcode
= yield self
.dec2
.dec
.op
.asmcode
1187 print("get assembly name asmcode", asmcode
, hex(dec_insn
))
1188 asmop
= insns
.get(asmcode
, None)
1189 int_op
= yield self
.dec2
.dec
.op
.internal_op
1191 # sigh reconstruct the assembly instruction name
1192 if hasattr(self
.dec2
.e
.do
, "oe"):
1193 ov_en
= yield self
.dec2
.e
.do
.oe
.oe
1194 ov_ok
= yield self
.dec2
.e
.do
.oe
.ok
1198 if hasattr(self
.dec2
.e
.do
, "rc"):
1199 rc_en
= yield self
.dec2
.e
.do
.rc
.rc
1200 rc_ok
= yield self
.dec2
.e
.do
.rc
.ok
1204 # grrrr have to special-case MUL op (see DecodeOE)
1205 print("ov %d en %d rc %d en %d op %d" %
1206 (ov_ok
, ov_en
, rc_ok
, rc_en
, int_op
))
1207 if int_op
in [MicrOp
.OP_MUL_H64
.value
, MicrOp
.OP_MUL_H32
.value
]:
1212 if not asmop
.endswith("."): # don't add "." to "andis."
1215 if hasattr(self
.dec2
.e
.do
, "lk"):
1216 lk
= yield self
.dec2
.e
.do
.lk
1219 print("int_op", int_op
)
1220 if int_op
in [MicrOp
.OP_B
.value
, MicrOp
.OP_BC
.value
]:
1221 AA
= yield self
.dec2
.dec
.fields
.FormI
.AA
[0:-1]
1225 spr_msb
= yield from self
.get_spr_msb()
1226 if int_op
== MicrOp
.OP_MFCR
.value
:
1231 # XXX TODO: for whatever weird reason this doesn't work
1232 # https://bugs.libre-soc.org/show_bug.cgi?id=390
1233 if int_op
== MicrOp
.OP_MTCRF
.value
:
1240 def get_spr_msb(self
):
1241 dec_insn
= yield self
.dec2
.e
.do
.insn
1242 return dec_insn
& (1 << 20) != 0 # sigh - XFF.spr[-1]?
1244 def call(self
, name
):
1245 """call(opcode) - the primary execution point for instructions
1247 name
= name
.strip() # remove spaces if not already done so
1249 print("halted - not executing", name
)
1252 # TODO, asmregs is from the spec, e.g. add RT,RA,RB
1253 # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
1254 asmop
= yield from self
.get_assembly_name()
1255 print("call", name
, asmop
)
1258 int_op
= yield self
.dec2
.dec
.op
.internal_op
1259 spr_msb
= yield from self
.get_spr_msb()
1261 instr_is_privileged
= False
1262 if int_op
in [MicrOp
.OP_ATTN
.value
,
1263 MicrOp
.OP_MFMSR
.value
,
1264 MicrOp
.OP_MTMSR
.value
,
1265 MicrOp
.OP_MTMSRD
.value
,
1267 MicrOp
.OP_RFID
.value
]:
1268 instr_is_privileged
= True
1269 if int_op
in [MicrOp
.OP_MFSPR
.value
,
1270 MicrOp
.OP_MTSPR
.value
] and spr_msb
:
1271 instr_is_privileged
= True
1273 print("is priv", instr_is_privileged
, hex(self
.msr
.value
),
1275 # check MSR priv bit and whether op is privileged: if so, throw trap
1276 if instr_is_privileged
and self
.msr
[MSRb
.PR
] == 1:
1277 self
.TRAP(0x700, PIb
.PRIV
)
1278 self
.namespace
['NIA'] = self
.trap_nia
1279 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1282 # check halted condition
1287 # check illegal instruction
1289 if name
not in ['mtcrf', 'mtocrf']:
1290 illegal
= name
!= asmop
1293 print("illegal", name
, asmop
)
1294 self
.TRAP(0x700, PIb
.ILLEG
)
1295 self
.namespace
['NIA'] = self
.trap_nia
1296 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1297 print("name %s != %s - calling ILLEGAL trap, PC: %x" %
1298 (name
, asmop
, self
.pc
.CIA
.value
))
1301 info
= self
.instrs
[name
]
1302 yield from self
.prep_namespace(info
.form
, info
.op_fields
)
1304 # preserve order of register names
1305 input_names
= create_args(list(info
.read_regs
) +
1306 list(info
.uninit_regs
))
1309 # get SVP64 entry for the current instruction
1310 sv_rm
= self
.svp64rm
.instrs
.get(name
)
1311 if sv_rm
is not None:
1312 dest_cr
, src_cr
, src_byname
, dest_byname
= decode_extra(sv_rm
)
1314 dest_cr
, src_cr
, src_byname
, dest_byname
= False, False, {}, {}
1315 print ("sv rm", sv_rm
, dest_cr
, src_cr
, src_byname
, dest_byname
)
1317 # get SVSTATE srcstep. TODO: dststep (twin predication)
1318 srcstep
= self
.svstate
.srcstep
.asint(msb0
=True)
1319 vl
= self
.svstate
.vl
.asint(msb0
=True)
1320 mvl
= self
.svstate
.maxvl
.asint(msb0
=True)
1322 # VL=0 in SVP64 mode means "do nothing: skip instruction"
1323 if self
.is_svp64_mode
and vl
== 0:
1324 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1325 print("end of call", self
.namespace
['CIA'], self
.namespace
['NIA'])
1328 # main input registers (RT, RA ...)
1330 for name
in input_names
:
1331 # using PowerDecoder2, first, find the decoder index.
1332 # (mapping name RA RB RC RS to in1, in2, in3)
1333 regnum
, is_vec
= yield from get_pdecode_idx_in(self
.dec2
, name
)
1335 # doing this is not part of svp64, it's because output
1336 # registers, to be modified, need to be in the namespace.
1337 regnum
, is_vec
= yield from get_pdecode_idx_out(self
.dec2
, name
)
1338 # here's where we go "vector". TODO: zero-testing (RA_IS_ZERO)
1339 # XXX already done by PowerDecoder2, now
1341 # regnum += srcstep # TODO, elwidth overrides
1343 # in case getting the register number is needed, _RA, _RB
1344 regname
= "_" + name
1345 self
.namespace
[regname
] = regnum
1346 print('reading reg %s %d' % (name
, regnum
), is_vec
)
1347 reg_val
= self
.gpr(regnum
)
1348 inputs
.append(reg_val
)
1350 # "special" registers
1351 for special
in info
.special_regs
:
1352 if special
in special_sprs
:
1353 inputs
.append(self
.spr
[special
])
1355 inputs
.append(self
.namespace
[special
])
1357 # clear trap (trap) NIA
1358 self
.trap_nia
= None
1360 print("inputs", inputs
)
1361 results
= info
.func(self
, *inputs
)
1362 print("results", results
)
1364 # "inject" decorator takes namespace from function locals: we need to
1365 # overwrite NIA being overwritten (sigh)
1366 if self
.trap_nia
is not None:
1367 self
.namespace
['NIA'] = self
.trap_nia
1369 print("after func", self
.namespace
['CIA'], self
.namespace
['NIA'])
1371 # detect if CA/CA32 already in outputs (sra*, basically)
1374 output_names
= create_args(info
.write_regs
)
1375 for name
in output_names
:
1381 print("carry already done?", bin(already_done
))
1382 if hasattr(self
.dec2
.e
.do
, "output_carry"):
1383 carry_en
= yield self
.dec2
.e
.do
.output_carry
1387 yield from self
.handle_carry_(inputs
, results
, already_done
)
1389 # detect if overflow was in return result
1392 for name
, output
in zip(output_names
, results
):
1393 if name
== 'overflow':
1396 if hasattr(self
.dec2
.e
.do
, "oe"):
1397 ov_en
= yield self
.dec2
.e
.do
.oe
.oe
1398 ov_ok
= yield self
.dec2
.e
.do
.oe
.ok
1402 print("internal overflow", overflow
, ov_en
, ov_ok
)
1404 yield from self
.handle_overflow(inputs
, results
, overflow
)
1406 if hasattr(self
.dec2
.e
.do
, "rc"):
1407 rc_en
= yield self
.dec2
.e
.do
.rc
.rc
1411 regnum
, is_vec
= yield from get_pdecode_cr_out(self
.dec2
, "CR0")
1412 self
.handle_comparison(results
, regnum
)
1414 # any modified return results?
1416 for name
, output
in zip(output_names
, results
):
1417 if name
== 'overflow': # ignore, done already (above)
1419 if isinstance(output
, int):
1420 output
= SelectableInt(output
, 256)
1421 if name
in ['CA', 'CA32']:
1423 print("writing %s to XER" % name
, output
)
1424 self
.spr
['XER'][XER_bits
[name
]] = output
.value
1426 print("NOT writing %s to XER" % name
, output
)
1427 elif name
in info
.special_regs
:
1428 print('writing special %s' % name
, output
, special_sprs
)
1429 if name
in special_sprs
:
1430 self
.spr
[name
] = output
1432 self
.namespace
[name
].eq(output
)
1434 print('msr written', hex(self
.msr
.value
))
1436 regnum
, is_vec
= yield from get_pdecode_idx_out(self
.dec2
,
1439 # temporary hack for not having 2nd output
1440 regnum
= yield getattr(self
.decoder
, name
)
1442 print('writing reg %d %s' % (regnum
, str(output
)), is_vec
)
1443 if output
.bits
> 64:
1444 output
= SelectableInt(output
.value
, 64)
1445 self
.gpr
[regnum
] = output
1447 # check if it is the SVSTATE.src/dest step that needs incrementing
1448 # this is our Sub-Program-Counter loop from 0 to VL-1
1449 if self
.is_svp64_mode
:
1450 # XXX twin predication TODO
1451 vl
= self
.svstate
.vl
.asint(msb0
=True)
1452 mvl
= self
.svstate
.maxvl
.asint(msb0
=True)
1453 srcstep
= self
.svstate
.srcstep
.asint(msb0
=True)
1454 print (" svstate.vl", vl
)
1455 print (" svstate.mvl", mvl
)
1456 print (" svstate.srcstep", srcstep
)
1457 # check if srcstep needs incrementing by one, stop PC advancing
1458 # svp64 loop can end early if the dest is scalar
1459 svp64_dest_vector
= not (yield self
.dec2
.no_out_vec
)
1460 if svp64_dest_vector
and srcstep
!= vl
-1:
1461 self
.svstate
.srcstep
+= SelectableInt(1, 7)
1462 self
.pc
.NIA
.value
= self
.pc
.CIA
.value
1463 self
.namespace
['NIA'] = self
.pc
.NIA
1464 print("end of sub-pc call", self
.namespace
['CIA'],
1465 self
.namespace
['NIA'])
1466 return # DO NOT allow PC to update whilst Sub-PC loop running
1468 self
.svstate
.srcstep
[0:7] = 0
1469 print (" svstate.srcstep loop end (PC to update)")
1470 self
.pc
.update_nia(self
.is_svp64_mode
)
1471 self
.namespace
['NIA'] = self
.pc
.NIA
1473 # UPDATE program counter
1474 self
.pc
.update(self
.namespace
, self
.is_svp64_mode
)
1475 print("end of call", self
.namespace
['CIA'], self
.namespace
['NIA'])
1479 """Decorator factory.
1481 this decorator will "inject" variables into the function's namespace,
1482 from the *dictionary* in self.namespace. it therefore becomes possible
1483 to make it look like a whole stack of variables which would otherwise
1484 need "self." inserted in front of them (*and* for those variables to be
1485 added to the instance) "appear" in the function.
1487 "self.namespace['SI']" for example becomes accessible as just "SI" but
1488 *only* inside the function, when decorated.
1490 def variable_injector(func
):
1492 def decorator(*args
, **kwargs
):
1494 func_globals
= func
.__globals
__ # Python 2.6+
1495 except AttributeError:
1496 func_globals
= func
.func_globals
# Earlier versions.
1498 context
= args
[0].namespace
# variables to be injected
1499 saved_values
= func_globals
.copy() # Shallow copy of dict.
1500 func_globals
.update(context
)
1501 result
= func(*args
, **kwargs
)
1502 print("globals after", func_globals
['CIA'], func_globals
['NIA'])
1503 print("args[0]", args
[0].namespace
['CIA'],
1504 args
[0].namespace
['NIA'])
1505 args
[0].namespace
= func_globals
1506 #exec (func.__code__, func_globals)
1509 # func_globals = saved_values # Undo changes.
1515 return variable_injector
1518 # very quick test of maskgen function (TODO, move to util later)
1519 if __name__
== '__main__':
1520 shift
= SelectableInt(5, 6)
1521 mask
= genmask(shift
, 43)
1522 print (" mask", bin(mask
.value
))
1524 mem
= Mem(row_bytes
=8)
1525 mem
= RADIX(mem
, None)
1526 # -----------------------------------------------
1527 # |/|RTS1|/| RPDB | RTS2 | RPDS |
1528 # -----------------------------------------------
1529 # |0|1 2|3|4 55|56 58|59 63|
1530 data
= SelectableInt(0, 64)
1533 data
[59:64] = 0b01101 # mask
1535 (rts
, mbits
, pgbase
) = mem
._decode
_prte
(data
)
1536 print (" rts", bin(rts
.value
), rts
.bits
)
1537 print (" mbits", bin(mbits
.value
), mbits
.bits
)
1538 print (" pgbase", hex(pgbase
.value
), pgbase
.bits
)
1539 addr
= SelectableInt(0x1000, 64)
1540 check
= mem
._segment
_check
(addr
, mbits
, shift
)
1541 print (" segment check", check
)