+# SPDX-License-Identifier: LGPLv3+
+# Copyright (C) 2020, 2021 Luke Kenneth Casson Leighton <lkcl@lkcl.net>
+# Copyright (C) 2020 Michael Nolan
+# Funded by NLnet http://nlnet.nl
"""core of the python-based POWER9 simulator
this is part of a cycle-accurate POWER9 simulator. its primary purpose is
* https://bugs.libre-soc.org/show_bug.cgi?id=424
"""
+from nmigen.back.pysim import Settle
from functools import wraps
from copy import copy
from soc.decoder.orderedset import OrderedSet
from soc.decoder.selectable_int import (FieldSelectableInt, SelectableInt,
selectconcat)
from soc.decoder.power_enums import (spr_dict, spr_byname, XER_bits,
- insns, MicrOp)
+ insns, MicrOp, In1Sel, In2Sel, In3Sel,
+ OutSel, CROutSel)
from soc.decoder.helpers import exts, gtu, ltu, undefined
from soc.consts import PIb, MSRb # big-endian (PowerISA versions)
+from soc.decoder.power_svp64 import SVP64RM, decode_extra
from collections import namedtuple
import math
return retval
+"""
+ Get Root Page
+
+ //Accessing 2nd double word of partition table (pate1)
+ //Ref: Power ISA Manual v3.0B, Book-III, section 5.7.6.1
+ // PTCR Layout
+ // ====================================================
+ // -----------------------------------------------
+ // | /// | PATB | /// | PATS |
+ // -----------------------------------------------
+ // 0 4 51 52 58 59 63
+ // PATB[4:51] holds the base address of the Partition Table,
+ // right shifted by 12 bits.
+ // This is because the address of the Partition base is
+ // 4k aligned. Hence, the lower 12bits, which are always
+ // 0 are ommitted from the PTCR.
+ //
+ // Thus, The Partition Table Base is obtained by (PATB << 12)
+ //
+ // PATS represents the partition table size right-shifted by 12 bits.
+ // The minimal size of the partition table is 4k.
+ // Thus partition table size = (1 << PATS + 12).
+ //
+ // Partition Table
+ // ====================================================
+ // 0 PATE0 63 PATE1 127
+ // |----------------------|----------------------|
+ // | | |
+ // |----------------------|----------------------|
+ // | | |
+ // |----------------------|----------------------|
+ // | | | <-- effLPID
+ // |----------------------|----------------------|
+ // .
+ // .
+ // .
+ // |----------------------|----------------------|
+ // | | |
+ // |----------------------|----------------------|
+ //
+ // The effective LPID forms the index into the Partition Table.
+ //
+ // Each entry in the partition table contains 2 double words, PATE0, PATE1,
+ // corresponding to that partition.
+ //
+ // In case of Radix, The structure of PATE0 and PATE1 is as follows.
+ //
+ // PATE0 Layout
+ // -----------------------------------------------
+ // |1|RTS1|/| RPDB | RTS2 | RPDS |
+ // -----------------------------------------------
+ // 0 1 2 3 4 55 56 58 59 63
+ //
+ // HR[0] : For Radix Page table, first bit should be 1.
+ // RTS1[1:2] : Gives one fragment of the Radix treesize
+ // RTS2[56:58] : Gives the second fragment of the Radix Tree size.
+ // RTS = (RTS1 << 3 + RTS2) + 31.
+ //
+ // RPDB[4:55] = Root Page Directory Base.
+ // RPDS = Logarithm of Root Page Directory Size right shifted by 3.
+ // Thus, Root page directory size = 1 << (RPDS + 3).
+ // Note: RPDS >= 5.
+ //
+ // PATE1 Layout
+ // -----------------------------------------------
+ // |///| PRTB | // | PRTS |
+ // -----------------------------------------------
+ // 0 3 4 51 52 58 59 63
+ //
+ // PRTB[4:51] = Process Table Base. This is aligned to size.
+ // PRTS[59: 63] = Process Table Size right shifted by 12.
+ // Minimal size of the process table is 4k.
+ // Process Table Size = (1 << PRTS + 12).
+ // Note: PRTS <= 24.
+ //
+ // Computing the size aligned Process Table Base:
+ // table_base = (PRTB & ~((1 << PRTS) - 1)) << 12
+ // Thus, the lower 12+PRTS bits of table_base will
+ // be zero.
+
+
+ //Ref: Power ISA Manual v3.0B, Book-III, section 5.7.6.2
+ //
+ // Process Table
+ // ==========================
+ // 0 PRTE0 63 PRTE1 127
+ // |----------------------|----------------------|
+ // | | |
+ // |----------------------|----------------------|
+ // | | |
+ // |----------------------|----------------------|
+ // | | | <-- effPID
+ // |----------------------|----------------------|
+ // .
+ // .
+ // .
+ // |----------------------|----------------------|
+ // | | |
+ // |----------------------|----------------------|
+ //
+ // The effective Process id (PID) forms the index into the Process Table.
+ //
+ // Each entry in the partition table contains 2 double words, PRTE0, PRTE1,
+ // corresponding to that process
+ //
+ // In case of Radix, The structure of PRTE0 and PRTE1 is as follows.
+ //
+ // PRTE0 Layout
+ // -----------------------------------------------
+ // |/|RTS1|/| RPDB | RTS2 | RPDS |
+ // -----------------------------------------------
+ // 0 1 2 3 4 55 56 58 59 63
+ //
+ // RTS1[1:2] : Gives one fragment of the Radix treesize
+ // RTS2[56:58] : Gives the second fragment of the Radix Tree size.
+ // RTS = (RTS1 << 3 + RTS2) << 31,
+ // since minimal Radix Tree size is 4G.
+ //
+ // RPDB = Root Page Directory Base.
+ // RPDS = Root Page Directory Size right shifted by 3.
+ // Thus, Root page directory size = RPDS << 3.
+ // Note: RPDS >= 5.
+ //
+ // PRTE1 Layout
+ // -----------------------------------------------
+ // | /// |
+ // -----------------------------------------------
+ // 0 63
+ // All bits are reserved.
+
+
+"""
+
+# see qemu/target/ppc/mmu-radix64.c for reference
+class RADIX:
+ def __init__(self, mem, caller):
+ self.mem = mem
+ self.caller = caller
+
+ # cached page table stuff
+ self.pgtbl0 = 0
+ self.pt0_valid = False
+ self.pgtbl3 = 0
+ self.pt3_valid = False
+
+ def ld(self, address, width=8, swap=True, check_in_mem=False):
+ print("RADIX: ld from addr 0x{:x} width {:d}".format(address, width))
+
+ pte = self._walk_tree()
+ # use pte to caclculate phys address
+ #mem.ld(address,width,swap,check_in_mem)
+
+ # TODO implement
+ # def st(self, addr, v, width=8, swap=True):
+ # def memassign(self, addr, sz, val):
+ def _next_level(self):
+ return True
+ ## DSISR_R_BADCONFIG
+ ## read_entry
+ ## DSISR_NOPTE
+ ## Prepare for next iteration
+
+ def _walk_tree(self):
+ """walk tree
+
+ // vaddr 64 Bit
+ // vaddr |-----------------------------------------------------|
+ // | Unused | Used |
+ // |-----------|-----------------------------------------|
+ // | 0000000 | usefulBits = X bits (typically 52) |
+ // |-----------|-----------------------------------------|
+ // | |<--Cursize---->| |
+ // | | Index | |
+ // | | into Page | |
+ // | | Directory | |
+ // |-----------------------------------------------------|
+ // | |
+ // V |
+ // PDE |---------------------------| |
+ // |V|L|//| NLB |///|NLS| |
+ // |---------------------------| |
+ // PDE = Page Directory Entry |
+ // [0] = V = Valid Bit |
+ // [1] = L = Leaf bit. If 0, then |
+ // [4:55] = NLB = Next Level Base |
+ // right shifted by 8 |
+ // [59:63] = NLS = Next Level Size |
+ // | NLS >= 5 |
+ // | V
+ // | |--------------------------|
+ // | | usfulBits = X-Cursize |
+ // | |--------------------------|
+ // |---------------------><--NLS-->| |
+ // | Index | |
+ // | into | |
+ // | PDE | |
+ // |--------------------------|
+ // |
+ // If the next PDE obtained by |
+ // (NLB << 8 + 8 * index) is a |
+ // nonleaf, then repeat the above. |
+ // |
+ // If the next PDE is a leaf, |
+ // then Leaf PDE structure is as |
+ // follows |
+ // |
+ // |
+ // Leaf PDE |
+ // |------------------------------| |----------------|
+ // |V|L|sw|//|RPN|sw|R|C|/|ATT|EAA| | usefulBits |
+ // |------------------------------| |----------------|
+ // [0] = V = Valid Bit |
+ // [1] = L = Leaf Bit = 1 if leaf |
+ // PDE |
+ // [2] = Sw = Sw bit 0. |
+ // [7:51] = RPN = Real Page Number, V
+ // real_page = RPN << 12 -------------> Logical OR
+ // [52:54] = Sw Bits 1:3 |
+ // [55] = R = Reference |
+ // [56] = C = Change V
+ // [58:59] = Att = Physical Address
+ // 0b00 = Normal Memory
+ // 0b01 = SAO
+ // 0b10 = Non Idenmpotent
+ // 0b11 = Tolerant I/O
+ // [60:63] = Encoded Access
+ // Authority
+ //
+ """
+ # walk tree starts on prtbl
+ while True:
+ ret = self._next_level()
+ if ret: return ret
+
+ def _segment_check(self):
+ """checks segment valid
+ mbits := '0' & r.mask_size;
+ v.shift := r.shift + (31 - 12) - mbits;
+ nonzero := or(r.addr(61 downto 31) and not finalmask(30 downto 0));
+ if r.addr(63) /= r.addr(62) or nonzero = '1' then
+ v.state := RADIX_FINISH;
+ v.segerror := '1';
+ elsif mbits < 5 or mbits > 16 or mbits > (r.shift + (31 - 12)) then
+ v.state := RADIX_FINISH;
+ v.badtree := '1';
+ else
+ v.state := RADIX_LOOKUP;
+ """
+
+ def _check_perms(self):
+ """check page permissions
+ -- test leaf bit
+ if data(62) = '1' then
+ -- check permissions and RC bits
+ perm_ok := '0';
+ if r.priv = '1' or data(3) = '0' then
+ if r.iside = '0' then
+ perm_ok := data(1) or (data(2) and not r.store);
+ else
+ -- no IAMR, so no KUEP support for now
+ -- deny execute permission if cache inhibited
+ perm_ok := data(0) and not data(5);
+ end if;
+ end if;
+ rc_ok := data(8) and (data(7) or not r.store);
+ if perm_ok = '1' and rc_ok = '1' then
+ v.state := RADIX_LOAD_TLB;
+ else
+ v.state := RADIX_FINISH;
+ v.perm_err := not perm_ok;
+ -- permission error takes precedence over RC error
+ v.rc_error := perm_ok;
+ end if;
+ """
+
+
class Mem:
def __init__(self, row_bytes=8, initial_mem=None):
class GPR(dict):
- def __init__(self, decoder, regfile):
+ def __init__(self, decoder, isacaller, svstate, regfile):
dict.__init__(self)
self.sd = decoder
+ self.isacaller = isacaller
+ self.svstate = svstate
for i in range(32):
self[i] = SelectableInt(regfile[i], 64)
return rnum
def ___getitem__(self, attr):
- print("GPR getitem", attr)
+ """ XXX currently not used
+ """
rnum = self._get_regnum(attr)
+ offs = self.svstate.srcstep
+ print("GPR getitem", attr, rnum, "srcoffs", offs)
return self.regfile[rnum]
def dump(self):
class PC:
def __init__(self, pc_init=0):
self.CIA = SelectableInt(pc_init, 64)
- self.NIA = self.CIA + SelectableInt(4, 64)
+ self.NIA = self.CIA + SelectableInt(4, 64) # only true for v3.0B!
+
+ def update_nia(self, is_svp64):
+ increment = 8 if is_svp64 else 4
+ self.NIA = self.CIA + SelectableInt(increment, 64)
- def update(self, namespace):
+ def update(self, namespace, is_svp64):
+ """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
+ """
self.CIA = namespace['NIA'].narrow(64)
- self.NIA = self.CIA + SelectableInt(4, 64)
+ self.update_nia(is_svp64)
namespace['CIA'] = self.CIA
namespace['NIA'] = self.NIA
+# Simple-V: see https://libre-soc.org/openpower/sv
+class SVP64State:
+ def __init__(self, init=0):
+ self.spr = SelectableInt(init, 32)
+ # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
+ self.maxvl = FieldSelectableInt(self.spr, tuple(range(0,7)))
+ self.vl = FieldSelectableInt(self.spr, tuple(range(7,14)))
+ self.srcstep = FieldSelectableInt(self.spr, tuple(range(14,21)))
+ self.dststep = FieldSelectableInt(self.spr, tuple(range(21,28)))
+ self.subvl = FieldSelectableInt(self.spr, tuple(range(28,30)))
+ self.svstep = FieldSelectableInt(self.spr, tuple(range(30,32)))
+
+
+# SVP64 ReMap field
+class SVP64RMFields:
+ def __init__(self, init=0):
+ self.spr = SelectableInt(init, 24)
+ # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
+ self.mmode = FieldSelectableInt(self.spr, [0])
+ self.mask = FieldSelectableInt(self.spr, tuple(range(1,4)))
+ self.elwidth = FieldSelectableInt(self.spr, tuple(range(4,6)))
+ self.ewsrc = FieldSelectableInt(self.spr, tuple(range(6,8)))
+ self.subvl = FieldSelectableInt(self.spr, tuple(range(8,10)))
+ self.extra = FieldSelectableInt(self.spr, tuple(range(10,19)))
+ self.mode = FieldSelectableInt(self.spr, tuple(range(19,24)))
+ # these cover the same extra field, split into parts as EXTRA2
+ self.extra2 = list(range(4))
+ self.extra2[0] = FieldSelectableInt(self.spr, tuple(range(10,12)))
+ self.extra2[1] = FieldSelectableInt(self.spr, tuple(range(12,14)))
+ self.extra2[2] = FieldSelectableInt(self.spr, tuple(range(14,16)))
+ self.extra2[3] = FieldSelectableInt(self.spr, tuple(range(16,18)))
+ self.smask = FieldSelectableInt(self.spr, tuple(range(16,19)))
+ # and here as well, but EXTRA3
+ self.extra3 = list(range(3))
+ self.extra3[0] = FieldSelectableInt(self.spr, tuple(range(10,13)))
+ self.extra3[1] = FieldSelectableInt(self.spr, tuple(range(13,16)))
+ self.extra3[2] = FieldSelectableInt(self.spr, tuple(range(16,19)))
+
+
+SVP64RM_MMODE_SIZE = len(SVP64RMFields().mmode.br)
+SVP64RM_MASK_SIZE = len(SVP64RMFields().mask.br)
+SVP64RM_ELWIDTH_SIZE = len(SVP64RMFields().elwidth.br)
+SVP64RM_EWSRC_SIZE = len(SVP64RMFields().ewsrc.br)
+SVP64RM_SUBVL_SIZE = len(SVP64RMFields().subvl.br)
+SVP64RM_EXTRA2_SPEC_SIZE = len(SVP64RMFields().extra2[0].br)
+SVP64RM_EXTRA3_SPEC_SIZE = len(SVP64RMFields().extra3[0].br)
+SVP64RM_SMASK_SIZE = len(SVP64RMFields().smask.br)
+SVP64RM_MODE_SIZE = len(SVP64RMFields().mode.br)
+
+
+# SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
+class SVP64PrefixFields:
+ def __init__(self):
+ self.insn = SelectableInt(0, 32)
+ # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
+ self.major = FieldSelectableInt(self.insn, tuple(range(0,6)))
+ self.pid = FieldSelectableInt(self.insn, (7, 9)) # must be 0b11
+ rmfields = [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
+ self.rm = FieldSelectableInt(self.insn, rmfields)
+
+
+SV64P_MAJOR_SIZE = len(SVP64PrefixFields().major.br)
+SV64P_PID_SIZE = len(SVP64PrefixFields().pid.br)
+SV64P_RM_SIZE = len(SVP64PrefixFields().rm.br)
+
+
class SPR(dict):
def __init__(self, dec2, initial_sprs={}):
self.sd = dec2
def __call__(self, ridx):
return self[ridx]
+def get_pdecode_idx_in(dec2, name):
+ op = dec2.dec.op
+ in1_sel = yield op.in1_sel
+ in2_sel = yield op.in2_sel
+ in3_sel = yield op.in3_sel
+ # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
+ in1 = yield dec2.e.read_reg1.data
+ in2 = yield dec2.e.read_reg2.data
+ in3 = yield dec2.e.read_reg3.data
+ in1_isvec = yield dec2.in1_isvec
+ in2_isvec = yield dec2.in2_isvec
+ in3_isvec = yield dec2.in3_isvec
+ print ("get_pdecode_idx", in1_sel, In1Sel.RA.value, in1, in1_isvec)
+ # identify which regnames map to in1/2/3
+ if name == 'RA':
+ if (in1_sel == In1Sel.RA.value or
+ (in1_sel == In1Sel.RA_OR_ZERO.value and in1 != 0)):
+ return in1, in1_isvec
+ if in1_sel == In1Sel.RA_OR_ZERO.value:
+ return in1, in1_isvec
+ elif name == 'RB':
+ if in2_sel == In2Sel.RB.value:
+ return in2, in2_isvec
+ if in3_sel == In3Sel.RB.value:
+ return in3, in3_isvec
+ # XXX TODO, RC doesn't exist yet!
+ elif name == 'RC':
+ assert False, "RC does not exist yet"
+ elif name == 'RS':
+ if in1_sel == In1Sel.RS.value:
+ return in1, in1_isvec
+ if in2_sel == In2Sel.RS.value:
+ return in2, in2_isvec
+ if in3_sel == In3Sel.RS.value:
+ return in3, in3_isvec
+ return None, False
+
+
+def get_pdecode_cr_out(dec2, name):
+ op = dec2.dec.op
+ out_sel = yield op.cr_out
+ out_bitfield = yield dec2.dec_cr_out.cr_bitfield.data
+ sv_cr_out = yield op.sv_cr_out
+ spec = yield dec2.crout_svdec.spec
+ sv_override = yield dec2.dec_cr_out.sv_override
+ # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
+ out = yield dec2.e.write_cr.data
+ o_isvec = yield dec2.o_isvec
+ print ("get_pdecode_cr_out", out_sel, CROutSel.CR0.value, out, o_isvec)
+ print (" sv_cr_out", sv_cr_out)
+ print (" cr_bf", out_bitfield)
+ print (" spec", spec)
+ print (" override", sv_override)
+ # identify which regnames map to out / o2
+ if name == 'CR0':
+ if out_sel == CROutSel.CR0.value:
+ return out, o_isvec
+ print ("get_pdecode_idx_out not found", name)
+ return None, False
+
+
+def get_pdecode_idx_out(dec2, name):
+ op = dec2.dec.op
+ out_sel = yield op.out_sel
+ # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
+ out = yield dec2.e.write_reg.data
+ o_isvec = yield dec2.o_isvec
+ print ("get_pdecode_idx_out", out_sel, OutSel.RA.value, out, o_isvec)
+ # identify which regnames map to out / o2
+ if name == 'RA':
+ if out_sel == OutSel.RA.value:
+ return out, o_isvec
+ elif name == 'RT':
+ if out_sel == OutSel.RT.value:
+ return out, o_isvec
+ print ("get_pdecode_idx_out not found", name)
+ return None, False
+
+
+# XXX TODO
+def get_pdecode_idx_out2(dec2, name):
+ op = dec2.dec.op
+ print ("TODO: get_pdecode_idx_out2", name)
+ return None, False
+
class ISACaller:
# decoder2 - an instance of power_decoder2
# respect_pc - tracks the program counter. requires initial_insns
def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
initial_mem=None, initial_msr=0,
+ initial_svstate=0,
initial_insns=None, respect_pc=False,
disassembly=None,
initial_pc=0,
- bigendian=False):
+ bigendian=False,
+ mmu=False):
self.bigendian = bigendian
self.halted = False
+ self.is_svp64_mode = False
self.respect_pc = respect_pc
if initial_sprs is None:
initial_sprs = {}
self.disassembly[i*4 + disasm_start] = code
# set up registers, instruction memory, data memory, PC, SPRs, MSR
- self.gpr = GPR(decoder2, regfile)
+ self.svp64rm = SVP64RM()
+ if isinstance(initial_svstate, int):
+ initial_svstate = SVP64State(initial_svstate)
+ self.svstate = initial_svstate
+ self.gpr = GPR(decoder2, self, self.svstate, regfile)
self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
+ if mmu:
+ self.mem = RADIX(self.mem,self)
self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
self.pc = PC()
self.spr = SPR(decoder2, initial_sprs)
so = so | ov
self.spr['XER'][XER_bits['SO']] = so
- def handle_comparison(self, outputs):
+ def handle_comparison(self, outputs, cr_idx=0):
out = outputs[0]
assert isinstance(out, SelectableInt), \
"out zero not a SelectableInt %s" % repr(outputs)
SO = self.spr['XER'][XER_bits['SO']]
print("handle_comparison SO", SO)
cr_field = selectconcat(negative, positive, zero, SO)
- self.crl[0].eq(cr_field)
+ self.crl[cr_idx].eq(cr_field)
def set_pc(self, pc_val):
self.namespace['NIA'] = SelectableInt(pc_val, 64)
- self.pc.update(self.namespace)
+ self.pc.update(self.namespace, self.is_svp64_mode)
def setup_one(self):
"""set up one instruction
print("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
print("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
+ yield self.dec2.sv_rm.eq(0)
yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
yield self.dec2.dec.bigendian.eq(self.bigendian)
yield self.dec2.state.msr.eq(self.msr.value)
yield self.dec2.state.pc.eq(pc)
+ yield self.dec2.state.svstate.eq(self.svstate.spr.value)
+
+ # SVP64. first, check if the opcode is EXT001, and SVP64 id bits set
+ yield Settle()
+ opcode = yield self.dec2.dec.opcode_in
+ pfx = SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
+ pfx.insn.value = opcode
+ major = pfx.major.asint(msb0=True) # MSB0 inversion
+ print ("prefix test: opcode:", major, bin(major),
+ pfx.insn[7] == 0b1, pfx.insn[9] == 0b1)
+ self.is_svp64_mode = ((major == 0b000001) and
+ pfx.insn[7].value == 0b1 and
+ pfx.insn[9].value == 0b1)
+ self.pc.update_nia(self.is_svp64_mode)
+ if not self.is_svp64_mode:
+ return
+
+ # in SVP64 mode. decode/print out svp64 prefix, get v3.0B instruction
+ print ("svp64.rm", bin(pfx.rm.asint(msb0=True)))
+ print (" svstate.vl", self.svstate.vl.asint(msb0=True))
+ print (" svstate.mvl", self.svstate.maxvl.asint(msb0=True))
+ sv_rm = pfx.rm.asint(msb0=True)
+ ins = self.imem.ld(pc+4, 4, False, True)
+ print(" svsetup: 0x%x 0x%x %s" % (pc+4, ins & 0xffffffff, bin(ins)))
+ yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff) # v3.0B suffix
+ yield self.dec2.sv_rm.eq(sv_rm) # svp64 prefix
+ yield Settle()
def execute_one(self):
"""execute one instruction
"""
# get the disassembly code for this instruction
- code = self.disassembly[self._pc]
- print("sim-execute", hex(self._pc), code)
+ if self.is_svp64_mode:
+ code = self.disassembly[self._pc+4]
+ print(" svp64 sim-execute", hex(self._pc), code)
+ else:
+ code = self.disassembly[self._pc]
+ print("sim-execute", hex(self._pc), code)
opname = code.split(' ')[0]
yield from self.call(opname)
+ # don't use this except in special circumstances
if not self.respect_pc:
self.fake_pc += 4
+
print("execute one, CIA NIA", self.pc.CIA.value, self.pc.NIA.value)
def get_assembly_name(self):
return dec_insn & (1 << 20) != 0 # sigh - XFF.spr[-1]?
def call(self, name):
+ """call(opcode) - the primary execution point for instructions
+ """
name = name.strip() # remove spaces if not already done so
if self.halted:
print("halted - not executing", name)
if instr_is_privileged and self.msr[MSRb.PR] == 1:
self.TRAP(0x700, PIb.PRIV)
self.namespace['NIA'] = self.trap_nia
- self.pc.update(self.namespace)
+ self.pc.update(self.namespace, self.is_svp64_mode)
return
# check halted condition
print("illegal", name, asmop)
self.TRAP(0x700, PIb.ILLEG)
self.namespace['NIA'] = self.trap_nia
- self.pc.update(self.namespace)
+ self.pc.update(self.namespace, self.is_svp64_mode)
print("name %s != %s - calling ILLEGAL trap, PC: %x" %
(name, asmop, self.pc.CIA.value))
return
list(info.uninit_regs))
print(input_names)
- # main registers (RT, RA ...)
+ # get SVP64 entry for the current instruction
+ sv_rm = self.svp64rm.instrs.get(name)
+ if sv_rm is not None:
+ dest_cr, src_cr, src_byname, dest_byname = decode_extra(sv_rm)
+ else:
+ dest_cr, src_cr, src_byname, dest_byname = False, False, {}, {}
+ print ("sv rm", sv_rm, dest_cr, src_cr, src_byname, dest_byname)
+
+ # get SVSTATE srcstep. TODO: dststep (twin predication)
+ srcstep = self.svstate.srcstep.asint(msb0=True)
+ vl = self.svstate.vl.asint(msb0=True)
+ mvl = self.svstate.maxvl.asint(msb0=True)
+
+ # VL=0 in SVP64 mode means "do nothing: skip instruction"
+ if self.is_svp64_mode and vl == 0:
+ self.pc.update(self.namespace, self.is_svp64_mode)
+ print("end of call", self.namespace['CIA'], self.namespace['NIA'])
+ return
+
+ # main input registers (RT, RA ...)
inputs = []
for name in input_names:
- regnum = yield getattr(self.decoder, name)
+ # using PowerDecoder2, first, find the decoder index.
+ # (mapping name RA RB RC RS to in1, in2, in3)
+ regnum, is_vec = yield from get_pdecode_idx_in(self.dec2, name)
+ if regnum is None:
+ # doing this is not part of svp64, it's because output
+ # registers, to be modified, need to be in the namespace.
+ regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, name)
+ # here's where we go "vector". TODO: zero-testing (RA_IS_ZERO)
+ # XXX already done by PowerDecoder2, now
+ #if is_vec:
+ # regnum += srcstep # TODO, elwidth overrides
+
+ # in case getting the register number is needed, _RA, _RB
regname = "_" + name
self.namespace[regname] = regnum
- print('reading reg %d' % regnum)
- inputs.append(self.gpr(regnum))
+ print('reading reg %s %d' % (name, regnum), is_vec)
+ reg_val = self.gpr(regnum)
+ inputs.append(reg_val)
# "special" registers
for special in info.special_regs:
# clear trap (trap) NIA
self.trap_nia = None
- print(inputs)
+ print("inputs", inputs)
results = info.func(self, *inputs)
- print(results)
+ print("results", results)
# "inject" decorator takes namespace from function locals: we need to
# overwrite NIA being overwritten (sigh)
else:
rc_en = False
if rc_en:
- self.handle_comparison(results)
+ regnum, is_vec = yield from get_pdecode_cr_out(self.dec2, "CR0")
+ self.handle_comparison(results, regnum)
# any modified return results?
if info.write_regs:
if name == 'MSR':
print('msr written', hex(self.msr.value))
else:
- regnum = yield getattr(self.decoder, name)
- print('writing reg %d %s' % (regnum, str(output)))
+ regnum, is_vec = yield from get_pdecode_idx_out(self.dec2,
+ name)
+ if regnum is None:
+ # temporary hack for not having 2nd output
+ regnum = yield getattr(self.decoder, name)
+ is_vec = False
+ print('writing reg %d %s' % (regnum, str(output)), is_vec)
if output.bits > 64:
output = SelectableInt(output.value, 64)
self.gpr[regnum] = output
- print("end of call", self.namespace['CIA'], self.namespace['NIA'])
+ # check if it is the SVSTATE.src/dest step that needs incrementing
+ # this is our Sub-Program-Counter loop from 0 to VL-1
+ if self.is_svp64_mode:
+ # XXX twin predication TODO
+ vl = self.svstate.vl.asint(msb0=True)
+ mvl = self.svstate.maxvl.asint(msb0=True)
+ srcstep = self.svstate.srcstep.asint(msb0=True)
+ print (" svstate.vl", vl)
+ print (" svstate.mvl", mvl)
+ print (" svstate.srcstep", srcstep)
+ # check if srcstep needs incrementing by one, stop PC advancing
+ # svp64 loop can end early if the dest is scalar
+ svp64_dest_vector = not (yield self.dec2.no_out_vec)
+ if svp64_dest_vector and srcstep != vl-1:
+ self.svstate.srcstep += SelectableInt(1, 7)
+ self.pc.NIA.value = self.pc.CIA.value
+ self.namespace['NIA'] = self.pc.NIA
+ print("end of sub-pc call", self.namespace['CIA'],
+ self.namespace['NIA'])
+ return # DO NOT allow PC to update whilst Sub-PC loop running
+ # reset to zero
+ self.svstate.srcstep[0:7] = 0
+ print (" svstate.srcstep loop end (PC to update)")
+ self.pc.update_nia(self.is_svp64_mode)
+ self.namespace['NIA'] = self.pc.NIA
+
# UPDATE program counter
- self.pc.update(self.namespace)
+ self.pc.update(self.namespace, self.is_svp64_mode)
+ print("end of call", self.namespace['CIA'], self.namespace['NIA'])
def inject():