* https://bugs.libre-soc.org/show_bug.cgi?id=424
"""
-import re
-from nmigen.sim import Settle, Delay
+from collections import namedtuple
+from copy import deepcopy
from functools import wraps
-from copy import copy, deepcopy
-from openpower.decoder.orderedset import OrderedSet
-from openpower.decoder.selectable_int import (
- FieldSelectableInt,
- SelectableInt,
- selectconcat,
-)
-from openpower.decoder.power_insn import SVP64Instruction
-from openpower.decoder.power_enums import (spr_dict, spr_byname, XER_bits,
- insns, MicrOp,
- In1Sel, In2Sel, In3Sel,
- OutSel, CRInSel, CROutSel, LDSTMode,
- SVMode,
- SVP64RMMode, SVP64PredMode,
- SVP64PredInt, SVP64PredCR,
- SVP64LDSTmode, FPTRANS_INSNS)
-
-from openpower.decoder.power_enums import SVPtype
-
-from openpower.decoder.helpers import (exts, gtu, ltu, undefined,
- ISACallerHelper, ISAFPHelpers)
-from openpower.consts import PIb, MSRb # big-endian (PowerISA versions)
-from openpower.consts import (SVP64MODE, SVP64MODEb,
- SVP64CROffs,
- )
-from openpower.decoder.power_svp64 import SVP64RM, decode_extra
+from nmigen.sim import Settle
+from openpower.consts import (MSRb, PIb, # big-endian (PowerISA versions)
+ SVP64CROffs, SVP64MODEb)
+from openpower.decoder.helpers import (ISACallerHelper, ISAFPHelpers, exts,
+ gtu, undefined)
+from openpower.decoder.isa.mem import Mem, MemException
from openpower.decoder.isa.radixmmu import RADIX
-from openpower.decoder.isa.mem import Mem, swap_order, MemException
from openpower.decoder.isa.svshape import SVSHAPE
from openpower.decoder.isa.svstate import SVP64State
-
-
+from openpower.decoder.orderedset import OrderedSet
+from openpower.decoder.power_enums import (FPTRANS_INSNS, CRInSel, CROutSel,
+ In1Sel, In2Sel, In3Sel, LDSTMode,
+ MicrOp, OutSel, SVMode,
+ SVP64LDSTmode, SVP64PredCR,
+ SVP64PredInt, SVP64PredMode,
+ SVP64RMMode, SVPType, XER_bits,
+ insns, spr_byname, spr_dict,
+ BFP_FLAG_NAMES)
+from openpower.decoder.power_insn import SVP64Instruction
+from openpower.decoder.power_svp64 import SVP64RM, decode_extra
+from openpower.decoder.selectable_int import (FieldSelectableInt,
+ SelectableInt, selectconcat,
+ EFFECTIVELY_UNLIMITED)
+from openpower.fpscr import FPSCRState
+from openpower.xer import XERState
from openpower.util import LogKind, log
-from collections import namedtuple
-import math
-import sys
+LDST_UPDATE_INSNS = ['ldu', 'lwzu', 'lbzu', 'lhzu', 'lhau', 'lfsu', 'lfdu',
+ ]
instruction_info = namedtuple('instruction_info',
'func read_regs uninit_regs write_regs ' +
"CTR": 0,
"TAR": 0,
"MSR": 0,
+ "FPSCR": 0,
"SVSTATE": 0,
"SVSHAPE0": 0,
"SVSHAPE1": 0,
fregs = ['FRA', 'FRB', 'FRC', 'FRS', 'FRT']
+def get_masked_reg(regs, base, offs, ew_bits):
+ # rrrright. start by breaking down into row/col, based on elwidth
+ gpr_offs = offs // (64 // ew_bits)
+ gpr_col = offs % (64 // ew_bits)
+ # compute the mask based on ew_bits
+ mask = (1 << ew_bits) - 1
+ # now select the 64-bit register, but get its value (easier)
+ val = regs[base + gpr_offs]
+ # shift down so element we want is at LSB
+ val >>= gpr_col * ew_bits
+ # mask so we only return the LSB element
+ return val & mask
+
+
+def set_masked_reg(regs, base, offs, ew_bits, value):
+ # rrrright. start by breaking down into row/col, based on elwidth
+ gpr_offs = offs // (64//ew_bits)
+ gpr_col = offs % (64//ew_bits)
+ # compute the mask based on ew_bits
+ mask = (1 << ew_bits)-1
+ # now select the 64-bit register, but get its value (easier)
+ val = regs[base+gpr_offs]
+ # now mask out the bit we don't want
+ val = val & ~(mask << (gpr_col*ew_bits))
+ # then wipe the bit we don't want from the value
+ value = value & mask
+ # OR the new value in, shifted up
+ val |= value << (gpr_col*ew_bits)
+ regs[base+gpr_offs] = val
+
+
def create_args(reglist, extra=None):
retval = list(OrderedSet(reglist))
retval.sort(key=lambda reg: REG_SORT_ORDER.get(reg, 0))
# now select the 64-bit register, but get its value (easier)
val = self[ridx+gpr_offs].value
# now shift down and mask out
- val = val >> (gpr_col*elwidth) & ((1<<elwidth)-1)
+ val = val >> (gpr_col*elwidth) & ((1 << elwidth)-1)
# finally, return a SelectableInt at the required elwidth
log("GPR call", ridx, "isvec", is_vec, "offs", offs,
- "elwid", elwidth, "offs/col", gpr_offs, gpr_col, "val", hex(val))
+ "elwid", elwidth, "offs/col", gpr_offs, gpr_col, "val", hex(val))
return SelectableInt(val, elwidth)
def set_form(self, form):
gpr_offs = offs // (64//elwidth)
gpr_col = offs % (64//elwidth)
# compute the mask based on elwidth
- mask = (1<<elwidth)-1
+ mask = (1 << elwidth)-1
# now select the 64-bit register, but get its value (easier)
val = self[base+gpr_offs].value
# now mask out the bit we don't want
val |= value << (gpr_col*elwidth)
# finally put the damn value into the regfile
log("GPR write", base, "isvec", is_vec, "offs", offs,
- "elwid", elwidth, "offs/col", gpr_offs, gpr_col, "val", hex(val),
- "@", base+gpr_offs)
+ "elwid", elwidth, "offs/col", gpr_offs, gpr_col, "val", hex(val),
+ "@", base+gpr_offs)
dict.__setitem__(self, base+gpr_offs, SelectableInt(val, 64))
def __setitem__(self, rnum, value):
info = spr_dict[key]
else:
info = spr_byname[key]
- dict.__setitem__(self, key, SelectableInt(0, info.length))
+ self[key] = SelectableInt(0, info.length)
res = dict.__getitem__(self, key)
log("spr returning", key, res)
return res
self.__setitem__('SRR0', value)
if key == 'HSRR1': # HACK!
self.__setitem__('SRR1', value)
+ if key == 1:
+ value = XERState(value)
log("setting spr", key, value)
dict.__setitem__(self, key, value)
elif name == 'RC':
if in3_sel == In3Sel.RC.value:
return 3
- assert False, "RC does not exist yet"
elif name in ['EA', 'RS']:
if in1_sel == In1Sel.RS.value:
return 1
elif name == 'FRA':
if in1_sel == In1Sel.FRA.value:
return 1
+ if in3_sel == In3Sel.FRA.value:
+ return 3
elif name == 'FRB':
if in2_sel == In2Sel.FRB.value:
return 2
return 1
if in3_sel == In3Sel.FRS.value:
return 3
+ elif name == 'FRT':
+ if in1_sel == In1Sel.FRT.value:
+ return 1
+ elif name == 'RT':
+ if in1_sel == In1Sel.RT.value:
+ return 1
return None
if ewmode:
in1_base = yield dec2.e.read_reg1.base
in2_base = yield dec2.e.read_reg2.base
- in3_base = yield dec2.e.read_reg3.base
+ in3_base = yield dec2.e.read_reg3.base
in1_offs = yield dec2.e.read_reg1.offs
in2_offs = yield dec2.e.read_reg2.offs
- in3_offs = yield dec2.e.read_reg3.offs
+ in3_offs = yield dec2.e.read_reg3.offs
in1 = (in1, in1_base, in1_offs)
in2 = (in2, in2_base, in2_offs)
in3 = (in3, in3_base, in3_offs)
if name == 'CR1': # these are not actually calculated correctly
if out_sel == CROutSel.CR1.value:
return out, o_isvec
+ # check RC1 set? if so return implicit vector, this is a REAL bad hack
+ RC1 = yield dec2.rm_dec.RC1
+ if RC1:
+ log("get_cr_out RC1 mode")
+ if name == 'CR0':
+ return 0, True # XXX TODO: offset CR0 from SVSTATE SPR
+ if name == 'CR1':
+ return 1, True # XXX TODO: offset CR1 from SVSTATE SPR
+ # nope - not found.
log("get_cr_out not found", name)
return None, False
# TODO, really should just be using PowerDecoder2
-def get_idx_out(dec2, name, ewmode=False):
+def get_out_map(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
- if ewmode:
- offs = yield dec2.e.write_reg.offs
- base = yield dec2.e.write_reg.base
- out = (out, base, offs)
# identify which regnames map to out / o2
- if name == 'BF':
- log("get_idx_out", out_sel, out, o_isvec)
if name == 'RA':
- log("get_idx_out", out_sel, OutSel.RA.value, out, o_isvec)
if out_sel == OutSel.RA.value:
- return out, o_isvec
+ return True
elif name == 'RT':
- log("get_idx_out", out_sel, OutSel.RT.value,
- OutSel.RT_OR_ZERO.value, out, o_isvec,
- dec2.dec.RT)
if out_sel == OutSel.RT.value:
- return out, o_isvec
+ return True
if out_sel == OutSel.RT_OR_ZERO.value and out != 0:
- return out, o_isvec
+ return True
elif name == 'RT_OR_ZERO':
- log("get_idx_out", out_sel, OutSel.RT.value,
- OutSel.RT_OR_ZERO.value, out, o_isvec,
- dec2.dec.RT)
if out_sel == OutSel.RT_OR_ZERO.value:
- return out, o_isvec
+ return True
elif name == 'FRA':
- log("get_idx_out", out_sel, OutSel.FRA.value, out, o_isvec)
if out_sel == OutSel.FRA.value:
- return out, o_isvec
+ return True
+ elif name == 'FRS':
+ if out_sel == OutSel.FRS.value:
+ return True
elif name == 'FRT':
- log("get_idx_out", out_sel, OutSel.FRT.value,
- OutSel.FRT.value, out, o_isvec)
if out_sel == OutSel.FRT.value:
- return out, o_isvec
+ return True
+ return False
+
+
+# TODO, really should just be using PowerDecoder2
+def get_idx_out(dec2, name, ewmode=False):
+ 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
+ if ewmode:
+ offs = yield dec2.e.write_reg.offs
+ base = yield dec2.e.write_reg.base
+ out = (out, base, offs)
+ # identify which regnames map to out / o2
+ ismap = yield from get_out_map(dec2, name)
+ if ismap:
+ log("get_idx_out", name, out_sel, out, o_isvec)
+ return out, o_isvec
log("get_idx_out not found", name, out_sel, out, o_isvec)
return None, False
# TODO, really should just be using PowerDecoder2
-def get_idx_out2(dec2, name, ewmode=False):
+def get_out2_map(dec2, name):
# check first if register is activated for write
op = dec2.dec.op
out_sel = yield op.out_sel
out = yield dec2.e.write_ea.data
- if ewmode:
- offs = yield dec2.e.write_ea.offs
- base = yield dec2.e.write_ea.base
- out = (out, base, offs)
- o_isvec = yield dec2.o2_isvec
out_ok = yield dec2.e.write_ea.ok
- log("get_idx_out2", name, out_sel, out, out_ok, o_isvec)
if not out_ok:
- return None, False
+ return False
- if name == 'RA':
+ if name in ['EA', 'RA']:
if hasattr(op, "upd"):
# update mode LD/ST uses read-reg A also as an output
upd = yield op.upd
log("get_idx_out2", upd, LDSTMode.update.value,
out_sel, OutSel.RA.value,
- out, o_isvec)
+ out)
if upd == LDSTMode.update.value:
- return out, o_isvec
+ return True
if name == 'RS':
fft_en = yield dec2.implicit_rs
if fft_en:
log("get_idx_out2", out_sel, OutSel.RS.value,
- out, o_isvec)
- return out, o_isvec
+ out)
+ return True
if name == 'FRS':
fft_en = yield dec2.implicit_rs
if fft_en:
log("get_idx_out2", out_sel, OutSel.FRS.value,
- out, o_isvec)
- return out, o_isvec
+ out)
+ return True
+ return False
+
+
+# TODO, really should just be using PowerDecoder2
+def get_idx_out2(dec2, name, ewmode=False):
+ # check first if register is activated for write
+ op = dec2.dec.op
+ out_sel = yield op.out_sel
+ out = yield dec2.e.write_ea.data
+ if ewmode:
+ offs = yield dec2.e.write_ea.offs
+ base = yield dec2.e.write_ea.base
+ out = (out, base, offs)
+ o_isvec = yield dec2.o2_isvec
+ ismap = yield from get_out2_map(dec2, name)
+ if ismap:
+ log("get_idx_out2", name, out_sel, out, o_isvec)
+ return out, o_isvec
return None, False
if self.svstate.ssubstep == subvl: # end-point
log(" advance pack stop")
return
- break # exit inner loop
+ break # exit inner loop
self.svstate.srcstep += SelectableInt(1, 7) # advance ss
subvl = self.subvl
if self.svstate.ssubstep == subvl: # end-point
yield (self.svstate.ssubstep, srcstep)
if self.svstate.ssubstep == subvl: # end-point
self.svstate.ssubstep = SelectableInt(0, 2) # reset
- break # exit inner loop
+ break # exit inner loop
self.svstate.ssubstep += SelectableInt(1, 2)
vl = self.svstate.vl
if srcstep == vl-1: # end-point
self.svstate.srcstep = SelectableInt(srcstep, 7)
log(" advance src", self.svstate.srcstep, self.svstate.ssubstep,
- self.loopend)
+ self.loopend)
def dst_iterate(self):
"""dest step iterator
self.svstate.dststep = SelectableInt(dststep, 7)
log(" advance dst", self.svstate.dststep, self.svstate.dsubstep,
- self.loopend)
+ self.loopend)
def at_loopend(self):
"""tells if this is the last possible element. uses the cached values
TODO when Pack/Unpack is set, substep becomes the *outer* loop
"""
self.subvl = yield self.dec2.rm_dec.rm_in.subvl
- if self.loopend: # huhn??
+ if self.loopend: # huhn??
return
self.src_iterate()
self.dst_iterate()
pred_sz = yield self.dec2.rm_dec.pred_sz
if pmode == SVP64PredMode.INT.value:
srcmask = dstmask = get_predint(self.gpr, dstpred)
- if sv_ptype == SVPtype.P2.value:
+ if sv_ptype == SVPType.P2.value:
srcmask = get_predint(self.gpr, srcpred)
elif pmode == SVP64PredMode.CR.value:
srcmask = dstmask = get_predcr(self.crl, dstpred, vl)
- if sv_ptype == SVPtype.P2.value:
+ if sv_ptype == SVPType.P2.value:
srcmask = get_predcr(self.crl, srcpred, vl)
# work out if the ssubsteps are completed
ssubstart = ssubstep == 0
initial_pc=0,
bigendian=False,
mmu=False,
- icachemmu=False):
+ icachemmu=False,
+ initial_fpscr=0):
self.bigendian = bigendian
self.halted = False
self.last_op_svshape = False
# "raw" memory
- self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
+ self.mem = Mem(row_bytes=8, initial_mem=initial_mem, misaligned_ok=True)
self.mem.log_fancy(kind=LogKind.InstrInOuts)
self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
# MMU mode, redirect underlying Mem through RADIX
# FPR (same as GPR except for FP nums)
# 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
# note that mffs, mcrfs, mtfsf "manage" this FPSCR
+ self.fpscr = FPSCRState(initial_fpscr)
+
# 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
# note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
# -- Done
# create CR then allow portions of it to be "selectable" (below)
self.cr_fields = CRFields(initial_cr)
self.cr = self.cr_fields.cr
- self.cr_backup = 0 # sigh, dreadful hack: for fail-first (VLi)
+ self.cr_backup = 0 # sigh, dreadful hack: for fail-first (VLi)
# "undefined", just set to variable-bit-width int (use exts "max")
- # self.undefined = SelectableInt(0, 256) # TODO, not hard-code 256!
+ # self.undefined = SelectableInt(0, EFFECTIVELY_UNLIMITED)
self.namespace = {}
self.namespace.update(self.spr)
'SVSHAPE3': self.spr['SVSHAPE3'],
'CR': self.cr,
'MSR': self.msr,
+ 'FPSCR': self.fpscr,
'undefined': undefined,
'mode_is_64bit': True,
'SO': XER_bits['SO'],
'XLEN': 64 # elwidth overrides
})
+ for name in BFP_FLAG_NAMES:
+ setattr(self, name, 0)
+
# update pc to requested start point
self.set_pc(initial_pc)
self.decoder = decoder2.dec
self.dec2 = decoder2
- super().__init__(XLEN=self.namespace["XLEN"])
+ super().__init__(XLEN=self.namespace["XLEN"], FPSCR=self.fpscr)
@property
def XLEN(self):
return self.namespace["XLEN"]
+ @property
+ def FPSCR(self):
+ return self.fpscr
+
def call_trap(self, trap_addr, trap_bit):
"""calls TRAP and sets up NIA to the new execution location.
next instruction will begin at trap_addr.
self.namespace['XER'] = self.spr['XER']
self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
+ self.namespace['OV'] = self.spr['XER'][XER_bits['OV']].value
+ self.namespace['OV32'] = self.spr['XER'][XER_bits['OV32']].value
self.namespace['XLEN'] = xlen
# add some SVSTATE convenience variables
if self.is_svp64_mode and insn_name.startswith("sv.bc"):
# blegh grab bits manually
mode = yield self.dec2.rm_dec.rm_in.mode
- mode = SelectableInt(mode, 5) # convert to SelectableInt before test
+ # convert to SelectableInt before test
+ mode = SelectableInt(mode, 5)
bc_vlset = mode[SVP64MODEb.BC_VLSET] != 0
bc_vli = mode[SVP64MODEb.BC_VLI] != 0
bc_snz = mode[SVP64MODEb.BC_SNZ] != 0
self.namespace['sz'] = SelectableInt(sz, 1)
self.namespace['SNZ'] = SelectableInt(bc_snz, 1)
- def handle_carry_(self, inputs, output, ca, ca32):
+ def get_kludged_op_add_ca_ov(self, inputs, inp_ca_ov):
+ """ this was not at all necessary to do. this function massively
+ duplicates - in a laborious and complex fashion - the contents of
+ the CSV files that were extracted two years ago from microwatt's
+ source code. A-inversion is the "inv A" column, output inversion
+ is the "inv out" column, carry-in equal to 0 or 1 or CA is the
+ "cry in" column
+
+ all of that information is available in
+ self.instrs[ins_name].op_fields
+ where info is usually assigned to self.instrs[ins_name]
+
+ https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=openpower/isatables/minor_31.csv;hb=HEAD
+
+ the immediate constants are *also* decoded correctly and placed
+ usually by DecodeIn2Imm into operand2, as part of power_decoder2.py
+ """
+ def ca(a, b, ca_in, width):
+ mask = (1 << width) - 1
+ y = (a & mask) + (b & mask) + ca_in
+ return y >> width
+
+ asmcode = yield self.dec2.dec.op.asmcode
+ insn = insns.get(asmcode)
+ SI = yield self.dec2.dec.SI
+ SI &= 0xFFFF
+ CA, OV = inp_ca_ov
+ inputs = [i.value for i in inputs]
+ if SI & 0x8000:
+ SI -= 0x10000
+ if insn in ("add", "addo", "addc", "addco"):
+ a = inputs[0]
+ b = inputs[1]
+ ca_in = 0
+ elif insn == "addic" or insn == "addic.":
+ a = inputs[0]
+ b = SI
+ ca_in = 0
+ elif insn in ("subf", "subfo", "subfc", "subfco"):
+ a = ~inputs[0]
+ b = inputs[1]
+ ca_in = 1
+ elif insn == "subfic":
+ a = ~inputs[0]
+ b = SI
+ ca_in = 1
+ elif insn == "adde" or insn == "addeo":
+ a = inputs[0]
+ b = inputs[1]
+ ca_in = CA
+ elif insn == "subfe" or insn == "subfeo":
+ a = ~inputs[0]
+ b = inputs[1]
+ ca_in = CA
+ elif insn == "addme" or insn == "addmeo":
+ a = inputs[0]
+ b = ~0
+ ca_in = CA
+ elif insn == "addze" or insn == "addzeo":
+ a = inputs[0]
+ b = 0
+ ca_in = CA
+ elif insn == "subfme" or insn == "subfmeo":
+ a = ~inputs[0]
+ b = ~0
+ ca_in = CA
+ elif insn == "subfze" or insn == "subfzeo":
+ a = ~inputs[0]
+ b = 0
+ ca_in = CA
+ elif insn == "addex":
+ # CA[32] aren't actually written, just generate so we have
+ # something to return
+ ca64 = ov64 = ca(inputs[0], inputs[1], OV, 64)
+ ca32 = ov32 = ca(inputs[0], inputs[1], OV, 32)
+ return ca64, ca32, ov64, ov32
+ elif insn == "neg" or insn == "nego":
+ a = ~inputs[0]
+ b = 0
+ ca_in = 1
+ else:
+ raise NotImplementedError(
+ "op_add kludge unimplemented instruction: ", asmcode, insn)
+
+ ca64 = ca(a, b, ca_in, 64)
+ ca32 = ca(a, b, ca_in, 32)
+ ov64 = ca64 != ca(a, b, ca_in, 63)
+ ov32 = ca32 != ca(a, b, ca_in, 31)
+ return ca64, ca32, ov64, ov32
+
+ def handle_carry_(self, inputs, output, ca, ca32, inp_ca_ov):
+ op = yield self.dec2.e.do.insn_type
+ if op == MicrOp.OP_ADD.value and ca is None and ca32 is None:
+ retval = yield from self.get_kludged_op_add_ca_ov(
+ inputs, inp_ca_ov)
+ ca, ca32, ov, ov32 = retval
+ asmcode = yield self.dec2.dec.op.asmcode
+ if insns.get(asmcode) == 'addex':
+ # TODO: if 32-bit mode, set ov to ov32
+ self.spr['XER'][XER_bits['OV']] = ov
+ self.spr['XER'][XER_bits['OV32']] = ov32
+ else:
+ # TODO: if 32-bit mode, set ca to ca32
+ self.spr['XER'][XER_bits['CA']] = ca
+ self.spr['XER'][XER_bits['CA32']] = ca32
+ return
inv_a = yield self.dec2.e.do.invert_in
if inv_a:
inputs[0] = ~inputs[0]
log(gts)
cy = 1 if any(gts) else 0
log("CA", cy, gts)
- if ca is None: # already written
+ if ca is None: # already written
self.spr['XER'][XER_bits['CA']] = cy
# 32 bit carry
gts.append(gt)
cy32 = 1 if any(gts) else 0
log("CA32", cy32, gts)
- if ca32 is None: # already written
+ if ca32 is None: # already written
self.spr['XER'][XER_bits['CA32']] = cy32
- def handle_overflow(self, inputs, output, div_overflow):
+ def handle_overflow(self, inputs, output, div_overflow, inp_ca_ov):
+ op = yield self.dec2.e.do.insn_type
+ if op == MicrOp.OP_ADD.value:
+ retval = yield from self.get_kludged_op_add_ca_ov(
+ inputs, inp_ca_ov)
+ ca, ca32, ov, ov32 = retval
+ # TODO: if 32-bit mode, set ov to ov32
+ self.spr['XER'][XER_bits['OV']] = ov
+ self.spr['XER'][XER_bits['OV32']] = ov32
+ self.spr['XER'][XER_bits['SO']] |= ov
+ return
if hasattr(self.dec2.e.do, "invert_in"):
inv_a = yield self.dec2.e.do.invert_in
if inv_a:
SO = SelectableInt(1, 0)
else:
SO = self.spr['XER'][XER_bits['SO']]
- log("handle_comparison SO overflow", SO, overflow)
+ log("handle_comparison SO", SO.value,
+ "overflow", overflow,
+ "zero", zero.value,
+ "+ve", positive.value,
+ "-ve", negative.value)
# alternative overflow checking (setvl mainly at the moment)
if overflow is not None and overflow == 1:
SO = SelectableInt(1, 1)
opcode = yield self.dec2.dec.opcode_in
opcode = SelectableInt(value=opcode, bits=32)
pfx = SVP64Instruction.Prefix(opcode)
- log("prefix test: opcode:", pfx.po, bin(pfx.po), pfx.id)
- self.is_svp64_mode = bool((pfx.po == 0b000001) and (pfx.id == 0b11))
+ log("prefix test: opcode:", pfx.PO, bin(pfx.PO), pfx.id)
+ self.is_svp64_mode = bool((pfx.PO == 0b000001) and (pfx.id == 0b11))
self.pc.update_nia(self.is_svp64_mode)
# set SVP64 decode
yield self.dec2.is_svp64_mode.eq(self.is_svp64_mode)
def execute_one(self):
"""execute one instruction
"""
+ self.insnlog = [] # log the instruction
# get the disassembly code for this instruction
if not self.disassembly:
code = yield from self.get_assembly_name()
else:
offs, dbg = 0, ""
if self.is_svp64_mode:
- offs, dbg = 4, "svp64 "
+ offs, dbg = 4, "svp64 "
code = self.disassembly[self._pc+offs]
log(" %s sim-execute" % dbg, hex(self._pc), code)
+ self.insnlog.append(code)
opname = code.split(' ')[0]
try:
yield from self.call(opname) # execute the instruction
except MemException as e: # check for memory errors
if e.args[0] == 'unaligned': # alignment error
- # run a Trap but set DAR first
- print("memory unaligned exception, DAR", e.dar)
+ # run a Trap but set DAR first
+ print("memory unaligned exception, DAR", e.dar, repr(e))
self.spr['DAR'] = SelectableInt(e.dar, 64)
self.call_trap(0x600, PIb.PRIV) # 0x600, privileged
return
elif e.args[0] == 'invalid': # invalid
- # run a Trap but set DAR first
+ # run a Trap but set DAR first
log("RADIX MMU memory invalid error, mode %s" % e.mode)
if e.mode == 'EXECUTE':
# XXX TODO: must set a few bits in SRR1,
# not supported yet:
raise e # ... re-raise
+ # append the log file
+ with open("/tmp/insnlog.txt", "a+") as f:
+ f.write(" ".join(self.insnlog)+"\n")
+
log("gprs after code", code)
self.gpr.dump()
crs = []
# list of instructions not being supported by binutils (.long)
dotstrp = asmop[:-1] if asmop[-1] == '.' else asmop
if dotstrp in [*FPTRANS_INSNS,
+ *LDST_UPDATE_INSNS,
'ffmadds', 'fdmadds', 'ffadds',
- 'mins', 'maxs', 'minu', 'maxu',
+ 'minmax',
'setvl', 'svindex', 'svremap', 'svstep',
'svshape', 'svshape2',
'grev', 'ternlogi', 'bmask', 'cprop',
'absdu', 'absds', 'absdacs', 'absdacu', 'avgadd',
'fmvis', 'fishmv', 'pcdec', "maddedu", "divmod2du",
- "dsld", "dsrd",
+ "dsld", "dsrd", "maddedus",
+ "shadd", "shaddw", "shadduw",
+ "fcvttg", "fcvttgo", "fcvttgs", "fcvttgso",
+ "fmvtg", "fmvtgs",
+ "fcvtfg", "fcvtfgs",
+ "fmvfg", "fmvfgs",
+ "maddsubrs", "maddrs"
]:
illegal = False
ins_name = dotstrp
if self.is_svp64_mode:
ew_src = yield self.dec2.rm_dec.ew_src
ew_dst = yield self.dec2.rm_dec.ew_dst
- ew_src = 8 << (3-int(ew_src)) # convert to bitlength
- ew_dst = 8 << (3-int(ew_dst)) # convert to bitlength
+ ew_src = 8 << (3-int(ew_src)) # convert to bitlength
+ ew_dst = 8 << (3-int(ew_dst)) # convert to bitlength
xlen = max(ew_src, ew_dst)
log("elwdith", ew_src, ew_dst)
log("XLEN:", self.is_svp64_mode, xlen)
# look up instruction in ISA.instrs, prepare namespace
- if ins_name == 'pcdec': # grrrr yes there are others ("stbcx." etc.)
+ if ins_name == 'pcdec': # grrrr yes there are others ("stbcx." etc.)
info = self.instrs[ins_name+"."]
+ elif asmop[-1] == '.' and asmop in self.instrs:
+ info = self.instrs[asmop]
else:
info = self.instrs[ins_name]
yield from self.prep_namespace(ins_name, info.form, info.op_fields,
end_loop = no_in_vec or srcstep == vl-1 or dststep == vl-1
self.namespace['end_loop'] = SelectableInt(end_loop, 1)
+ inp_ca_ov = (self.spr['XER'][XER_bits['CA']].value,
+ self.spr['XER'][XER_bits['OV']].value)
+
# execute actual instruction here (finally)
log("inputs", inputs)
results = info.func(self, *inputs)
# detect if CA/CA32 already in outputs (sra*, basically)
ca = outs.get("CA")
- ca32 = outs.get("CA32 ")
+ ca32 = outs.get("CA32")
log("carry already done?", ca, ca32, output_names)
carry_en = yield self.dec2.e.do.output_carry
if carry_en:
- yield from self.handle_carry_(inputs, results[0], ca, ca32)
+ yield from self.handle_carry_(
+ inputs, results[0], ca, ca32, inp_ca_ov=inp_ca_ov)
# get outout named "overflow" and "CR0"
overflow = outs.get('overflow')
ov_ok = yield self.dec2.e.do.oe.ok
log("internal overflow", ins_name, overflow, "en?", ov_en, ov_ok)
if ov_en & ov_ok:
- yield from self.handle_overflow(inputs, results[0], overflow)
+ yield from self.handle_overflow(
+ inputs, results[0], overflow, inp_ca_ov=inp_ca_ov)
# only do SVP64 dest predicated Rc=1 if dest-pred is not enabled
rc_en = False
ff_inv = yield self.dec2.rm_dec.inv
cr_bit = yield self.dec2.rm_dec.cr_sel
RC1 = yield self.dec2.rm_dec.RC1
- vli_ = yield self.dec2.rm_dec.vli # VL inclusive if truncated
+ vli_ = yield self.dec2.rm_dec.vli # VL inclusive if truncated
log(" ff rm_mode", rc_en, rm_mode, SVP64RMMode.FFIRST.value)
log(" inv", ff_inv)
log(" RC1", RC1)
is_setvl = ins_name in ('svstep', 'setvl')
if is_setvl:
result = SelectableInt(result.vl, 64)
- else:
- overflow = None # do not override overflow except in setvl
+ #else:
+ # overflow = None # do not override overflow except in setvl
# if there was not an explicit CR0 in the pseudocode, do implicit Rc=1
if cr0 is None:
self.crl[regnum].eq(cr0)
def do_outregs_nia(self, asmop, ins_name, info, outs,
- carry_en, rc_en, ffirst_hit, ew_dst):
+ ca_en, rc_en, ffirst_hit, ew_dst):
ffirst_hit, vli = ffirst_hit
- # write out any regs for this instruction
- for name, output in outs.items():
- yield from self.check_write(info, name, output, carry_en, ew_dst)
+ # write out any regs for this instruction, but only if fail-first is ok
+ # XXX TODO: allow CR-vector to be written out even if ffirst fails
+ if not ffirst_hit or vli:
+ for name, output in outs.items():
+ yield from self.check_write(info, name, output, ca_en, ew_dst)
# restore the CR value on non-VLI failfirst (from sv.cmp and others
# which write directly to CR in the pseudocode (gah, what a mess)
- #if ffirst_hit and not vli:
+ # if ffirst_hit and not vli:
# self.cr.value = self.cr_backup
if ffirst_hit:
else:
# check advancement of src/dst/sub-steps and if PC needs updating
nia_update = (yield from self.check_step_increment(rc_en,
- asmop, ins_name))
+ asmop, ins_name))
if nia_update:
self.update_pc_next()
if op == MicrOp.OP_LOAD.value:
if remap_active:
offsmul = yield self.dec2.in1_step
- log("D-field REMAP src", imm, offsmul)
+ log("D-field REMAP src", imm, offsmul, ldstmode)
else:
offsmul = (srcstep * (subvl+1)) + ssubstep
- log("D-field src", imm, offsmul)
+ log("D-field src", imm, offsmul, ldstmode)
elif op == MicrOp.OP_STORE.value:
# XXX NOTE! no bit-reversed STORE! this should not ever be used
offsmul = (dststep * (subvl+1)) + dsubstep
- log("D-field dst", imm, offsmul)
+ log("D-field dst", imm, offsmul, ldstmode)
# Unit-Strided LD/ST adds offset*width to immediate
if ldstmode == SVP64LDSTmode.UNITSTRIDE.value:
ldst_len = yield self.dec2.e.do.data_len
if isinstance(regnum, tuple):
(regnum, base, offs) = regnum
else:
- base, offs = regnum, 0 # temporary HACK
+ base, offs = regnum, 0 # temporary HACK
# in case getting the register number is needed, _RA, _RB
# (HACK: only in straight non-svp64-mode for now, or elwidth == 64)
if name in fregs:
reg_val = SelectableInt(self.fpr(base, is_vec, offs, ew_src))
log("read reg %d/%d: 0x%x" % (base, offs, reg_val.value))
+ self.insnlog.append("rFPR:%d.%d/%d" % (base, offs, ew_src))
elif name is not None:
reg_val = SelectableInt(self.gpr(base, is_vec, offs, ew_src))
+ self.insnlog.append("rGPR:%d.%d/%d" % (base, offs, ew_src))
log("read reg %d/%d: 0x%x" % (base, offs, reg_val.value))
else:
log('zero input reg %s %s' % (name, str(regnum)), is_vec)
mi2 = self.svstate.mi2
mo0 = self.svstate.mo0
mo1 = self.svstate.mo1
- steps = [(self.dec2.in1_step, mi0), # RA
- (self.dec2.in2_step, mi1), # RB
- (self.dec2.in3_step, mi2), # RC
- (self.dec2.o_step, mo0), # RT
- (self.dec2.o2_step, mo1), # EA
+ steps = [[self.dec2.in1_step, mi0], # RA
+ [self.dec2.in2_step, mi1], # RB
+ [self.dec2.in3_step, mi2], # RC
+ [self.dec2.o_step, mo0], # RT
+ [self.dec2.o2_step, mo1], # EA
]
+ if False: # TODO
+ rnames = ['RA', 'RB', 'RC', 'RT', 'RS']
+ for i, reg in enumerate(rnames):
+ idx = yield from get_idx_map(self.dec2, reg)
+ if idx is None:
+ idx = yield from get_idx_map(self.dec2, "F"+reg)
+ if idx == 1: # RA
+ steps[i][0] = self.dec2.in1_step
+ elif idx == 2: # RB
+ steps[i][0] = self.dec2.in2_step
+ elif idx == 3: # RC
+ steps[i][0] = self.dec2.in3_step
+ log("remap step", i, reg, idx, steps[i][1])
remap_idxs = self.remap_idxs
rremaps = []
# now cross-index the required SHAPE for each of 3-in 2-out regs
if shape.value == 0x0:
continue
# now set the actual requested step to the current index
- yield dstep.eq(remap_idx)
+ if dstep is not None:
+ yield dstep.eq(remap_idx)
# debug printout info
- rremaps.append((shape.mode, hex(shape.value),
+ rremaps.append((shape.mode, hex(shape.value), dstep,
i, rnames[i], shape_idx, remap_idx))
for x in rremaps:
log("shape remap", x)
if name == 'CR0': # ignore, done already (above)
return
if isinstance(output, int):
- output = SelectableInt(output, 256)
+ output = SelectableInt(output, EFFECTIVELY_UNLIMITED)
# write carry flafs
if name in ['CA', 'CA32']:
if carry_en:
# check zeroing due to predicate bit being zero
if self.is_svp64_mode and self.pred_dst_zero:
log('zeroing reg %s %s' % (str(regnum), str(output)), is_vec)
- output = SelectableInt(0, 256)
+ output = SelectableInt(0, EFFECTIVELY_UNLIMITED)
log("write reg %s%s 0x%x ew %d" % (reg_prefix, str(regnum),
output.value, ew_dst),
kind=LogKind.InstrInOuts)
# zero-extend tov64 bit begore storing (should use EXT oh well)
if output.bits > 64:
output = SelectableInt(output.value, 64)
+ rnum, base, offset = regnum
if name in fregs:
self.fpr.write(regnum, output, is_vec, ew_dst)
+ self.insnlog.append("wFPR:%d.%d/%d" % (rnum, offset, ew_dst))
else:
self.gpr.write(regnum, output, is_vec, ew_dst)
+ self.insnlog.append("wGPR:%d.%d/%d" % (rnum, offset, ew_dst))
def check_step_increment(self, rc_en, asmop, ins_name):
# check if it is the SVSTATE.src/dest step that needs incrementing
log(" reverse", reverse_gear)
log(" out_vec", out_vec)
log(" in_vec", in_vec)
- log(" sv_ptype", sv_ptype, sv_ptype == SVPtype.P2.value)
+ log(" sv_ptype", sv_ptype, sv_ptype == SVPType.P2.value)
# check if this was an sv.bc* and if so did it succeed
if self.is_svp64_mode and insn_name.startswith("sv.bc"):
end_loop = self.namespace['end_loop']
# check if srcstep needs incrementing by one, stop PC advancing
# but for 2-pred both src/dest have to be checked.
# XXX this might not be true! it may just be LD/ST
- if sv_ptype == SVPtype.P2.value:
+ if sv_ptype == SVPType.P2.value:
svp64_is_vector = (out_vec or in_vec)
else:
svp64_is_vector = out_vec