from nmigen.lib.coding import PriorityEncoder
-from soc.decoder.power_decoder import create_pdecode
-from soc.decoder.power_decoder2 import PowerDecode2, SVP64PrefixDecoder
-from soc.decoder.decode2execute1 import IssuerDecode2ToOperand
-from soc.decoder.decode2execute1 import Data
+from openpower.decoder.power_decoder import create_pdecode
+from openpower.decoder.power_decoder2 import PowerDecode2, SVP64PrefixDecoder
+from openpower.decoder.decode2execute1 import IssuerDecode2ToOperand
+from openpower.decoder.decode2execute1 import Data
+from openpower.decoder.power_enums import (MicrOp, SVP64PredInt, SVP64PredCR,
+ SVP64PredMode)
+from openpower.state import CoreState
+from openpower.consts import (CR, SVP64CROffs)
from soc.experiment.testmem import TestMemory # test only for instructions
from soc.regfile.regfiles import StateRegs, FastRegs
from soc.simple.core import NonProductionCore
from soc.config.test.test_loadstore import TestMemPspec
from soc.config.ifetch import ConfigFetchUnit
-from soc.decoder.power_enums import (MicrOp, SVP64PredInt, SVP64PredCR,
- SVP64PredMode)
from soc.debug.dmi import CoreDebug, DMIInterface
from soc.debug.jtag import JTAG
from soc.config.pinouts import get_pinspecs
-from soc.config.state import CoreState
from soc.interrupts.xics import XICS_ICP, XICS_ICS
from soc.bus.simple_gpio import SimpleGPIO
from soc.bus.SPBlock512W64B8W import SPBlock512W64B8W
from soc.clock.select import ClockSelect
from soc.clock.dummypll import DummyPLL
-from soc.sv.svstate import SVSTATERec
+from openpower.sv.svstate import SVSTATERec
from nmutil.util import rising_edge
invert = Signal(name=name+"crinvert")
with m.Switch(mask):
with m.Case(SVP64PredCR.LT.value):
- comb += idx.eq(0)
- comb += invert.eq(1)
- with m.Case(SVP64PredCR.GE.value):
- comb += idx.eq(0)
+ comb += idx.eq(CR.LT)
comb += invert.eq(0)
- with m.Case(SVP64PredCR.GT.value):
- comb += idx.eq(1)
+ with m.Case(SVP64PredCR.GE.value):
+ comb += idx.eq(CR.LT)
comb += invert.eq(1)
- with m.Case(SVP64PredCR.LE.value):
- comb += idx.eq(1)
+ with m.Case(SVP64PredCR.GT.value):
+ comb += idx.eq(CR.GT)
comb += invert.eq(0)
- with m.Case(SVP64PredCR.EQ.value):
- comb += idx.eq(2)
+ with m.Case(SVP64PredCR.LE.value):
+ comb += idx.eq(CR.GT)
comb += invert.eq(1)
- with m.Case(SVP64PredCR.NE.value):
- comb += idx.eq(1)
+ with m.Case(SVP64PredCR.EQ.value):
+ comb += idx.eq(CR.EQ)
comb += invert.eq(0)
- with m.Case(SVP64PredCR.SO.value):
- comb += idx.eq(3)
+ with m.Case(SVP64PredCR.NE.value):
+ comb += idx.eq(CR.EQ)
comb += invert.eq(1)
- with m.Case(SVP64PredCR.NS.value):
- comb += idx.eq(3)
+ with m.Case(SVP64PredCR.SO.value):
+ comb += idx.eq(CR.SO)
comb += invert.eq(0)
+ with m.Case(SVP64PredCR.NS.value):
+ comb += idx.eq(CR.SO)
+ comb += invert.eq(1)
return idx, invert
be done through multiple reads, extracting one relevant at a time.
later, a faster way would be to use the 32-bit-wide CR port but
this is more complex decoding, here. equivalent code used in
- ISACaller is "from soc.decoder.isa.caller import get_predcr"
+ ISACaller is "from openpower.decoder.isa.caller import get_predcr"
note: this ENTIRE FSM is not to be called when svp64 is disabled
"""
cur_state = self.cur_state
srcstep = cur_state.svstate.srcstep
dststep = cur_state.svstate.dststep
-
- # elif predmode == CR:
- # CR-src sidx, sinvert = get_predcr(m, srcpred)
- # CR-dst didx, dinvert = get_predcr(m, dstpred)
- # TODO read CR-src and CR-dst into self.srcmask+dstmask with loop
- # has to cope with first one then the other
- # for cr_idx = FSM-state-loop(0..VL-1):
- # FSM-state-trigger-CR-read:
- # cr_ren = (1<<7-(cr_idx+SVP64CROffs.CRPred))
- # comb += cr_pred.ren.eq(cr_ren)
- # FSM-state-1-clock-later-actual-Read:
- # cr_field = Signal(4)
- # cr_bit = Signal(1)
- # # read the CR field, select the appropriate bit
- # comb += cr_field.eq(cr_pred.data_o)
- # comb += cr_bit.eq(cr_field.bit_select(idx)))
- # # just like in branch BO tests
- # comd += self.srcmask[cr_idx].eq(inv ^ cr_bit)
+ cur_vl = cur_state.svstate.vl
# decode predicates
sregread, sinvert, sunary, sall1s = get_predint(m, srcpred, 's')
sidx, scrinvert = get_predcr(m, srcpred, 's')
didx, dcrinvert = get_predcr(m, dstpred, 'd')
+ # store fetched masks, for either intpred or crpred
+ # when src/dst step is not zero, the skipped mask bits need to be
+ # shifted-out, before actually storing them in src/dest mask
+ new_srcmask = Signal(64, reset_less=True)
+ new_dstmask = Signal(64, reset_less=True)
+
with m.FSM(name="fetch_predicate"):
with m.State("FETCH_PRED_IDLE"):
with m.If(predmode == SVP64PredMode.INT):
# skip fetching destination mask register, when zero
with m.If(dall1s):
- sync += self.dstmask.eq(-1)
+ sync += new_dstmask.eq(-1)
# directly go to fetch source mask register
# guaranteed not to be zero (otherwise predmode
# would be SVP64PredMode.ALWAYS, not INT)
comb += int_pred.addr.eq(dregread)
comb += int_pred.ren.eq(1)
m.next = "INT_DST_READ"
+ with m.Elif(predmode == SVP64PredMode.CR):
+ # go fetch masks from the CR register file
+ sync += new_srcmask.eq(0)
+ sync += new_dstmask.eq(0)
+ m.next = "CR_READ"
with m.Else():
sync += self.srcmask.eq(-1)
sync += self.dstmask.eq(-1)
with m.State("INT_DST_READ"):
# store destination mask
inv = Repl(dinvert, 64)
- new_dstmask = Signal(64)
with m.If(dunary):
# set selected mask bit for 1<<r3 mode
dst_shift = Signal(range(64))
comb += dst_shift.eq(self.int_pred.data_o & 0b111111)
- comb += new_dstmask.eq(1 << dst_shift)
+ sync += new_dstmask.eq(1 << dst_shift)
with m.Else():
# invert mask if requested
- comb += new_dstmask.eq(self.int_pred.data_o ^ inv)
- # shift-out already used mask bits
- sync += self.dstmask.eq(new_dstmask >> dststep)
+ sync += new_dstmask.eq(self.int_pred.data_o ^ inv)
# skip fetching source mask register, when zero
with m.If(sall1s):
- sync += self.srcmask.eq(-1)
- m.next = "FETCH_PRED_DONE"
+ sync += new_srcmask.eq(-1)
+ m.next = "FETCH_PRED_SHIFT_MASK"
# fetch source predicate register
with m.Else():
comb += int_pred.addr.eq(sregread)
with m.State("INT_SRC_READ"):
# store source mask
inv = Repl(sinvert, 64)
- new_srcmask = Signal(64)
with m.If(sunary):
# set selected mask bit for 1<<r3 mode
src_shift = Signal(range(64))
comb += src_shift.eq(self.int_pred.data_o & 0b111111)
- comb += new_srcmask.eq(1 << src_shift)
+ sync += new_srcmask.eq(1 << src_shift)
with m.Else():
# invert mask if requested
- comb += new_srcmask.eq(self.int_pred.data_o ^ inv)
- # shift-out already used mask bits
+ sync += new_srcmask.eq(self.int_pred.data_o ^ inv)
+ m.next = "FETCH_PRED_SHIFT_MASK"
+
+ # fetch masks from the CR register file
+ # implements the following loop:
+ # idx, inv = get_predcr(mask)
+ # mask = 0
+ # for cr_idx in range(vl):
+ # cr = crl[cr_idx + SVP64CROffs.CRPred] # takes one cycle
+ # if cr[idx] ^ inv:
+ # mask |= 1 << cr_idx
+ # return mask
+ with m.State("CR_READ"):
+ # CR index to be read, which will be ready by the next cycle
+ cr_idx = Signal.like(cur_vl, reset_less=True)
+ # submit the read operation to the regfile
+ with m.If(cr_idx != cur_vl):
+ # the CR read port is unary ...
+ # ren = 1 << cr_idx
+ # ... in MSB0 convention ...
+ # ren = 1 << (7 - cr_idx)
+ # ... and with an offset:
+ # ren = 1 << (7 - off - cr_idx)
+ idx = SVP64CROffs.CRPred + cr_idx
+ comb += cr_pred.ren.eq(1 << (7 - idx))
+ # signal data valid in the next cycle
+ cr_read = Signal(reset_less=True)
+ sync += cr_read.eq(1)
+ # load the next index
+ sync += cr_idx.eq(cr_idx + 1)
+ with m.Else():
+ # exit on loop end
+ sync += cr_read.eq(0)
+ sync += cr_idx.eq(0)
+ m.next = "FETCH_PRED_SHIFT_MASK"
+ with m.If(cr_read):
+ # compensate for the one cycle delay on the regfile
+ cur_cr_idx = Signal.like(cur_vl)
+ comb += cur_cr_idx.eq(cr_idx - 1)
+ # read the CR field, select the appropriate bit
+ cr_field = Signal(4)
+ scr_bit = Signal()
+ dcr_bit = Signal()
+ comb += cr_field.eq(cr_pred.data_o)
+ comb += scr_bit.eq(cr_field.bit_select(sidx, 1) ^ scrinvert)
+ comb += dcr_bit.eq(cr_field.bit_select(didx, 1) ^ dcrinvert)
+ # set the corresponding mask bit
+ bit_to_set = Signal.like(self.srcmask)
+ comb += bit_to_set.eq(1 << cur_cr_idx)
+ with m.If(scr_bit):
+ sync += new_srcmask.eq(new_srcmask | bit_to_set)
+ with m.If(dcr_bit):
+ sync += new_dstmask.eq(new_dstmask | bit_to_set)
+
+ with m.State("FETCH_PRED_SHIFT_MASK"):
+ # shift-out skipped mask bits
sync += self.srcmask.eq(new_srcmask >> srcstep)
+ sync += self.dstmask.eq(new_dstmask >> dststep)
m.next = "FETCH_PRED_DONE"
with m.State("FETCH_PRED_DONE"):
ports += list(self.dbg.dmi.ports())
ports += list(self.imem.ibus.fields.values())
- ports += list(self.core.l0.cmpi.lsmem.lsi.slavebus.fields.values())
+ ports += list(self.core.l0.cmpi.wb_bus().fields.values())
if self.sram4x4k:
for sram in self.sram4k:
self.pll_en = hasattr(pspec, "use_pll") and pspec.use_pll
if self.pll_en:
self.pll_18_o = Signal(reset_less=True)
+ self.clk_sel_i = Signal(reset_less=True)
def elaborate(self, platform):
m = Module()
# output 18 mhz PLL test signal
comb += self.pll_18_o.eq(pll.pll_18_o)
+ # input to pll clock selection
+ comb += Cat(pll.sel_a0_i, pll.sel_a1_i).eq(self.clk_sel_i)
+
# now wire up ResetSignals. don't mind them being in this domain
pll_rst = ResetSignal("pllclk")
comb += pll_rst.eq(ResetSignal())
ports.append(ClockSignal())
ports.append(ResetSignal())
if self.pll_en:
- ports.append(self.pll.clk_sel_i)
+ ports.append(self.clk_sel_i)
ports.append(self.pll_18_o)
- ports.append(self.pll.pll_lck_o)
+ ports.append(self.pll.pll_ana_o)
return ports