"""
from nmigen import (Elaboratable, Module, Signal, ClockSignal, ResetSignal,
- ClockDomain, DomainRenamer, Mux, Const)
+ ClockDomain, DomainRenamer, Mux, Const, Repl)
from nmigen.cli import rtlil
from nmigen.cli import main
import sys
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
+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
comb += res.eq(regfile.data_o)
return res
-def get_predint(m, mask):
+def get_predint(m, mask, name):
"""decode SVP64 predicate integer mask field to reg number and invert
this is identical to the equivalent function in ISACaller except that
it doesn't read the INT directly, it just decodes "what needs to be done"
i.e. which INT reg, whether it is shifted and whether it is bit-inverted.
+
+ * all1s is set to indicate that no mask is to be applied.
+ * regread indicates the GPR register number to be read
+ * invert is set to indicate that the register value is to be inverted
+ * unary indicates that the contents of the register is to be shifted 1<<r3
"""
- regread = Signal(5)
- invert = Signal()
- unary = Signal()
+ comb = m.d.comb
+ regread = Signal(5, name=name+"regread")
+ invert = Signal(name=name+"invert")
+ unary = Signal(name=name+"unary")
+ all1s = Signal(name=name+"all1s")
with m.Switch(mask):
with m.Case(SVP64PredInt.ALWAYS.value):
- comb += regread.eq(0)
- comb += invert.eq(1)
+ comb += all1s.eq(1) # use 0b1111 (all ones)
with m.Case(SVP64PredInt.R3_UNARY.value):
comb += regread.eq(3)
- comb += unary.eq(1)
+ comb += unary.eq(1) # 1<<r3 - shift r3 (single bit)
with m.Case(SVP64PredInt.R3.value):
comb += regread.eq(3)
with m.Case(SVP64PredInt.R3_N.value):
with m.Case(SVP64PredInt.R30_N.value):
comb += regread.eq(30)
comb += invert.eq(1)
- return regread, invert, unary
+ return regread, invert, unary, all1s
-def get_predcr(m, mask):
+def get_predcr(m, mask, name):
"""decode SVP64 predicate CR to reg number field and invert status
this is identical to _get_predcr in ISACaller
"""
- idx = Signal(2)
- invert = Signal()
+ comb = m.d.comb
+ idx = Signal(2, name=name+"idx")
+ invert = Signal(name=name+"crinvert")
with m.Switch(mask):
with m.Case(SVP64PredCR.LT.value):
comb += idx.eq(0)
class TestIssuerInternal(Elaboratable):
"""TestIssuer - reads instructions from TestMemory and issues them
- efficiency and speed is not the main goal here: functional correctness is.
+ efficiency and speed is not the main goal here: functional correctness
+ and code clarity is. optimisations (which almost 100% interfere with
+ easy understanding) come later.
"""
def __init__(self, pspec):
fetch_pc_ready_o, fetch_pc_valid_i,
fetch_insn_valid_o, fetch_insn_ready_i):
"""fetch FSM
+
this FSM performs fetch of raw instruction data, partial-decodes
it 32-bit at a time to detect SVP64 prefixes, and will optionally
read a 2nd 32-bit quantity if that occurs.
sync += dec_opcode_i.eq(insn)
m.next = "INSN_READY"
# TODO: probably can start looking at pdecode2.rm_dec
- # here (or maybe even in INSN_READ state, if svp64_mode
+ # here or maybe even in INSN_READ state, if svp64_mode
# detected, in order to trigger - and wait for - the
# predicate reading.
- pmode = pdecode2.rm_dec.predmode
- sv_ptype = pdecode2.dec.op.SV_Ptype
- srcpred = pdecode2.rm_dec.srcpred
- dstpred = pdecode2.rm_dec.dstpred
+ if self.svp64_en:
+ pmode = pdecode2.rm_dec.predmode
+ """
+ if pmode != SVP64PredMode.ALWAYS.value:
+ fire predicate loading FSM and wait before
+ moving to INSN_READY
+ else:
+ sync += self.srcmask.eq(-1) # set to all 1s
+ sync += self.dstmask.eq(-1) # set to all 1s
+ m.next = "INSN_READY"
+ """
with m.State("INSN_READY"):
# hand over the instruction, to be decoded
with m.If(fetch_insn_ready_i):
m.next = "IDLE"
- def fetch_predicate_fsm(self, m, core, TODO):
+ def fetch_predicate_fsm(self, m,
+ pred_insn_valid_i, pred_insn_ready_o,
+ pred_mask_valid_o, pred_mask_ready_i):
"""fetch_predicate_fsm - obtains (constructs in the case of CR)
src/dest predicate masks
predmode = rm_dec.predmode
srcpred, dstpred = rm_dec.srcpred, rm_dec.dstpred
cr_pred, int_pred = self.cr_pred, self.int_pred # read regfiles
- # if predmode == INT:
- # INT-src sregread, sinvert, sunary = get_predint(m, srcpred)
- # INT-dst dregread, dinvert, dunary = get_predint(m, dstpred)
- # TODO read INT-src and INT-dst into self.srcmask+dstmask
+
# 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
- # else
- # sync += self.srcmask.eq(-1) # set to all 1s
- # sync += self.dstmask.eq(-1) # set to all 1s
+ # 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)
+
+ # decode predicates
+ sregread, sinvert, sunary, sall1s = get_predint(m, srcpred, 's')
+ dregread, dinvert, dunary, dall1s = get_predint(m, dstpred, 'd')
+ sidx, scrinvert = get_predcr(m, srcpred, 's')
+ didx, dcrinvert = get_predcr(m, dstpred, 'd')
+
+ with m.FSM(name="fetch_predicate"):
+
+ with m.State("FETCH_PRED_IDLE"):
+ comb += pred_insn_ready_o.eq(1)
+ with m.If(pred_insn_valid_i):
+ with m.If(predmode == SVP64PredMode.INT):
+ # skip fetching destination mask register, when zero
+ with m.If(dall1s):
+ sync += self.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(sregread)
+ comb += int_pred.ren.eq(1)
+ m.next = "INT_SRC_READ"
+ # fetch destination predicate register
+ with m.Else():
+ comb += int_pred.addr.eq(dregread)
+ comb += int_pred.ren.eq(1)
+ m.next = "INT_DST_READ"
+ with m.Else():
+ sync += self.srcmask.eq(-1)
+ sync += self.dstmask.eq(-1)
+ m.next = "FETCH_PRED_DONE"
+
+ with m.State("INT_DST_READ"):
+ # store destination mask
+ inv = Repl(dinvert, 64)
+ sync += self.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"
+ # fetch source predicate register
+ with m.Else():
+ comb += int_pred.addr.eq(sregread)
+ comb += int_pred.ren.eq(1)
+ m.next = "INT_SRC_READ"
+
+ with m.State("INT_SRC_READ"):
+ # store source mask
+ inv = Repl(sinvert, 64)
+ sync += self.srcmask.eq(self.int_pred.data_o ^ inv)
+ m.next = "FETCH_PRED_DONE"
+
+ with m.State("FETCH_PRED_DONE"):
+ comb += pred_mask_valid_o.eq(1)
+ with m.If(pred_mask_ready_i):
+ m.next = "FETCH_PRED_IDLE"
def issue_fsm(self, m, core, pc_changed, sv_changed, nia,
dbg, core_rst, is_svp64_mode,
fetch_pc_ready_o, fetch_pc_valid_i,
fetch_insn_valid_o, fetch_insn_ready_i,
+ pred_insn_valid_i, pred_insn_ready_o,
+ pred_mask_valid_o, pred_mask_ready_i,
exec_insn_valid_i, exec_insn_ready_o,
exec_pc_valid_o, exec_pc_ready_i):
"""issue FSM
comb += self.insn_done.eq(1)
m.next = "ISSUE_START"
with m.Else():
- m.next = "INSN_EXECUTE" # move to "execute"
+ if self.svp64_en:
+ m.next = "PRED_START" # start fetching predicate
+ else:
+ m.next = "INSN_EXECUTE" # skip predication
+
+ with m.State("PRED_START"):
+ comb += pred_insn_valid_i.eq(1) # tell fetch_pred to start
+ with m.If(pred_insn_ready_o): # fetch_pred acknowledged us
+ m.next = "MASK_WAIT"
+
+ with m.State("MASK_WAIT"):
+ comb += pred_mask_ready_i.eq(1) # ready to receive the masks
+ with m.If(pred_mask_valid_o): # predication masks are ready
+ m.next = "INSN_EXECUTE"
# handshake with execution FSM, move to "wait" once acknowledged
with m.State("INSN_EXECUTE"):
with m.If(is_svp64_mode):
- pred_src_zero = pdecode2.rm_dec.pred_sz
- pred_dst_zero = pdecode2.rm_dec.pred_dz
+ if self.svp64_en:
+ pred_src_zero = pdecode2.rm_dec.pred_sz
+ pred_dst_zero = pdecode2.rm_dec.pred_dz
"""
if not pred_src_zero:
fetch_insn_valid_o = Signal()
fetch_insn_ready_i = Signal()
+ # predicate fetch FSM decodes and fetches the predicate
+ pred_insn_valid_i = Signal()
+ pred_insn_ready_o = Signal()
+
+ # predicate fetch FSM delivers the masks
+ pred_mask_valid_o = Signal()
+ pred_mask_ready_i = Signal()
+
# issue FSM delivers the instruction to the be executed
exec_insn_valid_i = Signal()
exec_insn_ready_o = Signal()
# (as opposed to using sync - which would be on a clock's delay)
# this includes the actual opcode, valid flags and so on.
- # Fetch, then Issue, then Execute. Issue is where the VL for-loop
- # lives. the ready/valid signalling is used to communicate between
- # the three.
+ # Fetch, then predicate fetch, then Issue, then Execute.
+ # Issue is where the VL for-loop # lives. the ready/valid
+ # signalling is used to communicate between the four.
self.fetch_fsm(m, core, pc, svstate, nia, is_svp64_mode,
fetch_pc_ready_o, fetch_pc_valid_i,
dbg, core_rst, is_svp64_mode,
fetch_pc_ready_o, fetch_pc_valid_i,
fetch_insn_valid_o, fetch_insn_ready_i,
+ pred_insn_valid_i, pred_insn_ready_o,
+ pred_mask_valid_o, pred_mask_ready_i,
exec_insn_valid_i, exec_insn_ready_o,
exec_pc_valid_o, exec_pc_ready_i)
+ if self.svp64_en:
+ self.fetch_predicate_fsm(m,
+ pred_insn_valid_i, pred_insn_ready_o,
+ pred_mask_valid_o, pred_mask_ready_i)
+
self.execute_fsm(m, core, pc_changed, sv_changed,
exec_insn_valid_i, exec_insn_ready_o,
exec_pc_valid_o, exec_pc_ready_i)
return m
def do_dmi(self, m, dbg):
+ """deals with DMI debug requests
+
+ currently only provides read requests for the INT regfile, CR and XER
+ it will later also deal with *writing* to these regfiles.
+ """
comb = m.d.comb
sync = m.d.sync
dmi, d_reg, d_cr, d_xer, = dbg.dmi, dbg.d_gpr, dbg.d_cr, dbg.d_xer