from nmigen.compat.sim import run_simulation
from nmigen.cli import verilog, rtlil
from nmigen import Module, Signal, Mux, Cat, Elaboratable, Array
+from nmigen.hdl.rec import Record, Layout
from nmutil.latch import SRLatch, latchregister
+from soc.experiment.compalu_multi import go_record
from soc.experiment.testmem import TestMemory
from soc.decoder.power_enums import InternalOp
-from .alu_hier import CompALUOpSubset
+from soc.experiment.alu_hier import CompALUOpSubset
+from soc.decoder.power_enums import InternalOp, Function, CryIn
-# internal opcodes. hypothetically this could do more combinations.
-# meanings:
-# * bit 0: 0 = ADD , 1 = SUB
-# * bit 1: 0 = src1, 1 = IMM
-# * bit 2: 1 = LD
-# * bit 3: 1 = ST
-BIT0_ADD = 0
-BIT1_SRC = 1
-BIT2_ST = 2
-BIT3_LD = 3
-# convenience thingies.
-LDST_OP_ADD = 0b0000 # plain ADD (src1 + src2) - use this ALU as an ADD
-LDST_OP_SUB = 0b0001 # plain SUB (src1 - src2) - use this ALU as a SUB
-LDST_OP_ADDI = 0b0010 # immed ADD (imm + src1)
-LDST_OP_SUBI = 0b0011 # immed SUB (imm - src1)
-LDST_OP_ST = 0b0110 # immed ADD plus LD op. ADD result is address
-LDST_OP_LD = 0b1010 # immed ADD plus ST op. ADD result is address
+import operator
+
+
+class CompLDSTOpSubset(Record):
+ """CompLDSTOpSubset
+
+ a copy of the relevant subset information from Decode2Execute1Type
+ needed for LD/ST operations. use with eq_from_execute1 (below) to
+ grab subsets.
+ """
+ def __init__(self, name=None):
+ layout = (('insn_type', InternalOp),
+ ('imm_data', Layout((("imm", 64), ("imm_ok", 1)))),
+ ('is_32bit', 1),
+ ('is_signed', 1),
+ ('data_len', 4), # TODO: should be in separate CompLDSTSubset
+ ('byte_reverse', 1),
+ ('sign_extend', 1),
+ ('update', 1))
+
+ Record.__init__(self, Layout(layout), name=name)
+
+ # grrr. Record does not have kwargs
+ self.insn_type.reset_less = True
+ self.is_32bit.reset_less = True
+ self.is_signed.reset_less = True
+ self.data_len.reset_less = True
+ self.byte_reverse.reset_less = True
+ self.sign_extend.reset_less = True
+ self.update.reset_less = True
+
+ def eq_from_execute1(self, other):
+ """ use this to copy in from Decode2Execute1Type
+ """
+ res = []
+ for fname, sig in self.fields.items():
+ eqfrom = other.fields[fname]
+ res.append(sig.eq(eqfrom))
+ return res
+
+ def ports(self):
+ return [self.insn_type,
+ self.is_32bit,
+ self.is_signed,
+ self.data_len,
+ self.byte_reverse,
+ self.sign_extend,
+ self.update,
+ ]
class LDSTCompUnit(Elaboratable):
* :addr_o: Address out (LD or ST)
"""
- def __init__(self, rwid, alu, mem, n_src=2, n_dst=1):
+ def __init__(self, rwid, alu, mem, n_src=2, n_dst=1, debugtest=False):
self.rwid = rwid
self.alu = alu
self.mem = mem
+ self.debugtest = debugtest
self.counter = Signal(4)
src = []
j = i + 1 # name numbering to match dest1/2...
dst.append(Signal(rwid, name="dest%d_i" % j, reset_less=True))
- self.go_rd_i = Signal(n_src, reset_less=True) # go read in
- self.go_ad_i = Signal(n_dst, reset_less=True) # go address in
- self.go_wr_i = Signal(reset_less=True) # go write in
+ self.rd = go_record(n_src, name="rd") # read in, req out
+ self.wr = go_record(n_dst, name="wr") # write in, req out
+ self.go_rd_i = self.rd.go # temporary naming
+ self.go_wr_i = self.wr.go # temporary naming
+ self.rd_rel_o = self.rd.rel # temporary naming
+ self.req_rel_o = self.wr.rel # temporary naming
+
+ self.go_ad_i = Signal(reset_less=True) # go address in
self.go_st_i = Signal(reset_less=True) # go store in
self.issue_i = Signal(reset_less=True) # fn issue in
self.isalu_i = Signal(reset_less=True) # fn issue as ALU in
self.busy_o = Signal(reset_less=True) # fn busy out
self.dest = Array(dst)
self.data_o = dst[0] # Dest out
- self.rd_rel_o = Signal(n_src, reset_less=True) # request src1/src2
self.adr_rel_o = Signal(reset_less=True) # request address (from mem)
self.sto_rel_o = Signal(reset_less=True) # request store (to mem)
- self.req_rel_o = Signal(n_dst, reset_less=True) # req write (result)
self.done_o = Signal(reset_less=True) # final release signal
self.data_o = Signal(rwid, reset_less=True) # Dest out (LD or ALU)
self.addr_o = Signal(rwid, reset_less=True) # Address out (LD or ST)
reset_a = Signal(reset_less=True)
reset_s = Signal(reset_less=True)
reset_r = Signal(reset_less=True)
- comb += reset_b.eq(self.go_st_i | self.go_wr_i |
+ comb += reset_b.eq(self.go_st_i | self.wr.go |
self.go_ad_i | self.go_die_i)
- comb += reset_w.eq(self.go_wr_i | self.go_die_i)
+ comb += reset_w.eq(self.wr.go | self.go_die_i)
comb += reset_s.eq(self.go_st_i | self.go_die_i)
- comb += reset_r.eq(self.go_rd_i | self.go_die_i)
- # this one is slightly different, issue_alu_i selects go_wr_i)
- a_sel = Mux(self.isalu_i, self.go_wr_i, self.go_ad_i)
+ comb += reset_r.eq(self.rd.go | self.go_die_i)
+ # this one is slightly different, issue_alu_i selects wr.go)
+ a_sel = Mux(self.isalu_i, self.wr.go, self.go_ad_i)
comb += reset_a.eq(a_sel | self.go_die_i)
# opcode decode
op_is_st = Signal(reset_less=True)
op_ldst = Signal(reset_less=True)
op_is_imm = Signal(reset_less=True)
+
+ # ALU/LD data output control
alulatch = Signal(reset_less=True)
+ ldlatch = Signal(reset_less=True)
# src2 register
src2_r = Signal(self.rwid, reset_less=True)
sync += src_l.r.eq(reset_r)
# addr latch
- sync += adr_l.s.eq(self.go_rd_i)
+ sync += adr_l.s.eq(self.rd.go)
sync += adr_l.r.eq(reset_a)
# dest operand latch
- sync += req_l.s.eq(self.go_ad_i | self.go_st_i | self.go_wr_i)
+ sync += req_l.s.eq(self.go_ad_i | self.go_st_i | self.wr.go)
sync += req_l.r.eq(reset_w)
# store latch
- sync += sto_l.s.eq(self.go_rd_i) # XXX not sure which
+ sync += sto_l.s.eq(self.rd.go) # XXX not sure which
sync += sto_l.r.eq(reset_s)
# create a latch/register for the operand
oper_r = CompALUOpSubset() # Dest register
- latchregister(m, self.oper_i, oper_r, self.issue_i, name="oper_r")
+ latchregister(m, self.oper_i, oper_r, self.issue_i, name="oper_l")
# and one for the output from the ALU
data_r = Signal(self.rwid, reset_less=True) # Dest register
# outputs: busy and release signals
busy_o = self.busy_o
comb += self.busy_o.eq(opc_l.q) # busy out
- comb += self.rd_rel_o.eq(src_l.q & busy_o) # src1/src2 req rel
+ comb += self.rd.rel.eq(src_l.q & busy_o) # src1/src2 req rel
comb += self.sto_rel_o.eq(sto_l.q & busy_o & self.shadown_i & op_is_st)
# request release enabled based on if op is a LD/ST or a plain ALU
comb += wr_q.eq(req_l.q & (~op_ldst | op_is_ld))
comb += alulatch.eq((op_ldst & self.adr_rel_o) |
- (~op_ldst & self.req_rel_o))
+ (~op_ldst & self.wr.rel))
# select immediate if opcode says so. however also change the latch
# to trigger *from* the opcode latch instead.
self.src2_i))
# create a latch/register for src1/src2 (include immediate select)
- latchregister(m, self.src1_i, self.alu.a, src_l.q, name="src1_r")
- latchregister(m, self.src2_i, src2_r, src_l.q, name="src2_r")
+ latchregister(m, self.src1_i, self.alu.a, src_l.q, name="src1_l")
+ latchregister(m, self.src2_i, src2_r, src_l.q, name="src2_l")
latchregister(m, src2_or_imm, self.alu.b, src_sel, name="imm_r")
# decode bits of operand (latched)
# on a go_read, tell the ALU we're accepting data.
# NOTE: this spells TROUBLE if the ALU isn't ready!
# go_read is only valid for one clock!
- with m.If(self.go_rd_i): # src operands ready, GO!
+ with m.If(self.rd.go): # src operands ready, GO!
with m.If(~self.alu.p_ready_o): # no ACK yet
m.d.comb += self.alu.p_valid_i.eq(1) # so indicate valid
# only proceed if ALU says its output is valid
with m.If(self.alu.n_valid_o):
# write req release out. waits until shadow is dropped.
- comb += self.req_rel_o.eq(wr_q & busy_o & self.shadown_i)
+ comb += self.wr.rel.eq(wr_q & busy_o & self.shadown_i)
# address release only happens on LD/ST, and is shadowed.
comb += self.adr_rel_o.eq(adr_l.q & op_ldst & busy_o &
self.shadown_i)
# when output latch is ready, and ALU says ready, accept ALU output
- with m.If(self.req_rel_o):
+ with m.If(self.wr.rel):
# tells ALU "thanks got it"
m.d.comb += self.alu.n_ready_i.eq(1)
# provide "done" signal: select req_rel for non-LD/ST, adr_rel for LD/ST
- comb += self.done_o.eq((self.req_rel_o & ~op_ldst) |
+ comb += self.done_o.eq((self.wr.rel & ~op_ldst) |
(self.adr_rel_o & op_ldst))
# put the register directly onto the output bus on a go_write
# this is "ALU mode". go_wr_i *must* be deasserted on next clock
- with m.If(self.go_wr_i):
+ with m.If(self.wr.go):
comb += self.data_o.eq(data_r)
# "LD/ST" mode: put the register directly onto the *address* bus
# TODO: think about moving these to another module
+ if self.debugtest:
+ return m
+
# connect ST to memory. NOTE: unit *must* be set back
# to start again by dropping go_st_i on next clock
with m.If(self.stwd_mem_o):
# connect LD to memory. NOTE: unit *must* be set back
# to start again by dropping go_ad_i on next clock
+ rdport = self.mem.rdport
+ ldd_r = Signal(self.rwid, reset_less=True) # Dest register
+ # latch LD-out
+ latchregister(m, rdport.data, ldd_r, ldlatch, "ldo_r")
+ sync += ldlatch.eq(self.load_mem_o)
with m.If(self.load_mem_o):
- rdport = self.mem.rdport
comb += rdport.addr.eq(self.addr_o)
- comb += self.data_o.eq(rdport.data)
# comb += rdport.en.eq(1) # only when transparent=False
+ # if LD-latch, put ld-reg out onto output
+ with m.If(ldlatch | self.load_mem_o):
+ comb += self.data_o.eq(ldd_r)
+
return m
def __iter__(self):
- yield self.go_rd_i
+ yield self.rd.go
yield self.go_ad_i
- yield self.go_wr_i
+ yield self.wr.go
yield self.go_st_i
yield self.issue_i
yield self.isalu_i
yield from self.oper_i.ports()
yield from self.src_i
yield self.busy_o
- yield self.rd_rel_o
+ yield self.rd.rel
yield self.adr_rel_o
yield self.sto_rel_o
- yield self.req_rel_o
+ yield self.wr.rel
yield self.data_o
yield self.load_mem_o
yield self.stwd_mem_o
yield
yield dut.issue_i.eq(0)
yield
- yield dut.go_rd_i.eq(0b11)
- yield from wait_for(dut.rd_rel_o)
- yield dut.go_rd_i.eq(0)
+ yield dut.rd.go.eq(0b11)
+ yield from wait_for(dut.rd.rel)
+ yield dut.rd.go.eq(0)
yield from wait_for(dut.adr_rel_o)
yield dut.go_st_i.eq(1)
yield from wait_for(dut.sto_rel_o)
yield
yield dut.issue_i.eq(0)
yield
- yield dut.go_rd_i.eq(0b11)
- yield from wait_for(dut.rd_rel_o)
- yield dut.go_rd_i.eq(0)
+ yield dut.rd.go.eq(0b11)
+ yield from wait_for(dut.rd.rel)
+ yield dut.rd.go.eq(0)
yield from wait_for(dut.adr_rel_o)
yield dut.go_ad_i.eq(1)
yield from wait_for(dut.busy_o)
yield
yield dut.issue_i.eq(0)
yield
- yield dut.go_rd_i.eq(1)
- yield from wait_for(dut.rd_rel_o)
- yield dut.go_rd_i.eq(0)
- yield from wait_for(dut.req_rel_o)
- yield dut.go_wr_i.eq(1)
+ yield dut.rd.go.eq(1)
+ yield from wait_for(dut.rd.rel)
+ yield dut.rd.go.eq(0)
+ yield from wait_for(dut.wr.rel)
+ yield dut.wr.go.eq(1)
yield from wait_for(dut.busy_o)
yield
data = (yield dut.data_o)
- yield dut.go_wr_i.eq(0)
+ yield dut.wr.go.eq(0)
yield
# wait_for(dut.stwd_mem_o)
return data