from nmigen.compat.sim import run_simulation
from nmigen.cli import verilog, rtlil
-from nmigen import Module, Signal, Elaboratable
+from nmigen import Module, Signal, Mux, Elaboratable
from nmutil.latch import SRLatch, latchregister
+""" Computation Unit (aka "ALU Manager").
+
+ This module runs a "revolving door" set of three latches, based on
+ * Issue
+ * Go_Read
+ * Go_Write
+ where one of them cannot be set on any given cycle.
+ (Note however that opc_l has been inverted (and qn used), due to SRLatch
+ default reset state being "0" rather than "1")
+
+ * When issue is first raised, a busy signal is sent out.
+ The src1 and src2 registers and the operand can be latched in
+ at this point
+
+ * Read request is set, which is acknowledged through the Scoreboard
+ to the priority picker, which generates (one and only one) Go_Read
+ at a time. One of those will (eventually) be this Computation Unit.
+
+ * Once Go_Read is set, the src1/src2/operand latch door shuts (locking
+ src1/src2/operand in place), and the ALU is told to proceed.
+
+ * As this is currently a "demo" unit, a countdown timer is activated
+ to simulate an ALU "pipeline", which activates "write request release",
+ and the ALU's output is captured into a temporary register.
+
+ * Write request release will go through a similar process as Read request,
+ resulting (eventually) in Go_Write being asserted.
+
+ * When Go_Write is asserted, two things happen: (1) the data in the temp
+ register is placed combinatorially onto the output, and (2) the
+ req_l latch is cleared, busy is dropped, and the Comp Unit is back
+ through its revolving door to do another task.
+
+ Notes on oper_i:
+
+ * bits[0:2] are for the ALU, add=0, sub=1, shift=2, mul=3
+ * bit[2] are the immediate (bit[2]=1 == immediate mode)
+"""
class ComputationUnitNoDelay(Elaboratable):
def __init__(self, rwid, opwid, alu):
+ self.opwid = opwid
self.rwid = rwid
self.alu = alu
- self.counter = Signal(3)
+ self.counter = Signal(4)
self.go_rd_i = Signal(reset_less=True) # go read in
self.go_wr_i = Signal(reset_less=True) # go write in
self.issue_i = Signal(reset_less=True) # fn issue in
+ self.shadown_i = Signal(reset=1) # shadow function, defaults to ON
+ self.go_die_i = Signal() # go die (reset)
self.oper_i = Signal(opwid, reset_less=True) # opcode in
+ self.imm_i = Signal(rwid, reset_less=True) # immediate in
self.src1_i = Signal(rwid, reset_less=True) # oper1 in
self.src2_i = Signal(rwid, reset_less=True) # oper2 in
m.submodules.opc_l = opc_l = SRLatch(sync=False)
m.submodules.req_l = req_l = SRLatch(sync=False)
+ # shadow/go_die
+ reset_w = Signal(reset_less=True)
+ reset_r = Signal(reset_less=True)
+ m.d.comb += reset_w.eq(self.go_wr_i | self.go_die_i)
+ m.d.comb += reset_r.eq(self.go_rd_i | self.go_die_i)
+
# This is fascinating and very important to observe that this
# is in effect a "3-way revolving door". At no time may all 3
# latches be set at the same time.
# opcode latch (not using go_rd_i) - inverted so that busy resets to 0
- m.d.comb += opc_l.s.eq(self.issue_i) # XXX NOTE: INVERTED FROM book!
- m.d.comb += opc_l.r.eq(self.go_wr_i) # XXX NOTE: INVERTED FROM book!
+ m.d.sync += opc_l.s.eq(self.issue_i) # XXX NOTE: INVERTED FROM book!
+ m.d.sync += opc_l.r.eq(reset_w) # XXX NOTE: INVERTED FROM book!
# src operand latch (not using go_wr_i)
- m.d.comb += src_l.s.eq(self.issue_i)
- m.d.comb += src_l.r.eq(self.go_rd_i)
+ m.d.sync += src_l.s.eq(self.issue_i)
+ m.d.sync += src_l.r.eq(reset_r)
# dest operand latch (not using issue_i)
- m.d.comb += req_l.s.eq(self.go_rd_i)
- m.d.comb += req_l.r.eq(self.go_wr_i)
+ m.d.sync += req_l.s.eq(self.go_rd_i)
+ m.d.sync += req_l.r.eq(reset_w)
- # XXX
- # XXX NOTE: sync on req_rel_o and data_o due to simulation lock-up
- # XXX
- # outputs
- m.d.comb += self.busy_o.eq(opc_l.q) # busy out
- m.d.comb += self.rd_rel_o.eq(src_l.q & opc_l.q) # src1/src2 req rel
+ # create a latch/register for the operand
+ oper_r = Signal(self.opwid+1, reset_less=True) # opcode reg
+ latchregister(m, self.oper_i, oper_r, self.issue_i)
- with m.If(req_l.qn & opc_l.q & (self.counter == 0)):
- m.d.sync += self.counter.eq(3)
- with m.If(self.counter > 0):
- m.d.sync += self.counter.eq(self.counter - 1)
- with m.If((self.counter == 1) | (self.counter == 0)):
- m.d.comb += self.req_rel_o.eq(req_l.q & opc_l.q) # req release out
+ # and one for the output from the ALU
+ data_r = Signal(self.rwid, reset_less=True) # Dest register
+ latchregister(m, self.alu.o, data_r, req_l.q)
- # create a latch/register for src1/src2
- latchregister(m, self.src1_i, self.alu.a, src_l.q)
- latchregister(m, self.src2_i, self.alu.b, src_l.q)
- #with m.If(src_l.qn):
- # m.d.comb += self.alu.op.eq(self.oper_i)
+ # get the top 2 bits for the ALU
+ m.d.comb += self.alu.op.eq(oper_r[0:2])
- # create a latch/register for the operand
- latchregister(m, self.oper_i, self.alu.op, src_l.q)
+ # 3rd bit is whether this is an immediate or not
+ op_is_imm = Signal(reset_less=True)
+ m.d.comb += op_is_imm.eq(oper_r[2])
- # and one for the output from the ALU
- data_o = Signal(self.rwid, reset_less=True) # Dest register
- latchregister(m, self.alu.o, data_o, req_l.q)
+ # select immediate if opcode says so. however also change the latch
+ # to trigger *from* the opcode latch instead.
+ src2_or_imm = Signal(self.rwid, reset_less=True)
+ src_sel = Signal(reset_less=True)
+ m.d.comb += src_sel.eq(Mux(op_is_imm, opc_l.qn, src_l.q))
+ m.d.comb += src2_or_imm.eq(Mux(op_is_imm, self.imm_i, self.src2_i))
+ # create a latch/register for src1/src2
+ latchregister(m, self.src1_i, self.alu.a, src_l.q)
+ latchregister(m, src2_or_imm, self.alu.b, src_sel)
+
+ # -----
+ # outputs
+ # -----
+
+ # all request signals gated by busy_o. prevents picker problems
+ busy_o = self.busy_o
+ m.d.comb += busy_o.eq(opc_l.q) # busy out
+ m.d.comb += self.rd_rel_o.eq(src_l.q & busy_o) # src1/src2 req rel
+
+ # 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.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):
+ # when ALU ready, write req release out. waits for shadow
+ m.d.comb += self.req_rel_o.eq(req_l.q & busy_o & self.shadown_i)
+ # when output latch is ready, and ALU says ready, accept ALU output
+ with m.If(self.req_rel_o):
+ m.d.comb += self.alu.n_ready_i.eq(1) # tells ALU "thanks got it"
+
+ # output the data from the latch on go_write
with m.If(self.go_wr_i):
- m.d.comb += self.data_o.eq(data_o)
+ m.d.comb += self.data_o.eq(data_r)
return m
+ def __iter__(self):
+ yield self.go_rd_i
+ yield self.go_wr_i
+ yield self.issue_i
+ yield self.shadown_i
+ yield self.go_die_i
+ yield self.oper_i
+ yield self.imm_i
+ yield self.src1_i
+ yield self.src2_i
+ yield self.busy_o
+ yield self.rd_rel_o
+ yield self.req_rel_o
+ yield self.data_o
+
+ def ports(self):
+ return list(self)
+
+
def scoreboard_sim(dut):
yield dut.dest_i.eq(1)
yield dut.issue_i.eq(1)
yield
def test_scoreboard():
- dut = Scoreboard(32, 8)
+ from alu_hier import ALU
+ alu = ALU(16)
+ dut = ComputationUnitNoDelay(16, 8, alu)
vl = rtlil.convert(dut, ports=dut.ports())
- with open("test_scoreboard.il", "w") as f:
+ with open("test_compalu.il", "w") as f:
f.write(vl)
- run_simulation(dut, scoreboard_sim(dut), vcd_name='test_scoreboard.vcd')
+ run_simulation(dut, scoreboard_sim(dut), vcd_name='test_compalu.vcd')
if __name__ == '__main__':
test_scoreboard()