def __init__(self, bitwidth, n_regs):
self.bitwidth = bitwidth
self.nregs = n_regs
- self.regwidth = bitwidth // n_regs
+ self.regwidth = regwidth = bitwidth // n_regs
super().__init__(self.regwidth, n_regs)
- # create suite of 8 write ports
- self.w_ports = [ self.write_port(f"extw_{i}") for i in range(n_regs) ]
- self.r_ports = [ self.read_port(f"extr_{i}") for i in range(n_regs) ]
- # now start again: one set will be internal, the other external
- self._rdports = []
- self._wrports = []
- for i in range(n_regs):
- self.write_port(f"intw_{i}")
- self.read_port(f"intr_{i}")
- # and append the "full" depth variant to the "external" ports
- self.w_ports.append(RecordObject([("wen", n_regs),
- ("data_i", bitwidth)], # *full* wid
- name="full"))
- self.r_ports.append(RecordObject([("ren", n_regs),
- ("data_o", bitwidth)], # *full* wid
- name="full"))
-
+ # "full" depth variant of the "external" port
+ self.full_wr = RecordObject([("wen", n_regs),
+ ("data_i", bitwidth)], # *full* wid
+ name="full_wr")
+ self.full_rd = RecordObject([("ren", n_regs),
+ ("data_o", bitwidth)], # *full* wid
+ name="full_rd")
def elaborate(self, platform):
- m = Module()
- comb, sync = m.d.comb, m.d.sync
-
- # connect up: detect if read is requested on large (full) port
- # nothing fancy needed because reads are same-cycle
- rlast = self.r_ports[-1]
- print (rlast)
- with m.If(self._get_en_sig([rlast], "ren") != 0):
- # wire up the enable signals and accumulate the data
- l = []
- print (self._rdports)
- for i, port in enumerate(self._rdports[:-1]):
- print (port)
- comb += port.ren.eq(1<<i) # port indices are *unary*-indexed
- l.append(port.data_o)
- comb += rlast.data_o.eq(Cat(*l)) # we like Cat on lists
- with m.Else():
- # allow request through the corresponding lower indexed ports
- for i, port in enumerate(self._rdports[:-1]):
- comb += port.eq(self.r_ports[i])
-
- # connect up: detect if write is requested on large (full) port
- # however due to the delay (1 clock) on write, we also need to
- # delay the test. enable is not-delayed, but data is.
- en_sig = Signal(reset_less=True) # combinatorial
- data_sig = Signal(reset_less=True) # sync (one clock delay)
-
- wlast = self.w_ports[-1]
- comb += en_sig.eq(self._get_en_sig(wlast, "wen") != 0)
- sync += data_sig.eq(en_sig)
-
- with m.If(en_sig):
- # wire up the enable signals
- for i, port in enumerate(self._wrports[:-1]):
- comb += port.wen.eq(1<<i) # port indices are *unary*-indexed
- with m.Else():
- # allow request through the corresponding lower indexed ports
- for i, port in enumerate(self._wrports[:-1]):
- comb += port.wen.eq(self.w_ports[i].wen)
-
- # and (sigh) data is on one clock-delay, connect that too
- with m.If(data_sig):
- # get list of all data_i and assign to them via Cat
- l = map(lambda port: port.data_i, self._wrports[:-1])
- comb += Cat(*l).eq(wlast.data_i)
- with m.Else():
- # allow data through the corresponding lower indexed ports
- for i, port in enumerate(self._wrports[:-1]):
- comb += self.w_ports[i].data_i.eq(port.data_i)
+ m = super().elaborate(platform)
+ comb = m.d.comb
+
+ # for internal use only.
+ wr_regs = self.write_reg_port(f"w")
+ rd_regs = self.read_reg_port(f"r")
+
+ # connect up full read port
+ rfull = self.full_rd
+
+ # wire up the enable signals and chain-accumulate the data
+ l = map(lambda port: port.data_o, rd_regs) # get port data(s)
+ le = map(lambda port: port.ren, rd_regs) # get port ren(s)
+
+ comb += rfull.data_o.eq(Cat(*l)) # we like Cat on lists
+ comb += Cat(*le).eq(rfull.ren)
+
+ # connect up full write port
+ wfull = self.full_wr
+
+ # wire up the enable signals from the large (full) port
+ l = map(lambda port: port.data_i, wr_regs)
+ le = map(lambda port: port.wen, wr_regs) # get port wen(s)
+
+ # get list of all data_i (and wens) and assign to them via Cat
+ comb += Cat(*l).eq(wfull.data_i)
+ comb += Cat(*le).eq(wfull.wen)
+
+ return m
+
+ def __iter__(self):
+ yield from super().__iter__()
+ yield from self.full_wr
+ yield from self.full_rd
+
+
+def regfile_array_sim(dut, rp1, rp2, rp3, wp):
+ # part-port write
+ yield wp.data_i.eq(2)
+ yield wp.wen.eq(1<<1)
+ yield
+ yield wp.wen.eq(0)
+ # part-port read
+ yield rp1.ren.eq(1<<1)
+ yield
+ data = yield rp1.data_o
+ print (data)
+ assert data == 2
+
+ # simultaneous read/write - should be pass-thru
+ yield rp1.ren.eq(1<<5)
+ yield rp2.ren.eq(1<<1)
+ yield wp.wen.eq(1<<5)
+ yield wp.data_i.eq(6)
+ yield
+ yield wp.wen.eq(0)
+ yield rp1.ren.eq(0)
+ yield rp2.ren.eq(0)
+ data1 = yield rp1.data_o
+ print (data1)
+ assert data1 == 6, data1
+ data2 = yield rp2.data_o
+ print (data2)
+ assert data2 == 2, data2
+ yield
+ data = yield rp1.data_o
+ print (data)
+
+ # full port read (whole reg)
+ yield dut.full_rd.ren.eq(0xff)
+ yield
+ yield dut.full_rd.ren.eq(0)
+ data = yield dut.full_rd.data_o
+ print (hex(data))
+
+ # full port read (part reg)
+ yield dut.full_rd.ren.eq(0x1<<5)
+ yield
+ yield dut.full_rd.ren.eq(0)
+ data = yield dut.full_rd.data_o
+ print (hex(data))
+
+ # full port part-write (part masked reg)
+ yield dut.full_wr.wen.eq(0x1<<1)
+ yield dut.full_wr.data_i.eq(0xe0)
+ yield
+ yield dut.full_wr.wen.eq(0x0)
+
+ # full port read (whole reg)
+ yield dut.full_rd.ren.eq(0xff)
+ yield
+ yield dut.full_rd.ren.eq(0)
+ data = yield dut.full_rd.data_o
+ print (hex(data))
+
+ # full port write
+ yield dut.full_wr.wen.eq(0xff)
+ yield dut.full_wr.data_i.eq(0xcafeface)
+ yield
+ yield dut.full_wr.wen.eq(0x0)
+
+ # full port read (whole reg)
+ yield dut.full_rd.ren.eq(0xff)
+ yield
+ yield dut.full_rd.ren.eq(0)
+ data = yield dut.full_rd.data_o
+ print (hex(data))
+
+ # part write
+ yield wp.data_i.eq(2)
+ yield wp.wen.eq(1<<1)
+ yield
+ yield wp.wen.eq(0)
+ yield rp1.ren.eq(1<<1)
+ yield
+ data = yield rp1.data_o
+ print (hex(data))
+ assert data == 2
+
+ # full port read (whole reg)
+ yield dut.full_rd.ren.eq(0xff)
+ yield
+ yield dut.full_rd.ren.eq(0)
+ data = yield dut.full_rd.data_o
+ print (hex(data))
+
+ # simultaneous read/write: full-write, part-write, 3x part-read
+ yield rp1.ren.eq(1<<5)
+ yield rp2.ren.eq(1<<1)
+ yield rp3.ren.eq(1<<3)
+ yield wp.wen.eq(1<<3)
+ yield wp.data_i.eq(6)
+ yield dut.full_wr.wen.eq((1<<1) | (1<<5))
+ yield dut.full_wr.data_i.eq((0xa<<(1*4)) | (0x3<<(5*4)))
+ yield
+ yield dut.full_wr.wen.eq(0)
+ yield wp.wen.eq(0)
+ yield rp1.ren.eq(0)
+ yield rp2.ren.eq(0)
+ yield rp3.ren.eq(0)
+ data1 = yield rp1.data_o
+ print (hex(data1))
+ assert data1 == 0x3
+ data2 = yield rp2.data_o
+ print (hex(data2))
+ assert data2 == 0xa
+ data3 = yield rp3.data_o
+ print (hex(data3))
+ assert data3 == 0x6
def test_regfile():
dut = VirtualRegPort(32, 8)
- vl = rtlil.convert(dut, ports=dut.ports() + dut.w_ports + dut.r_ports)
+ rp1 = dut.read_port("read1")
+ rp2 = dut.read_port("read2")
+ rp3 = dut.read_port("read3")
+ wp = dut.write_port("write")
+
+ ports=dut.ports()
+ print ("ports", ports)
+ vl = rtlil.convert(dut, ports=ports)
with open("test_virtualregfile.il", "w") as f:
f.write(vl)
+ run_simulation(dut, regfile_array_sim(dut, rp1, rp2, rp3, wp),
+ vcd_name='test_regfile_array.vcd')
+
if __name__ == '__main__':
test_regfile()
+
projects / soc.git / blobdiff