46 lines
1.7 KiB
Plaintext
46 lines
1.7 KiB
Plaintext
package PinSync;
|
|
|
|
// mkPinSync builds a synchronizer for use with asynchronous inputs.
|
|
//
|
|
// You should only use this to capture asynchronous inputs coming from
|
|
// outside your design. For clock domain crossing within your design,
|
|
// use the dual-clocked synchronizers found in Bluespec's standard
|
|
// library.
|
|
//
|
|
// As the name suggests, mkPinSync is intended to be used to
|
|
// synchronize data coming into your design from an external pin, such
|
|
// as the RX line of a UART. Such signals do not run according to a
|
|
// known clock, so the regular stdlib synchronizers cannot be used as
|
|
// there's no "source" clock we can provide them.
|
|
//
|
|
// You can think of mkPinSync as the output end of a standard
|
|
// synchronizer, without the initial register that's clocked by the
|
|
// source domain. Conceptually, we assume that register exists outside
|
|
// our design and is driving the input of mkPinSync, so we just need
|
|
// the metastability mitigation within our own domain.
|
|
module mkPinSync(val init_value, Reg#(val) ifc)
|
|
provisos(Bits#(val, _));
|
|
|
|
Reg#(val) r1 <- mkReg(init_value);
|
|
Reg#(val) r2 <- mkReg(init_value);
|
|
|
|
// To break write+read conflicts. Without this, a rule that
|
|
// atomically reads the sync while also writing it fails to
|
|
// schedule vs. the 'every' rule below. This shouldn't really
|
|
// happen in real designs, but it's a convenient idiom in
|
|
// testing. The wire is free in terms of logic, so might as well
|
|
// make atomic read+write work.
|
|
Wire#(val) out <- mkBypassWire();
|
|
|
|
(* no_implicit_conditions, fire_when_enabled *)
|
|
rule every;
|
|
out <= r2;
|
|
r2 <= r1;
|
|
endrule
|
|
|
|
method _read = out._read;
|
|
method _write = r1._write;
|
|
endmodule
|
|
|
|
endpackage
|