flake.nix: grab the GUI version of nextpnr

In practice the flow control is unusable on ULX3S dev boards because
the CTS line isn't hooked up (it's instead wired to JTAG_TDO, to enable
the USB<>UART chip to serve a dual purpose as a bitbanged JTAG programmer)

Still, support for flow control is nice, for the future. And the UART
itself also works regardless of flow control, which is of course nice.

debugger/UART.bsv

@@ -0,0 +1,243 @@
+package UART;
+
+import Cntrs::*;
+import GetPut::*;
+import FIFOF::*;
+import SpecialFIFOs::*;
+import StmtFSM::*;
+import Connectable::*;
+
+import PinSync::*;
+import GlitchFilter::*;
+import Strobe::*;
+
+(* always_enabled *)
+interface UART_RX_PHY;
+   (* prefix="" *)
+   method Action rx_in((* port="rx_in" *) bit b);
+   (* result="cts" *)
+   method Bool stop_sending();
+endinterface
+
+(* always_enabled *)
+interface UART_TX_PHY;
+   (* result="tx_out" *)
+   method bit tx_out();
+   (* prefix="" *)
+   method Action can_send((* port="rts" *) Bool send);
+endinterface
+
+(* always_enabled *)
+interface UART_PHY;
+   (* prefix="" *)
+   method Action rx_in((* port="rx_in" *) bit b);
+   (* result="tx_out" *)
+   method bit tx_out();
+   (* result="cts" *)
+   method Bool stop_sending();
+   (* prefix="" *)
+   method Action can_send((* port="rts" *) Bool send);
+endinterface
+
+interface UART_RX;
+   interface UART_RX_PHY phy;
+   interface Get#(Bit#(8)) receive;
+endinterface
+
+interface UART_TX;
+   interface UART_TX_PHY phy;
+   interface Put#(Bit#(8)) send;
+endinterface
+
+interface UART;
+   interface UART_PHY phy;
+   interface Put#(Bit#(8)) send;
+   interface Get#(Bit#(8)) receive;
+endinterface
+
+typedef enum {
+   WaitIdle,
+   Idle,
+   Read,
+   Stop,
+   Cork
+} RXState deriving (Bits, Eq);
+
+module mkUARTReceiver(Integer clock_frequency, Integer uart_bitrate, UART_RX ifc);
+   Reg#(bit) rx_sync <- mkPinSync(0);
+   let rx_in <- mkGlitchFilter(3, 0);
+   mkConnection(toGet(asReg(rx_sync)), toPut(asReg(rx_in)));
+
+   Reg#(RXState) rx_state <- mkReg(WaitIdle);
+   Strobe bit_16x_strobe <- mkStrobe(clock_frequency, 16*uart_bitrate);
+   Count#(UInt#(4)) cnt <- mkCount(0);
+   Reg#(Bit#(8)) shift_in <- mkReg(0);
+   FIFOF#(Bit#(8)) rx <- mkBypassFIFOF();
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule rx_counter (bit_16x_strobe);
+      cnt.incr(1);
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule rx_wait_idle (rx_state == WaitIdle && bit_16x_strobe && rx_in == 1);
+      rx_state <= Idle;
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule rx_idle (rx_state == Idle && bit_16x_strobe && rx_in == 0);
+      rx_state <= Read;
+      cnt <= 1;
+      shift_in <= 8'hFF;
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule rx_read (rx_state == Read && bit_16x_strobe && cnt == 7);
+      let shifted_out = shift_in[0];
+      shift_in <= {rx_in, shift_in[7:1]};
+      if (shifted_out == 0) begin // start bit reached end of shiftreg
+         rx_state <= Stop;
+      end
+   endrule
+
+   (* fire_when_enabled *)
+   rule rx_stop (rx_state == Stop);
+      if (rx.notFull) begin
+         rx.enq(shift_in);
+         rx_state <= WaitIdle;
+      end
+      else
+         rx_state <= Cork;
+   endrule
+
+   (* fire_when_enabled *)
+   rule rx_cork (rx_state == Cork);
+      rx.enq(shift_in);
+      rx_state <= WaitIdle;
+   endrule
+
+   interface UART_RX_PHY phy;
+      method rx_in = rx_sync._write;
+      method Bool stop_sending();
+         // Signal is active low, so 1 == "stop sending"
+         return rx_state == Cork;
+      endmethod
+   endinterface
+
+   interface receive = toGet(rx);
+endmodule
+
+typedef enum {
+   Idle,
+   Ready,
+   Send
+} TXState deriving (Bits, Eq);
+
+module mkUARTTransmitter(Integer clock_frequency, Integer uart_bitrate, UART_TX ifc);
+   Reg#(bit) cts_sync <- mkPinSync(0);
+   let cts_in <- mkGlitchFilter(3, 0);
+   mkConnection(toGet(asReg(cts_sync)), toPut(asReg(cts_in)));
+
+   Reg#(TXState) tx_state <- mkReg(Idle);
+   Strobe bit_strobe <- mkStrobe(clock_frequency, uart_bitrate);
+   Reg#(UInt#(4)) cnt <- mkReg(0);
+   Reg#(Bit#(9)) shift_out <- mkReg(9'h1FF);
+   FIFOF#(Bit#(8)) tx <- mkPipelineFIFOF();
+
+   (* fire_when_enabled *)
+   rule tx_idle (tx_state == Idle && tx.notEmpty);
+      shift_out <= {tx.first, 0};
+      tx.deq();
+      tx_state <= Ready;
+      bit_strobe.reset();
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule rx_ready (tx_state == Ready && bit_strobe && cts_in == 1);
+      tx_state <= Send;
+      cnt <= 0;
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule tx_send (tx_state == Send && bit_strobe);
+      shift_out <= {1'b1, shift_out[8:1]};
+      cnt <= cnt+1;
+      if (cnt == 9)
+         tx_state <= Idle;
+   endrule
+
+   interface UART_TX_PHY phy;
+      method bit tx_out();
+         if (tx_state == Send)
+            return shift_out[0];
+         else
+            return 1'b1;
+      endmethod
+      method Action can_send(b);
+         cts_sync <= pack(b);
+      endmethod
+   endinterface
+
+   interface Put send = toPut(tx);
+endmodule
+
+module mkUART(Integer clock_frequency, Integer uart_bitrate, UART ifc);
+   let _rx <- mkUARTReceiver(clock_frequency, uart_bitrate);
+   let _tx <- mkUARTTransmitter(clock_frequency, uart_bitrate);
+
+   interface UART_PHY phy;
+      method rx_in = _rx.phy.rx_in;
+      method tx_out = _tx.phy.tx_out;
+      method stop_sending = _rx.phy.stop_sending;
+      method can_send = _tx.phy.can_send;
+   endinterface
+   interface send = _tx.send;
+   interface receive = _rx.receive;
+endmodule
+
+typeclass FlowControlled#(type ifc);
+   module disableFlowControl(ifc i, Empty ret);
+endtypeclass
+
+instance FlowControlled#(UART_RX_PHY);
+   module disableFlowControl(UART_RX_PHY phy, Empty ifc);
+   endmodule
+endinstance
+
+instance FlowControlled#(UART_TX_PHY);
+   module disableFlowControl(UART_TX_PHY phy, Empty ifc);
+      (* no_implicit_conditions,fire_when_enabled *)
+      rule always_send;
+         phy.can_send(True);
+      endrule
+   endmodule
+endinstance
+
+instance FlowControlled#(UART_PHY);
+   module disableFlowControl(UART_PHY phy, Empty ifc);
+      (* no_implicit_conditions,fire_when_enabled *)
+      rule always_send;
+         phy.can_send(True);
+      endrule
+   endmodule
+endinstance
+
+instance FlowControlled#(UART_RX);
+   module disableFlowControl(UART_RX rx, Empty ifc);
+      disableFlowControl(rx.phy);
+   endmodule
+endinstance
+
+instance FlowControlled#(UART_TX);
+   module disableFlowControl(UART_TX tx, Empty ifc);
+      disableFlowControl(tx.phy);
+   endmodule
+endinstance
+
+instance FlowControlled#(UART);
+   module disableFlowControl(UART uart, Empty ifc);
+      disableFlowControl(uart.phy);
+   endmodule
+endinstance
+
+endpackage

debugger/UART_Test.bsv

@@ -0,0 +1,265 @@
+package UART_Test;
+
+import Assert::*;
+import StmtFSM::*;
+import Connectable::*;
+import GetPut::*;
+import Probe::*;
+
+import UART::*;
+import Testing::*;
+import Strobe::*;
+
+interface Test;
+   method Action start();
+   method Bool done();
+endinterface
+
+module mkTestReceiver(Test);
+   Reg#(Bool) running <- mkReg(False);
+
+   let testflags <- mkTestFlags();
+   let cycles <- mkCycleCounter();
+
+   let dut <- mkUARTReceiver(25_000_000, 115_200);
+   Reg#(bit) tx <- mkReg(1);
+   mkConnection(toGet(asReg(tx)), toPut(dut.phy.rx_in));
+
+   Probe#(Bit#(8)) read_probe <- mkProbe();
+   Probe#(Bool) cork_probe <- mkProbe();
+   rule record_cork;
+      cork_probe <= dut.phy.stop_sending();
+   endrule
+
+   Reg#(Bit#(40)) shift[2] <- mkCReg(2, 40'hFFFFFFFFFF);
+   let shift_in_strobe <- mkStrobe(25_000_000, 115_200);
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule shift_in (running && shift_in_strobe && (tx == 0 || !dut.phy.stop_sending));
+      if (testflags.verbose)
+         $display("%0d (%0d): to UART: %0d", cycles.all, cycles, shift[0][0]);
+      tx <= shift[0][0];
+      shift[0] <= {1'b1, shift[0][39:1]};
+   endrule
+
+   function Action start_tx(Bit#(40) val);
+      return action
+                shift[1] <= val;
+             endaction;
+   endfunction
+
+   function Bool tx_idle();
+      return shift[0] == 40'hFFFFFFFFFF;
+   endfunction
+
+   function Bool tx_is(Bit#(40) val);
+      return shift[0] == val;
+   endfunction
+
+   let fsm <- mkFSM(seq
+      running <= True;
+      // Let UART initialize and settle
+      repeat (61) noAction;
+      // desynchronize sender and receiver strobes, to emulate a
+      // real setup.
+      shift_in_strobe.reset();
+
+      action
+         dynamicAssert(tx_idle(), "transmitter not idle");
+         dynamicAssert(!dut.phy.stop_sending(), "receiver not ready to receive");
+      endaction
+
+      // Single byte transmission
+      action
+         start_tx({10'h3FF, 10'h3FF, 10'h3FF, 1'b1, 8'd77, 1'b0});
+         cycles.reset();
+      endaction
+
+      // Read received byte
+      action
+         let got <- dut.receive.get();
+         read_probe <= got;
+         if (testflags.verbose)
+            $display("%0d (%0d): UART.rx = %0d, want %0d", cycles.all, cycles, got, 77);
+         dynamicAssert(got == 77, "wrong byte received");
+         dynamicAssert(cycles < 2100, "byte not received during stop bit");
+      endaction
+      dynamicAssert(!dut.phy.stop_sending(), "receiver not ready to receive");
+
+      await(shift_in_strobe);
+      action
+         dynamicAssert(tx_idle(), "transmitter not idle");
+         dynamicAssert(!dut.phy.stop_sending(), "receiver not ready to receive");
+      endaction
+      await(shift_in_strobe);
+
+      // Send 4 bytes back to back, to check flow control
+      action
+         start_tx({
+            1'b1, 8'd25, 1'b0,
+            1'b1, 8'd14, 1'b0,
+            1'b1, 8'd59, 1'b0,
+            1'b1, 8'd42, 1'b0});
+         cycles.reset();
+      endaction
+
+      // Once two bytes are sent, the receiver should cork.
+      await(tx_is({10'h3FF, 10'h3FF, 1'b1, 8'd25, 1'b0, 1'b1, 8'd14, 1'b0}));
+      repeat (1000) noAction;
+      dynamicAssert(dut.phy.stop_sending(), "receiver did not cork sender");
+
+      // Read out one byte, verify that one more byte can transmit.
+      action
+         let got <- dut.receive.get();
+         read_probe <= got;
+         if (testflags.verbose)
+            $display("%0d (%0d): UART.rx = %0d, want %0d", cycles.all, cycles, got, 42);
+         dynamicAssert(got == 42, "wrong byte received");
+         dynamicAssert(cycles < 5400, "byte not received during stop bit");
+      endaction
+      await(tx_is({10'h3FF, 10'h3FF, 10'h3FF, 1'b1, 8'd25, 1'b0}));
+      dynamicAssert(dut.phy.stop_sending(), "receiver did not cork sender");
+
+      // Read out one more byte, check the final byte transmits.
+      action
+         let got <- dut.receive.get();
+         read_probe <= got;
+         if (testflags.verbose)
+            $display("%0d (%0d): UART.rx = %0d, want %0d", cycles.all, cycles, got, 59);
+         dynamicAssert(got == 59, "wrong byte received");
+         dynamicAssert(cycles < 7600, "byte not received during stop bit");
+      endaction
+      await(tx_is(40'hFFFFFFFFFF));
+      repeat (1000) noAction;
+      dynamicAssert(dut.phy.stop_sending(), "receiver did not cork sender");
+
+      // Read the final two bytes from the receiver.
+      action
+         let got <- dut.receive.get();
+         read_probe <= got;
+         if (testflags.verbose)
+            $display("%0d (%0d): UART.rx = %0d, want %0d", cycles.all, cycles, got, 14);
+         dynamicAssert(got == 14, "wrong byte received");
+         dynamicAssert(cycles < 10500, "byte not received during stop bit");
+      endaction
+      repeat (1000) noAction;
+      dynamicAssert(!dut.phy.stop_sending(), "receiver did not uncork sender");
+
+      action
+         let got <- dut.receive.get();
+         read_probe <= got;
+         if (testflags.verbose)
+            $display("%0d (%0d): UART.rx = %0d, want %0d", cycles.all, cycles, got, 25);
+         dynamicAssert(got == 25, "wrong byte received");
+         dynamicAssert(cycles < 11500, "byte not received during stop bit");
+      endaction
+      repeat (1000) noAction;
+      dynamicAssert(!dut.phy.stop_sending(), "receiver did not uncork sender");
+      running <= False;
+   endseq);
+   method start = fsm.start;
+   method done = fsm.done;
+endmodule
+
+module mkTestTransmitter(Test);
+   let testflags <- mkTestFlags();
+   let cycles <- mkCycleCounter();
+
+   let receiver <- mkUARTReceiver(25_000_000, 115_200);
+   let dut <- mkUARTTransmitter(25_000_000, 115_200);
+   mkConnection(toGet(dut.phy.tx_out), toPut(receiver.phy.rx_in));
+   mkConnection(toGet(True), toPut(dut.phy.can_send));
+
+   Probe#(bit) tx_probe <- mkProbe();
+   Probe#(Bit#(8)) recv_probe <- mkProbe();
+   rule record_tx;
+      tx_probe <= dut.phy.tx_out;
+   endrule
+
+   let fsm <- mkFSM(seq
+      // Let UART initialize and settle
+      repeat (61) noAction;
+
+      action
+         dut.send.put(42);
+         cycles.reset();
+         if (testflags.verbose)
+            $display("%0d: send", cycles.all);
+      endaction
+      action
+         let got <- receiver.receive.get();
+         recv_probe <= got;
+         if (testflags.verbose)
+            $display("%0d (%0d): received %0d", cycles.all, cycles, got);
+         dynamicAssert(got == 42, "wrong value received");
+         dynamicAssert(cycles < 2000, "value received too late");
+      endaction
+
+      repeat(1000) noAction;
+      par
+         seq
+            dut.send.put(1);
+            dut.send.put(2);
+            dut.send.put(5);
+            dut.send.put(3);
+            dut.send.put(4);
+         endseq
+
+         seq
+            action
+               let got <- receiver.receive.get();
+               recv_probe <= got;
+               if (testflags.verbose)
+                  $display("%0d (%0d): received %0d", cycles.all, cycles, got);
+               dynamicAssert(got == 1, "wrong byte received");
+            endaction
+            action
+               let got <- receiver.receive.get();
+               recv_probe <= got;
+               if (testflags.verbose)
+                  $display("%0d (%0d): received %0d", cycles.all, cycles, got);
+               dynamicAssert(got == 2, "wrong byte received");
+            endaction
+            action
+               let got <- receiver.receive.get();
+               recv_probe <= got;
+               if (testflags.verbose)
+                  $display("%0d (%0d): received %0d", cycles.all, cycles, got);
+               dynamicAssert(got == 5, "wrong byte received");
+            endaction
+            action
+               let got <- receiver.receive.get();
+               recv_probe <= got;
+               if (testflags.verbose)
+                  $display("%0d (%0d): received %0d", cycles.all, cycles, got);
+               dynamicAssert(got == 3, "wrong byte received");
+            endaction
+            action
+               let got <- receiver.receive.get();
+               recv_probe <= got;
+               if (testflags.verbose)
+                  $display("%0d (%0d): received %0d", cycles.all, cycles, got);
+               dynamicAssert(got == 4, "wrong byte received");
+            endaction
+         endseq
+      endpar
+   endseq);
+
+   method start = fsm.start;
+   method done = fsm.done;
+endmodule
+
+module mkTB();
+   let rx_test <- mkTestReceiver();
+   let tx_test <- mkTestTransmitter();
+
+   runTest(30000,
+      mkTest("UART", seq
+         rx_test.start();
+         await(rx_test.done);
+
+         tx_test.start();
+         await(tx_test.done);
+      endseq));
+endmodule
+
+endpackage

experiments/uart/Top.bsv

@@ -0,0 +1,33 @@
+package Top;
+
+import GetPut::*;
+
+import UART::*;
+
+(* always_enabled *)
+interface Top;
+   (* prefix="" *)
+   method Action rx_in((* port="rx_in" *) bit b);
+   method bit tx_out();
+   method Bit#(8) leds();
+endinterface
+
+(* synthesize *)
+module mkTop(Top);
+   let _ret <- mkUART(100_000_000, 115_200);
+   disableFlowControl(_ret);
+
+   Reg#(Bit#(8)) leds_out <- mkReg(0);
+
+   rule recv;
+      let b <- _ret.receive.get();
+      _ret.send.put(b);
+      leds_out <= b;
+   endrule
+
+   method rx_in = _ret.phy.rx_in;
+   method tx_out = _ret.phy.tx_out;
+   method leds = leds_out._read;
+endmodule
+
+endpackage

experiments/uart/pin_map.lpf

@@ -0,0 +1,33 @@
+BLOCK RESETPATHS;
+BLOCK ASYNCPATHS;
+
+LOCATE COMP "CLK" SITE "G2";
+IOBUF  PORT "CLK" PULLMODE=NONE IO_TYPE=LVCMOS33;
+FREQUENCY PORT "CLK" 150 MHZ;
+
+SYSCONFIG CONFIG_IOVOLTAGE=3.3 COMPRESS_CONFIG=ON MCCLK_FREQ=62 SLAVE_SPI_PORT=DISABLE MASTER_SPI_PORT=ENABLE SLAVE_PARALLEL_PORT=DISABLE;
+
+LOCATE COMP "tx_out" SITE "L4"; # FPGA transmits to ftdi
+LOCATE COMP "rx_in" SITE "M1"; # FPGA receives from ftdi
+IOBUF  PORT "tx_out" PULLMODE=UP IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "rx_in" PULLMODE=UP IO_TYPE=LVCMOS33;
+
+LOCATE COMP "leds[7]" SITE "H3";
+LOCATE COMP "leds[6]" SITE "E1";
+LOCATE COMP "leds[5]" SITE "E2";
+LOCATE COMP "leds[4]" SITE "D1";
+LOCATE COMP "leds[3]" SITE "D2";
+LOCATE COMP "leds[2]" SITE "C1";
+LOCATE COMP "leds[1]" SITE "C2";
+LOCATE COMP "leds[0]" SITE "B2";
+IOBUF  PORT "leds[0]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "leds[1]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "leds[2]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "leds[3]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "leds[4]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "leds[5]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "leds[6]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+IOBUF  PORT "leds[7]" PULLMODE=NONE IO_TYPE=LVCMOS33 DRIVE=4;
+
+LOCATE COMP "RST_N" SITE "B3";
+IOBUF  PORT "RST_N" PULLMODE=DOWN IO_TYPE=LVCMOS33;

flake.nix

@@ -24,7 +24,7 @@
             gotools
             gtkwave
             imagemagick
-            nextpnr
+            nextpnrWithGui
             openfpgaloader
             picocom
             (python3.withPackages (py-pkgs: [

lib/GlitchFilter.bsv

@@ -0,0 +1,30 @@
+package GlitchFilter;
+
+import Cntrs::*;
+
+(* always_enabled="_write",always_ready="_read" *)
+module mkGlitchFilter(Integer hysteresis, bit init_value, Reg#(bit) ifc);
+   Wire#(bit) in <- mkBypassWire();
+   UCount cnt <- mkUCount(init_value == 0 ? 0 : hysteresis, hysteresis);
+   Reg#(bit) out[2] <- mkCReg(2, init_value);
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule incr (cnt.isLessThan(fromInteger(hysteresis)) && in == 1);
+      cnt.incr(1);
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule decr (cnt.isGreaterThan(0) && in == 0);
+      cnt.decr(1);
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule set_out (cnt.isEqual(0) || cnt.isEqual(fromInteger(hysteresis)));
+      out[0] <= cnt.isEqual(0) ? 0 : 1;
+   endrule
+
+   method _write = in._write;
+   method _read = out[1]._read;
+endmodule
+
+endpackage

lib/GlitchFilter_Test.bsv

@@ -0,0 +1,76 @@
+package GlitchFilter_Test;
+
+import Assert::*;
+import StmtFSM::*;
+
+import GlitchFilter::*;
+import Testing::*;
+
+module mkTB();
+   let testflags <- mkTestFlags();
+   let cycles <- mkCycleCounter();
+
+   let dut <- mkGlitchFilter(3, 0);
+
+   // Slight indirection to appease the compiler: GlitchFilter
+   // requires a write on every cycle, and with just the test FSM, the
+   // compiler can't prove that this is satisfied. So, give it a
+   // default value when the test isn't driving the input, on the
+   // first few cycles of the test.
+   Wire#(bit) pin_in <- mkDWire(0);
+   (* no_implicit_conditions,fire_when_enabled *)
+   rule push_in;
+      dut <= pin_in;
+   endrule
+
+   function Action check_dut(bit in, bit want_out);
+      return action
+                if (testflags.verbose)
+                   $display("%0d: GlitchFilter(%0d) => %0d, want %0d", cycles.all, in, dut, want_out);
+                dynamicAssert(dut == want_out, "wrong GlitchFilter output");
+                pin_in <= in;
+             endaction;
+   endfunction
+
+   runTest(100,
+      mkTest("GlitchFilter", seq
+                                // Simple 0->1 transition
+                                check_dut(0, 0);
+                                check_dut(1, 0);
+                                check_dut(1, 0);
+                                check_dut(1, 0);
+                                check_dut(1, 1);
+                                check_dut(1, 1);
+
+                                // Simple 1->0 transition
+                                check_dut(0, 1);
+                                check_dut(0, 1);
+                                check_dut(0, 1);
+                                check_dut(0, 0);
+                                check_dut(0, 0);
+
+                                // Glitchy 0->1
+                                check_dut(1, 0);
+                                check_dut(1, 0);
+                                check_dut(0, 0);
+                                check_dut(1, 0);
+                                check_dut(0, 0);
+                                check_dut(1, 0);
+                                check_dut(1, 0);
+                                check_dut(1, 1);
+                                check_dut(1, 1);
+
+                                // Glitchy 1->0
+                                check_dut(0, 1);
+                                check_dut(0, 1);
+                                check_dut(1, 1);
+                                check_dut(0, 1);
+                                check_dut(1, 1);
+                                check_dut(0, 1);
+                                check_dut(0, 1);
+                                check_dut(0, 0);
+                                check_dut(0, 0);
+                             endseq));
+endmodule
+
+endpackage

lib/PinSync.bsv

@@ -1,11 +1,5 @@
 package PinSync;
 
-(* always_ready, always_enabled *)
-interface PinSync#(type val);
-   method Action _write(val a);
-   method val _read();
-endinterface
-
 // mkPinSync builds a synchronizer for use with asynchronous inputs.
 //
 // You should only use this to capture asynchronous inputs coming from
@@ -24,7 +18,8 @@ endinterface
 // 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, PinSync#(val) ifc)
+(* always_enabled="_write",always_ready="_read" *)
+module mkPinSync(val init_value, Reg#(val) ifc)
    provisos(Bits#(val, _));
 
    Reg#(val) r1 <- mkReg(init_value);

lib/PinSync_Test.bsv

@@ -10,7 +10,7 @@ module mkTB();
    let testflags <- mkTestFlags();
    let cycles <- mkCycleCounter();
 
-   PinSync#(UInt#(2)) dut <- mkPinSync(0);
+   Reg#(UInt#(2)) dut <- mkPinSync(0);
 
    function Action check_dut_val(UInt#(2) want_val);
       return action

lib/Strobe.bsv

@@ -2,6 +2,7 @@ package Strobe;
 
 import Real::*;
 import Printf::*;
+import DReg::*;
 
 // A Strobe provides a synchronization signal to other modules, when
 // an event happens at a cadence other than the module clock.
@@ -9,49 +10,134 @@ import Printf::*;
 interface Strobe;
    method Bool _read();
    // reset resets the strobe cycle, starting with a strobe on the
-   // cycle following reset.
+   // cycle following the reset.
    method Action reset();
 endinterface
 
+module mkStrobeRaw(UInt#(sz) increment, UInt#(sz) max, Strobe ifc);
+   Reg#(UInt#(sz)) cnt <- mkReg(0);
+   PulseWire want_reset <- mkPulseWire();
+   Reg#(Bool) strobe_out <- mkDReg(False);
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule increment;
+      if (want_reset) begin
+         cnt <= 0;
+         strobe_out <= True;
+      end
+      else begin
+         UInt#(TAdd#(sz, 1)) new_cnt = extend(cnt)+extend(increment);
+         if (new_cnt >= extend(max)) begin
+            new_cnt = new_cnt - extend(max);
+            strobe_out <= True;
+         end
+         cnt <= truncate(new_cnt);
+      end
+   endrule
+
+   method Bool _read = strobe_out._read;
+   method Action reset();
+      want_reset.send();
+   endmethod
+endmodule
+
+module mkStrobeRawOld(UInt#(sz) increment, UInt#(sz) max, Strobe ifc);
+   Reg#(UInt#(sz)) cnt[2] <- mkCReg(2, 0);
+   Reg#(Bool) pulse <- mkDReg(False);
+   PulseWire pw <- mkPulseWireOR();
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule increment;
+      UInt#(TAdd#(sz, 1)) new_cnt = extend(cnt[0])+extend(increment);
+      if (new_cnt >= extend(max)) begin
+         cnt[0] <= truncate(new_cnt - extend(max));
+         pulse <= True;
+      end
+      else
+         cnt[0] <= truncate(new_cnt);
+   endrule
+
+   (* no_implicit_conditions, fire_when_enabled *)
+   rule strobe (pulse);
+      pw.send();
+   endrule
+
+   method _read = pw._read;
+   method Action reset();
+      cnt[1] <= increment;
+      pw.send();
+   endmethod
+endmodule
+
+module mkStrobeRawRec(UInt#(n) hack, Integer increment, Integer max, Strobe ifc);
+   let register_width = ceil(log2(fromInteger(max)));
+   if (valueOf(n) < register_width) begin
+      UInt#(TAdd#(n, 1)) hack2 = 0;
+      let _ret <- mkStrobeRawRec(hack2, increment, max);
+      return _ret;
+   end
+   else begin
+      UInt#(n) hack2 = fromInteger(increment);
+      let _ret <- mkStrobeRaw(hack2, fromInteger(max));
+      return _ret;
+   end
+endmodule
+
 // mkStrobe returns a Strobe that triggers at the given
 // target_frequency, assuming mkStrobe is being clocked at the given
 // higher clock_frequency.
 module mkStrobe(Integer clock_frequency, Integer target_frequency, Strobe ifc);
    if (target_frequency > clock_frequency)
       error("mkStrobe target_frequency must be less than clock_frequency");
-   let strobe_every = round(fromInteger(clock_frequency)/fromInteger(target_frequency));
 
-   // Because we're using integer counters to divide frequencies,
-   // unless the clock and target frequencies divide cleanly we'll end
-   // up with a small amount of error.
+   // This is how many main clock ticks there are per strobe at the
+   // target frequency. For all but the simplest cases, this will
+   // include fractional clock cycles due to the two frequencies not
+   // dividing evenly.
+   Real target_ratio = fromInteger(clock_frequency)/fromInteger(target_frequency);
+
+   // At this point we could round the target ratio and instantiate a
+   // counter for that amount, but the resulting strobe frequency can
+   // be quite drastically off target (empirically I've seen up to 5%
+   // without trying).
    //
-   // Strobes like this tend to be used for relatively short
-   // operations before some other synchronization event happens
-   // (e.g. sending one byte on UART), so we can allow a small amount
-   // of frequency error. For now, the target frequency error is fixed
-   // at <=0.1%.
-   Real actual_frequency = fromInteger(clock_frequency)/fromInteger(strobe_every);
-   Real frequency_error_pct = abs(fromInteger(target_frequency)-actual_frequency) / fromInteger(target_frequency) * 100;
-   if (frequency_error_pct > 0.1)
-      error(sprintf("mkStrobe actual frequency is %0f, %0f%% error vs. requested %0d. Your clock_frequency and target_frequency are probably too near each other.", actual_frequency, frequency_error_pct, target_frequency));
+   // We can improve on this by counting in larger increments modulo
+   // some maximum, and strobing on every overflow. This will make the
+   // strobe period vary slightly by the main clock's reckoning,
+   // jittering around the "true" strobe time as error accumulates and
+   // is then subtracted back out.
+   Real target_error = 1/1000; // 0.1%
 
-   Reg#(UInt#(32)) cnt[2] <- mkCReg(2, 0);
+   Integer incr;
+   Integer max;
+   Real actual_error = 1;
+   for (incr=1; incr<=32 && actual_error > target_error; incr=incr+1) begin
+      Real mul_ratio = fromInteger(incr)*target_ratio;
+      max = round(mul_ratio);
+      actual_error = abs(mul_ratio - fromInteger(max))/fromInteger(max);
+   end
 
-   (* no_implicit_conditions, fire_when_enabled *)
-   rule increment;
-      if (cnt[0] == fromInteger(strobe_every-1))
-         cnt[0] <= 0;
-      else
-         cnt[0] <= cnt[0]+1;
-   endrule
+   // Exited the loop, so either we found a good counter width to hit
+   // the requested error, or we maxed out on a 64-bit counter.
+   if (actual_error > target_error)
+      error(sprintf("cannot construct a strobe running at %0dHz from a main clock of %0dHz without exceeding the target frequency error of %0d%%", target_frequency, clock_frequency, target_error*100));
 
-   method Bool _read();
-      return cnt[0] == 0;
-   endmethod
+   // The loop over-incremented by 1 after finding an acceptable error amount.
+   incr = incr-1;
 
-   method Action reset();
-      cnt[1] <= 0;
-   endmethod
+   // Debug messages, uncomment if you're wondering what the logic
+   // above decided for your main clock and target strobe rate.
+   Bool show_debug = True;
+   if (show_debug) begin
+      Real actual_ratio = fromInteger(max)/fromInteger(incr);
+      Real actual_freq_low = fromInteger(clock_frequency)/fromInteger(ceil(actual_ratio));
+      Real actual_freq_high = fromInteger(clock_frequency)/fromInteger(floor(actual_ratio));
+      Real actual_freq_avg = fromInteger(clock_frequency)/actual_ratio;
+      messageM(sprintf("Strobe at %0dHz given clock at %0dHz:\n  count by %0d mod %0d\n  strobing every %d to %d cycles\n  effective strobe frequency %0.2fHz to %0.2fHz\n  avg frequency %0.2fHz (%0.2f%% error)", target_frequency, clock_frequency, incr, max, floor(actual_ratio), ceil(actual_ratio), actual_freq_low, actual_freq_high, actual_freq_avg, actual_error*100));
+   end
+
+   let _ret <- mkStrobeRawRec(UInt#(1) ' (0), incr, max);
+   return _ret;
 endmodule
 
 endpackage

lib/Strobe_Test.bsv

@@ -11,10 +11,12 @@ module mkTB();
    let cycles <- mkCycleCounter();
 
    // For this test, we assume we're clocked at 25MHz, and want a
-   // 115_200bps strobe for a serial port. That translates to a strobe
-   // every 217 cycles.
-   let dut <- mkStrobe(25_000_000, 115_200);
-   let want_pulse_every = 217;
+   // 16x115_200bps strobe for a serial port. That translates to a
+   // strobe that happens every 13 or 14 cycles, depending on the
+   // amount of accumulated error from the imprecise frequency
+   // division.
+   let dut <- mkStrobe(25_000_000, 115_200*16);
+   let want_pulse_every = 14;
 
    function Action check_dut(Bool want);
       return action
@@ -24,29 +26,39 @@ module mkTB();
              endaction;
    endfunction
 
-   function Stmt check_one_cycle();
+   function Stmt check_one_cycle(Integer cycle_len);
       return seq
                 action
+                   $display("%0d: cycle start", cycles.all);
                    check_dut(True);
                    cycles.reset();
                 endaction
-                while (cycles < want_pulse_every)
+                while (cycles < fromInteger(cycle_len))
                    check_dut(False);
              endseq;
    endfunction
 
-   runTest(2000,
+   runTest(500,
       mkTest("Strobe", seq
          dut.reset();
-         repeat(3) check_one_cycle();
+         check_one_cycle(14);
+         check_one_cycle(14);
+         check_one_cycle(13);
+         check_one_cycle(14);
+         check_one_cycle(13);
+         check_one_cycle(14);
+         check_one_cycle(13);
+         check_one_cycle(14);
+         check_one_cycle(14);
+         check_one_cycle(13);
+         check_one_cycle(14);
+         check_one_cycle(13);
 
          // Reset should actually reset
          repeat(10) noAction;
-         par
-            check_dut(False);
-            dut.reset();
-         endpar
-         repeat(3) check_one_cycle();
+         dut.reset();
+         check_one_cycle(14);
+         check_one_cycle(14);
       endseq));
 endmodule
 

tasks.py

@@ -137,7 +137,7 @@ def build(c, target):
     return verilog_files
 
 @task
-def synth(c, target):
+def synth(c, target, gui=False):
     target = resolve_synth_target(target)
 
     out_info, out_verilog, out_bsc, out_yosys, out_nextpnr = ensure_build_dirs(target, "info", "verilog", "bsc", "yosys", "nextpnr")
@@ -192,11 +192,16 @@ def synth(c, target):
         pin_map_arg = f"--lpf {pin_map}"
     else:
         print(f"WARNING: no pin map at {pin_map}, executing place&route with no constraints")
-        pin_map_arg = f"--lpf-allow-unconstrained --lpf=lib/ulx3s_v20.lpf"
+        pin_map_arg = f"--lpf-allow-unconstrained --lpf=lib/ulx3s_v20.lpf --freq 100"
+    nextpnr_out_json = out_nextpnr / f"{module_name.stem}_routed.json"
     nextpnr_out = out_nextpnr / module_name.with_suffix(".pnr")
     nextpnr_log = out_nextpnr / module_name.with_suffix(".log")
     nextpnr_timing = out_nextpnr / module_name.with_suffix(".timing.log")
-    out = c.run(f"nextpnr-ecp5 --85k --detailed-timing-report -l {nextpnr_log} --report {nextpnr_timing} --json {yosys_json}  --package=CABGA381 {pin_map_arg} --speed=6 --textcfg {nextpnr_out}", hide='stderr')
+    if gui:
+        cmd = f"nextpnr-ecp5 --85k --detailed-timing-report --report {nextpnr_timing} --json {yosys_json} --gui --gui-no-aa --package=CABGA381 {pin_map_arg} --speed=6 --textcfg {nextpnr_out}"
+    else:
+        cmd = f"nextpnr-ecp5 --85k --detailed-timing-report -l {nextpnr_log} --report {nextpnr_timing} --json {yosys_json}  --write {nextpnr_out_json} --package=CABGA381 {pin_map_arg} --speed=6 --textcfg {nextpnr_out}"
+    out = c.run(cmd)
 
     print_filtered_paragraphs(out.stderr, "Device utilisation", "Critical path", "Max frequency", "Max delay", common_prefix="Info: ")
     print(f"   PNR : {nextpnr_out}")