OPC6 - a simple DIY cpu now running on BlackIce!

ken.boak 2017-08-25 10:06:00 UTC #1

Thanks to some remarkably fast work by our colleagues over on the anycpu forum (Dave, Ed and Richard) - we now have a port of their OPC6 cpu running on the BlackIce board.

We now have a soft cpu running on BlackIce, clocked at 25MHz that can execute code and run real programs.

There is a monitor program preloaded into RAM, and also a program to calculate Pi to a few digits.

This now gives us a repertoire of soft processors and machines that may be implemented on the BlackIce board. Elsewhere in this forum you will find Acorn Atom (6502 + BASIC) BBC Model B (6502 + BASIC), Jupiter Ace (Z80 with FORTH), Generic CP/M (Z80 with Microsoft BASIC).

The OPC6 offers an alternative to the vintage EDSAC processor that some forum members have been working on - and got running a couple of weeks ago.

There is also James Bowman's J1a Forth processor that has been ported to BlackIce.

With just a serial terminal we have interaction with another soft processor now a fairly comprehensive, but hardware efficient 16-bit register based machine.

The resources used in the ICE40 HX4K FPGA were as follows

After packing:
IOs 50 / 107
GBs 0 / 8
GB_IOs 0 / 8
LCs 822 / 7680 (about 10.7%)
DFF 173
CARRY 94
CARRY, DFF 0
DFF PASS 69
CARRY PASS 5
BRAMs 18 / 32
WARMBOOTs 0 / 1
PLLs 0 / 2

After placement:
PIOs 39 / 107
PLBs 179 / 960
BRAMs 18 / 32

And the timing estimate: 24.38 ns (41.02 MHz)

Not only have the guys got it working - but Dave (Hoglet) has done some exceptional detective work on some aspects of our platform that were causing some trouble - highlighting a conflict between out PLL pins and a couple used for the external RAM databus. In addition - he has found the root cause of the issues that we were having with the dfu-util when trying to update firmware - Cheers Dave!

Note that you will need to upgrade the BlackIce firmware to version 0.3 - as it uses the CH340 UART-USB connection - direct to the FPGA - and needs the STM32 uart lines set both as input pins.

OPC6 is a fairly compact cpu - and this is the first time that it has been tried on a Lattice ICE 40 (previously Xilinx FPGA & CPLD).

It is described on the project pages here https://revaldinho.github.io/opc/ - and OPC6 is the sixth major evolution of the processor. It feature 16 registers, one of which is the PC - and a 3 bit instruction predicate, allowing conditional execution of all alu operations.

Here are some notes from Dave about the cpu implementation

The verilog is in my fork of the OPC github project:

github.com

hoglet67/opc/blob/blackice/system/blackice/system.v

// `define use_pll

module system (
               input         clk100,
               output        led1,
               output        led2,
               output        led3,
               output        led4,
               input         sw1_1,
               input         sw1_2,
               input         sw2_1,
               input         sw2_2,
               input         sw3,
               input         sw4,
               output        RAMWE_b,
               output        RAMOE_b,
               output        RAMCS_b,
               output [17:0] ADR,
               inout [15:0]  DAT,
               input         rxd,

This file has been truncated. show original

github.com

hoglet67/opc/blob/blackice/system/blackice/ram_4k_16.v

module ram ( input [15:0] din, output [15:0] dout, input[11:0] address, input rnw, input clk, input cs_b);

 parameter MEM_INIT_FILE = "monitor_syn.mem";

   reg [15:0] ram [0:4095];
   reg [15:0] dout;

   initial begin
      if (MEM_INIT_FILE != "") begin
         $readmemh(MEM_INIT_FILE, ram);
      end
   end

   always @(posedge clk)
     if (!cs_b) begin
        if (!rnw)
          ram[address] <= din;
        dout <= ram[address];
     end
endmodule

github.com

hoglet67/opc/blob/blackice/system/src/uart.v

module uart ( input[15:0] din, output[15:0] dout, input a0, input rnw, input clk, input reset_b, input cs_b, input rxd, output txd);

   // CLKSPEED is the main clock speed
   parameter CLKSPEED = 32000000;

   // BAUD is the desired serial baud rate
   parameter BAUD = 115200;

   // DIVISOR is the number of clk cycles per bit time
   parameter DIVISOR = CLKSPEED / BAUD;

   // Registers
   reg [15:0] rx_bit_cnt;
   reg [15:0] tx_bit_cnt;
   reg [10:0] tx_shift_reg;
   reg [9:0]  rx_shift_reg;
   reg        rxd1;
   reg        rxd2;
   reg        rx_full;
   reg [7:0]  rx_data;

This file has been truncated. show original

github.com

hoglet67/opc/blob/blackice/opc6/opc6cpu.v

module opc6cpu(input[15:0] din,input clk,input reset_b,input[1:0] int_b,input clken,output vpa,output vda,output vio,output[15:0] dout,output[15:0] address,output rnw);
    parameter MOV=5'h0,AND=5'h1,OR=5'h2,XOR=5'h3,ADD=5'h4,ADC=5'h5,STO=5'h6,LD=5'h7,ROR=5'h8,JSR=5'h9,SUB=5'hA,SBC=5'hB,INC=5'hC,LSR=5'hD,DEC=5'hE,ASR=5'hF;
    parameter HLT=5'h10,BSWP=5'h11,PPSR=5'h12,GPSR=5'h13,RTI=5'h14,NOT=5'h15,OUT=5'h16,IN=5'h17,PUSH=5'h18,POP=5'h19,CMP=5'h1A,CMPC=5'h1B;    
    parameter FET0=3'h0,FET1=3'h1,EAD=3'h2,RDM=3'h3,EXEC=3'h4,WRM=3'h5,INT=3'h6;
    parameter EI=3,S=2,C=1,Z=0,P0=15,P1=14,P2=13,IRLEN=12,IRLD=16,IRSTO=17,IRNPRED=18,IRWBK=19,INT_VECTOR0=16'h0002,INT_VECTOR1=16'h0004;
    reg [15:0] OR_q,PC_q,PCI_q,result;
    reg [19:0] IR_q; (* RAM_STYLE="DISTRIBUTED" *)
    reg [15:0] RF_q[15:0];
    reg [2:0]  FSM_q;
    reg [3:0]  swiid,PSRI_q;
    reg [7:0]  PSR_q ;
    reg        zero,carry,sign,enable_int,reset_s0_b,reset_s1_b,pred_q;
    wire [4:0]  op            = {IR_q[IRNPRED],IR_q[11:8]};
    wire [4:0]  op_d          = { (din[15:13]==3'b001),din[11:8] };
    wire        pred_d        = (din[15:13]==3'b001) || (din[P2] ^ (din[P1] ? (din[P0] ? sign : zero): (din[P0] ? carry : 1))); // New data,new flags (in exec/fetch)
    wire        pred_din      = (din[15:13]==3'b001) || (din[P2] ^ (din[P1]?(din[P0]?PSR_q[S]:PSR_q[Z]):(din[P0]?PSR_q[C]:1))); // New data,old flags (in fetch0)
    wire [15:0] RF_w_p2       = (IR_q[7:4]==4'hF) ? PC_q: {16{(IR_q[7:4]!=4'h0)}} & RF_q[IR_q[7:4]];                          // Port 2 always reads source reg
    wire [15:0] RF_dout       = (IR_q[3:0]==4'hF) ? PC_q: {16{(IR_q[3:0]!=4'h0)}} & RF_q[IR_q[3:0]];                          // Port 1 always reads dest reg
    wire [15:0] operand       = (IR_q[IRLEN]||IR_q[IRLD]||(op==INC)||(op==DEC)||(IR_q[IRWBK]))?OR_q:RF_w_p2;     // One word instructions operand usu comes from RF
    assign {rnw,dout,address} = {!(FSM_q==WRM), RF_w_p2,(FSM_q==WRM||FSM_q==RDM)? ((op==POP)? RF_dout: OR_q)  : PC_q};

This file has been truncated. show original

The latest .pcf file is here:

github.com

hoglet67/opc/blob/blackice/system/blackice/blackice.pcf

# User Constraint File for BlackIce OPC6

# Onboard 100MHz oscillator
set_io clk100 129

# LEDs (active high)
set_io led1  71
set_io led2  67
set_io led3  68
set_io led4  70

# Switches
set_io sw1_2 37 # DIG19
set_io sw1_1 38 # DIG18
set_io sw2_2 39 # DIG17
set_io sw2_1 41 # DIG16
set_io sw3   63 # B1 (active low)
set_io sw4   64 # B2 (active low)

# PMOD UART connections

This file has been truncated. show original

Attached is the latest system.bin file (the bitstream file for the FPGA). - ( I have put this in my github repository https://github.com/monsonite/BackIce_FPGA - if anyone just want's to just load it up and play)

This uses the USB UART, so need to be used with IceBoot 0.3. Using with a earlier IceBoot may result is smoke!

You should be able to build this yourself (assuming you have a Linux machine with IceStorm tools installed):
- git clone -b blackice https://github.com/hoglet67/opc.git
- cd opc/system/blackice/opc6
- make clobber build

There is a simple Pi program included at address 0200.

To execute this, you type 0200g

We'd be happy to write any other hello world program you choose. We could also map some of the LEDs into I/O space so they could be flashed from a hello world program. Would that be of use?

The monitor is quite minimal, and not very user friendly.

The commands are:
- x - hex dump from address
- z - disassemble from address
- @ - set the current address
- s - single step instructions from current address
- , - write to the current address, which is auto incremented
# - toggle local echo
- r - register dump
- l - srecord loader (you can cut and paste srecords to load a new program)

,, are 16-bit hex values

BigEd 2017-08-25 15:56:42 UTC #2

This is great! I should note, in case anyone finds the source to the OPC machines a bit unusual, that OPC stands for One Page CPU - we imposed a limit of 66 lines of code for each of our sources. Perhaps the least strained source is Dave's one-page UART.

(It's easy enough to configure the Pi program to produce rather more digits, if that's felt to be a better demo. I think it will go up to 100 and more.)

hoglet 2017-08-26 08:53:31 UTC #3

Thanks for highlighting this work Ken.

It's now been merged back into the master branch of upstream repository, so the correct links are now of the form:

I also have some small progress to report on the Acorn Atom front, but I'll start a new thread for that later today.

Dave

gururise 2017-09-24 21:19:12 UTC #4

make clobber build on revaldiho's repo results in the following build error... any suggestions?

route...
  pass 1, 14 shared.
  pass 2, 3 shared.
  pass 3, 1 shared.
  pass 4, 0 shared.

After routing:
span_4     1891 / 29696
span_12    383 / 5632

  route time 4.90s
write_txt system.txt1...
icetime -d hx8k -P tq144:4k system.txt1
// Reading input .asc file..
// Reading 8k chipdb file..
// Creating timing netlist..
// Timing estimate: 24.38 ns (41.02 MHz)
icebram monitor_syn.mem monitor.mem < system.txt1 > system.txt2
Hexfiles have different number of words! (4096 vs. 0)
make: *** [../../../common/Makefile_ice40.inc:18: system.txt2] Error 1

BigEd 2017-09-24 21:23:34 UTC #5

It's possible you hit what I hit today, trying to build on a Mac. But in the meantime, try
make -B build
to force a full make.

gururise 2017-09-24 22:27:43 UTC #6

Thanks, tried with make -B build however results in the same exact error above.
Noticed that the system.txt2 file is 0 bytes in size, while the system.txt1 file is slightly over 1 meg in size.

BigEd 2017-09-25 05:05:53 UTC #7

I think icebram is rejecting the monitor.mem file for being too short, which causes the system.txt2 file not to be written and therefore have just zero size.

So, have a look to see if monitor.mem is zero size, delete it, and let's see if we can get it created successfully.

If you look back in the log you might see a command like this
xxd -r -p < monitor.hex | dd ibs=2 conv=swab skip=$((16#F000)) count=$((16#1000)) | od -An -w2 -tx2 -v > monitor.mem

which reads monitor.hex and creates monitor.mem - check monitor.hex's size. In my case I see

-rw-r--r--  1 ed  staff  330410 24 Sep 12:35 monitor.hex
-rw-r--r--  1 ed  staff   20480 25 Sep 06:02 monitor.mem

and yesterday I had a zero size monitor.mem because the 'od' command in the above line was failing. I was able to continue by using this command instead:
xxd -r -p < monitor.hex | dd ibs=2 skip=$((16#F000)) count=$((16#1000)) | xxd -p -c2 > monitor.mem

If your monitor.hex looks OK, perhaps you could try the same.

gururise 2017-09-25 15:22:38 UTC #8

Thanks a lot! Turns out I was missing xxd. After installing vim, everything worked!

BigEd 2017-09-25 16:13:18 UTC #9

Let us know what kinds of things you manage to do with OPC6.

There is now a BCPL (cross) compiler:
http://anycpu.org/forum/viewtopic.php?f=3&t=462