Reputation: 199
I want to write a module for GCD computing, using extended Euclidean algorithm. But the main problem is that I completely don't know how to do that without getting to the lowest (RTL) level. What I mean is to have FSM with three states:
However, when I try to separate FSM & computations into separate processes, like this:
module modinv(clk, reset, number, prime, finished, gcd, inverse_fail, inverse);
input [31:0] number, prime;
input wire clk, reset;
output integer gcd, inverse;
output reg finished, inverse_fail;
parameter [2:0] IDLE = 3'b001, COMPUTING = 3'b010, END = 3'b100;
reg [2:0] state, state_next;
integer a, b, c, q, p, r;
always @ (posedge clk, posedge reset)
begin
if (reset == 1)
begin
state <= IDLE;
end
else
begin
state <= state_next;
end
end
always @(state or b)
begin
finished <= 0;
inverse_fail <= 0;
case (state)
IDLE:
begin
a <= number;
b <= prime;
p <= 1;
r <= 0;
state_next <= COMPUTING;
end
COMPUTING:
begin
c = a % b;
q = a / b;
a = b;
b = c;
r = p - q * r;
p = r;
if (b == 0)
begin
state_next <= END;
end
else
begin
state_next <= COMPUTING;
end
end
END:
begin
gcd <= a;
inverse <= p;
finished <= 1;
if (gcd != 1)
begin
inverse_fail <= 1;
end
end
endcase
end
endmodule
And when I try to put computation in the second process, in COMPUTING state case, it only works once - what is correct in means of verilog, because until computing is done, state doesn't change, so the process isn't called again.
However, when I put computation in the first process, there isn't any non-ugly way to limit the computations only to correct STATE, which results in wrong output (as soon as FSM is in FINISHED state, the output is already incorrect - one step further).
So, my question is - how to do it correctly without using FSM + datapath (low-level RTL) solution?
My current waveform:
Upvotes: 3
Views: 7418
Reputation: 35933
You seem to be missing some clocked elements in your design.
From what I understand of your design, you seem to expect once the state goes to COMPUTING, it should keep iterating the values of a
and b
until b
reaches 0. But the only thing you're actually clocking on a clock edge is the state variable, so there's no memory of a and b from one state to the next. If you want variables like a
and b
to have memory from one clock cycle to the next, then you need to latch these variables as well:
I made some modifications to your program, it might not be 100% correct, but you should see what I'm getting at. See if this makes sense how you do the combinational logic in the second block, but you register the values on the posedge so that you can use them at the start of the next clock cycle.
module modinv(clk, reset, number, prime, finished, gcd, inverse_fail, inverse);
input [31:0] number, prime;
input wire clk, reset;
output integer gcd, inverse;
output reg finished, inverse_fail;
parameter [2:0] IDLE = 3'b001, COMPUTING = 3'b010, END = 3'b100;
reg [2:0] state, state_next;
integer a, b, c, q, p, r;
integer a_next, b_next, p_next, r_next;
always @ (posedge clk, posedge reset)
begin
if (reset == 1)
begin
state <= IDLE;
a <= 0;
b <= 0;
p <= 0;
r <= 0;
end
else
begin
state <= state_next;
a <= a_next;
b <= b_next;
p <= p_next;
r <= r_next;
end
end
always @* //just use the auto-triggered '@*' operator
begin
finished <= 0;
inverse_fail <= 0;
case (state)
IDLE:
begin
a_next <= number;
b_next <= prime;
p_next <= 1;
r_next <= 0;
state_next <= COMPUTING;
end
COMPUTING:
begin
c = a % b;
q = a / b;
a_next = b;
b_next = c;
r_next = p - q * r;
p_next = r;
if (b == 0)
begin
state_next <= END;
end
else
begin
state_next <= COMPUTING;
end
end
END:
begin
gcd <= a;
inverse <= p;
finished <= 1;
if (gcd != 1)
begin
inverse_fail <= 1;
end
end
endcase
end
endmodule
Upvotes: 3