While implementing a state machine on VHDL I was wondering how can I set the output / current state initial condition. I read on one of the questions on here.
One of the answers said we do the initialization before the case structure:
process(currentstate, a)
begin
b <= '1';
c <= '1';
case currentstate is
when s1 =>
if (a = '1') then
c <= '0';
end if;
nextstate <= s2;
However doesn't that make us automatically set b<='1' and c<='1' whenever we get into the process? So if we are at a state say A and we are at the conditions of moving to B whenever we enter the process this directly puts b<='1' and c<='1' isn't that true ?
Or does it actually just run once we start the program and then gets bounded in the case structure ?
Also check this link.
In their FSM implementation they did not specify the initial state how does the compiler or FPGA determine the start state ?
The lines you are looking at are not performing initialization.
b <= '1';
c <= '1';
Remember that VHDL is a hardware description language, not a programming language. What those two assignments do is to set a default assignment for those signals, unless something else contradicts these assignments later in the process. You can assign to the same signal several times in one process, and whichever assignment happens last will take priority. This saves having to write code like:
case State is
when s1 =>
a <= '0';
b <= '1';
c <= '1';
when s2 =>
a <= '1';
b <= '0';
c <= '1';
when s2 =>
a <= '1';
b <= '1';
c <= '0';
end case;
It can end up being quite repetitive and error prone to have the same assignments in many states, so default assignments can really tidy it up:
a <= '1';
b <= '1';
c <= '1';
case State is
when s1 =>
a <= '0';
when s2 =>
b <= '0';
when s2 =>
c <= '0';
end case;
The same pattern works for if statements where you don't want to cover every output signal in every logical branch.
If you want an initial state, there are two approaches that may be applicable depending on the scenario. Here you would assert reset at start-up to set the initial state. Note that the case statement is inside a clocked process:
process (clk)
begin
if (rising_edge(clk)) then
if (reset = '1') then
State <= s1;
else
case State is
when s1 =>
State <= s2;
when s2 =>
State <= s1;
end case;
end if;
end if;
end process;
The other option is to define your state signal with an initial value:
signal State : state_type := s1;
I won't go into the pros and cons of using initial values as there are existing questions that explore this.
Hello I have a state machine that reads from BRAM sends data to a compute core and then writes results back in the BRAM after that the address are incremented so that the next item in the bram can be fed to the compute core. The increment is happening erratically in the state machine. I simulated the code(after stripping down unnecessary stuff) and the increment looks fine in simulation. Can some one please help me to figure out whats wrong. Thanks a bunch for all the help
--The bulk of my glue logic to bring it all together.
process(Bus2IP_Clk,ap_rst,ap_done)
variable in_a_addrb : std_logic_vector(9 downto 0);
variable in_b_addrb : std_logic_vector(9 downto 0);
variable out_r_addrb : std_logic_vector(9 downto 0);
begin
if(ap_rst = '1' ) then
gcd_cs <= wait_for_reset;
in_a_addrb := "0000000000";
in_b_addrb := "0000110010";
out_r_addrb := "0001101110";
elsif (Bus2IP_Clk'event and Bus2IP_Clk = '1') then
gcd_cs <= gcd_ns;
end if;
case gcd_cs is
when wait_for_reset =>
intrpt <= '0';
slv_reg6(0) <= '0';
state_tracker <= "00000000000000000000000000000001";
if(ap_rst = '0')then
gcd_ns <= wait_for_bram_ready;
else
gcd_ns <= wait_for_reset;
end if;
when wait_for_bram_ready =>
state_tracker <= "00000000000000000000000000000010";
if(bram_ready = '1')then
web <= "0";
gcd_ns <= load_input_a;
else
gcd_ns <= wait_for_bram_ready;
end if;
when load_input_a =>
state_tracker <= "00000000000000000000000000000011";
addrb <= in_a_addrb;
in_a <= doutb;
gcd_ns <= load_input_b;
when load_input_b =>
state_tracker <= "00000000000000000000000000000100";
addrb <= in_b_addrb;
in_b <= doutb;
gcd_ns <= start_compute;
when start_compute =>
state_tracker <= "00000000000000000000000000000101";
ap_start <= '1';
gcd_ns <= wait_for_done;
when wait_for_done =>
state_tracker <= "00000000000000000000000000000110";
if(ap_done = '1')then
ap_start <= '0';
web <= "1";
addrb <= out_r_addrb;
dinb <= output;
gcd_ns <= increment_addresses_1;
else
gcd_ns <= wait_for_done;
end if;
when increment_addresses_1 =>
web <= "0";
state_tracker <= "00000000000000000000000000000111";
gcd_ns <= increment_addresses;
when increment_addresses =>
state_tracker <= "00000000000000000000000000001000";
if(in_a_addrb = "0000000111") then
gcd_ns <= stop_compute;
else
in_a_addrb := in_a_addrb +'1';
in_b_addrb := in_b_addrb +'1';
out_r_addrb := out_r_addrb + '1';
gcd_ns <= load_input_a;
end if;
when stop_compute =>
state_tracker <= "00000000000000000000000000001001";
gcd_ns <= stop_compute;
slv_reg6(0) <= '1';
intrpt <= '1';
end case;
in_a_addrb_d <= in_a_addrb;
in_b_addrb_d <= in_b_addrb;
out_r_addrb_d <= out_r_addrb;
end process;
All of your additions are done outside of the clock event. In simulation this may "work" because the process only gets simulated when a signal in the sensitivity list changes. The sensitivity list exists mostly to make simulation more computationally efficient. When you try to implement this into real hardware, items outside of the clock event will become combinational logic that "executes" constantly. This is similar to hooking the output of an inverter to itself without a register in between - it doesn't make any sense. Brian Drummond speaks truth when he says that you should convert this to a regular state machine form to avoid these kinds of errors.
State machines with this kind of architecture can work, despite being more difficult to read and use. The important part is that anything that needs to store information (such as the state) needs to be split into a current and a next, with that transition happening on the clock edge. You do this with the explicit state, but the values of the counters are also part of the machine's state data, and need to receive the same treatment. Using a standard form (with everything occurring on the clock edge) means that you don't have to duplicate all the signals/variables for your storage elements, nor figure out which signals are in fact storage elements, making everything easier.
This project is about adding user's custom peripheral core to MicroBlaze project on FPGA board "spartan 6 lx9". Using ISE Design Suite 14.6 and EDK.
My problem is being not enough experienced in writing VHDL code. I'm still getting 1-bit unintentional latches on signals: "data_bits" and "latest_value" from <0> til <15>, even though I have used recommended coding style for signal's assignment. I have set default values, but no success... Assignment of signal in each branch of case statement is not an option, since I want to retain value especially for "data_bits" since this vector is being build from several clock cycles. I'm trying to solve this problem for several days.
My questions are:
How I can fixed latches problem in this finite-state machine design? --Answered
I would like to get feedback on my state-machine design, styling etc. --Answered, but there is new code
Any design advice for staying on one state for several clocks cycles, using counters or there is a better technique? --Still expecting some advice
Initial Source Code:
library IEEE;
use IEEE.STD_LOGIC_1164.ALL;
entity adc_16bit is
port(
clk : in std_logic;
rst : in std_logic;
data_reg_out : out std_logic_vector(31 downto 0);
control_reg : in std_logic_vector(31 downto 0);
SDO : in std_logic;
SCK : out std_logic;
CONV : out std_logic
);
end adc_16bit;
architecture Behavioral of adc_16bit is
type adc_states is (idle, conversation, clocking_low, clocking_high, receiving_bit, update_data);
signal State, NextState : adc_states;
signal data_bits : std_logic_vector(15 downto 0) := (others => '0');
signal latest_value : std_logic_vector(15 downto 0) := (others => '0');
signal conv_cnt : integer range 0 to 501 := 0;
signal clk_cnt : integer range 0 to 14 := 0;
signal bit_cnt : integer range 0 to 17 := 0;
begin
----------------------------------------------------------------------------------------
-- State Machine Register
----------------------------------------------------------------------------------------
StateReg:
process(clk, rst)
begin
if(clk'event and clk = '1') then
if(rst = '0') then
State <= idle;
else
State <= NextState;
end if;
end if;
end process StateReg;
----------------------------------------------------------------------------------------
-- Signals Register
----------------------------------------------------------------------------------------
TimerReg:
process(clk, rst)
begin
if(clk'event and clk = '1') then
--!default
conv_cnt <= conv_cnt;
clk_cnt <= clk_cnt;
bit_cnt <= bit_cnt;
--latest_value <= latest_value;
--data_bits <= data_bits;
case State is
when idle =>
conv_cnt <= 0;
clk_cnt <= 0;
bit_cnt <= 0;
when conversation =>
if(conv_cnt = 501) then
conv_cnt <= 0;
else
conv_cnt <= conv_cnt + 1;
end if;
when clocking_low =>
if(clk_cnt = 14) then
clk_cnt <= 0;
else
clk_cnt <= clk_cnt + 1;
end if;
when clocking_high =>
if(clk_cnt = 14) then
clk_cnt <= 0;
else
clk_cnt <= clk_cnt + 1;
end if;
when receiving_bit =>
if(bit_cnt = 16) then
bit_cnt <= 0;
else
bit_cnt <= bit_cnt + 1;
end if;
when update_data =>
end case;
end if;
end process TimerReg;
----------------------------------------------------------------------------------------
-- FSM Logic
----------------------------------------------------------------------------------------
FSM_Proc:
process(State, control_reg, conv_cnt, clk_cnt, bit_cnt )
begin
case State is
when idle =>
if(control_reg(0) = '1') then
NextState <= conversation;
else
NextState <= idle;
end if;
when conversation =>
if(conv_cnt = 500) then
NextState <= clocking_low;
else
NextState <= conversation;
end if;
when clocking_low =>
if(clk_cnt = 13) then
NextState <= clocking_high;
else
NextState <= clocking_low;
end if;
when clocking_high =>
if(clk_cnt = 13) then
NextState <= receiving_bit;
else
NextState <= clocking_high;
end if;
when receiving_bit =>
if(bit_cnt = 15) then
NextState <= update_data;
else
NextState <= clocking_low;
end if;
when update_data =>
if(control_reg(0) = '1') then
NextState <= conversation;
else
NextState <= idle;
end if;
end case;
end process FSM_Proc;
----------------------------------------------------------------------------------------
-- FSM Output
----------------------------------------------------------------------------------------
FSM_Output:
process(NextState, latest_value, data_bits, bit_cnt, SDO )
begin
--!default
CONV <= '0';
SCK <= '0';
data_reg_out(31 downto 16) <= (others => '0');
data_reg_out(15 downto 0) <= latest_value;
--data_bits <= data_bits;
--latest_value <= latest_value;
case NextState is
when idle =>
latest_value <= (others => '0');
data_bits <= (others => '0');
when conversation =>
CONV <= '1';
when clocking_low =>
SCK <= '0';
when clocking_high =>
SCK <= '1';
when receiving_bit =>
SCK <= '1';
--data_bits <= data_bits;
data_bits(bit_cnt) <= SDO;
when update_data =>
latest_value <= data_bits;
when others =>
--latest_value <= latest_value;
--data_bits <= data_bits;
end case;
end process FSM_Output;
end Behavioral;
EDIT
Thank you for all your responses! I decided to rewrite my FSM on single process and to add more information regarding my problem in order to make it more understandable for others who has similar problems!
Block Diagram of system:
http://i.stack.imgur.com/odCwR.png
Note: that right now I just want to simulate and verify stand alone adc_core itself without MicroBlaze and AXI interconnection block.
FSM Diagram:
http://i.stack.imgur.com/5qFdN.png
Single process source code:
library IEEE;
use IEEE.STD_LOGIC_1164.ALL;
entity adc_chip_driver is
port(
clk : in std_logic;
rst : in std_logic;
data_reg_out : out std_logic_vector(31 downto 0);
control_reg : in std_logic_vector(31 downto 0);
SDO : in std_logic;
SCK : out std_logic;
CONV : out std_logic
);
end adc_chip_driver;
architecture Behavioral of adc_chip_driver is
type states is (idle, conversation, clocking_low, clocking_high, receiving_bit, update_data);
signal state : states;
signal data_bits : std_logic_vector(0 to 15) := (others => '0');
signal latest_value : std_logic_vector(15 downto 0) := (others => '0');
signal conv_cnt : integer range 0 to 500 := 0;
signal clk_cnt : integer range 0 to 13 := 0;
signal bit_cnt : integer range 0 to 15 := 0;
begin
process(clk, rst, control_reg)
begin
if(rst = '0') then
state <= idle;
data_bits <= (others => '0');
latest_value <= (others => '0');
data_reg_out <= (others => '0');
elsif(clk'event and clk = '1') then
--!Default Values
data_reg_out(31 downto 16) <= (others => '0'); --unused bits of register
data_reg_out(15 downto 0) <= latest_value; --data_reg_out is always tided to latast_value;
latest_value <= latest_value; --latest_value is being updated only once
data_bits <= data_bits; --has to retain value
conv_cnt <= conv_cnt;
clk_cnt <= clk_cnt;
bit_cnt <= bit_cnt;
case state is
when idle =>
--signals
conv_cnt <= 0;
clk_cnt <= 0;
bit_cnt <= 0;
--outputs
SCK <= '0';
CONV <= '0';
--logic
if(control_reg(0) = '1') then
state <= conversation;
else
state <= idle;
end if;
when conversation =>
--output
SCK <= '0';
CONV <= '1';
--logic
if(conv_cnt = 500) then
state <= clocking_low;
conv_cnt <= 0;
else
state <= conversation;
conv_cnt <= conv_cnt + 1;
end if;
when clocking_low =>
--ouput
SCK <= '0';
CONV <= '0';
--logic
if(clk_cnt = 13) then
clk_cnt <= 0;
state <= clocking_high;
else
clk_cnt <= clk_cnt + 1;
state <= clocking_low;
end if;
when clocking_high =>
--ouput
SCK <= '1';
CONV <= '0';
--logic
if(clk_cnt = 13) then
clk_cnt <= 0;
state <= receiving_bit;
else
clk_cnt <= clk_cnt + 1;
state <= clocking_high;
end if;
when receiving_bit =>
--signal
data_bits(bit_cnt) <= SDO;
--ouput
SCK <= '1';
CONV <= '0';
--logic
if(bit_cnt = 15) then
bit_cnt <= 0;
state <= update_data;
else
bit_cnt <= bit_cnt + 1;
state <= clocking_low;
end if;
when update_data =>
--signal
latest_value(15 downto 0) <= data_bits(0 to 15);
--ouput
SCK <= '0';
CONV <= '0';
--logic
if(control_reg(0) = '1') then
state <= conversation;
else
state <= idle;
end if;
end case;
end if;
end process;
end Behavioral;
Maybe I could receive some new feedback on single process design?
Also I still do you have unanswered question regarding usage of counters in specific FSM states. I have noticed that usually during second cycle on "clocking_low" and "clocking_high" counter actually starts at 1 instead of 0, I know that in this situation it's not a problem, but I can easily imagine where it could be important. I was thinking about after reset set counters to '-1', but maybe there is better solution?
Your code has a number of problems. To illustrate some of them, I tried to sketch your finite state machine in Figs. 1 and 2 below, based on the VHDL code that you provided.
First and most importantly, the design should begin with a top-level block diagram, showing the circuit ports (as in Fig. 1), followed by a detailed state transition diagram (as in Fig. 2 – incomplete here). Recall, for example, that the circuit outputs (data_reg_out, SCK, and CONV – Fig. 1) are the signals that the FSM is supposed to produce, so it is indispensable that these values be specified in all states (shown inside the state circles in Fig. 2). Once the diagram of Fig. 2 is fixed and completed, writing a corresponding VHDL code should be relatively straightforward (except for the timer - see comments below).
Other problems can be seen directly in the code. Some comments on the four processes follow.
The first process (StateReg), which stores the FSM state, is fine.
The second process (TimerReg) is also registered (under clk’event), which is necessary to build the timer. However, dealing with timers is one of the trickiest parts of any timed FSM, because you MUST devise a correct strategy for stopping/running the timer and also for zeroing it. For this, I suggest that you check reference 1 below, which deals with all possible kinds of FSM implementations from a hardware perspective, including an extensive study of timed FSMs.
The third process (FSM_Proc) defines the next state. It is not registered, which is as it should be. However, to check it, it is necessary to complete first the state transition diagram of Fig. 2.
The last process (FSM_Output) defines the machine outputs. It is not registered, which is as it should be in general. However, the list of outputs is not the same in all states, in spite of the default values. Note, for example, the existence of latest_value and data_bits in state idle, which do not appear in all states, thus causing the inference of latches. Additionally, this process is based on NextState instead of PresentState, which (besides being awkward) might reduce the circuit’s maximum speed.
I hope these comments motivate you to restart from the beginning.
1 V. A. Pedroni, Finite State Machines in Hardware: Theory and Design (with VHDL and SystemVerilog), MIT Press, Dec. 2013.
You get a latch if a signal is not assigned to on all possible paths, as it then becomes stateful.
To avoid the problem, make sure you always assign a value to the signal (one way is to assign a "default" value at the top of the process).
since I want to retain value especially for "data_bits" since this vector is being build from several clock cycles.
"Retaining a value" means state, not purely combinatorial logic. In which case, it should not be in your output process. It should be in your state-update process.
My solution to this has been to always use clocked processes for everything. There is no need to have a separate clocked process for the state register and a separate process for the state transitions. That's a style which was required years ago. In my opinion, you are better off putting everything into a single clocked process and then you cannot get latches.
If you must use two processes then get a VHDL 2008 compiler and use process(all) to ensure that all your signals are correctly in the sensitivity list, and then carefully ensure that every signal you assign to gets an assignment for every logical path through the process. The easiest way to achieve this is often to assign them all some 'default' values at the start of the process.
In a combinational process (like FSM_Output), you should never read a signal and write to the same signal. That is exactly what is going on here, for latest_value and data_bits.
Either create new signals latest_value_r and data_bits_r and copy the values in the clocked process, or stick to a single clocked process with no separate combinational process.
What hardware do you want for data_bits and latest_value? If you want to build a vector over several clock cycles, then you need a storage device. Your choices are: latch (level sensitive storage) and flip-flop (edge sensitive storage). If you don't want latches, then you must code flip-flops.
To code flip-flops use the "if clk='1' and clk'event then", just like you did in TimerReg process. You can alternatively use "if rising_edge(Clk) then" - I like this better for readablity, but the tools don't care either way.
I think where you went wrong is in your planning process. Code is just design capture. What is important is that you plan with a block diagram and know where your design requires flip-flops and where it requires combinational logic. Get this right and the rest is just applying coding templates. So make sure you understand this before you start coding.
It does not matter whether you code with only clocked processes or use a mix of clocked and combinational logic processes. I think the most important thing you do in your coding is make it readable. If you collect opinions, you will see they vary, #Martin and #Brian prefer a single clocked process. I prefer a 2 process statemachine - flip-flop and combinational (present state to next state and ouput decode). You used a 3 process statemachine - for me that is like drawing a bubble diagram to show state transitions and a separate one to show the output transitions. However at the end of the day, they all capture the same intent. As long it is clear to someone reading your code long after you have left, it should be ok.
I'm programming a N-bit non-restoring divider, but I faced a little problem.
I have an Operative Part (combinatorial) and a Control Part (Finite State Machine).
The Control Part has a 2 processes FSM, 1 for updating the next state and 1 for the "state sequence".
update: process(clk_in, next_state)
begin
if rising_edge(clk_in) then
current_state <= next_state;
end if;
end process;
And this is the second process:
control: process(current_state, start, S_in, counted)
variable sub_tmp : STD_LOGIC := '0';
begin
[...]
sub <= sub_tmp; -- sub is an output signal of my entity that goes in the Operative Part
case current_state is
when idle =>
if start='1' then
next_state <= init;
else
next_state <= idle;
end if;
when init =>
-- [...]
next_state <= subtract;
when subtract =>
en_A <= '1';
sub_tmp := '1';
next_state <= test;
when test => -- shift
en_Q <= '1';
if S_in='0' then
sub_tmp := '1';
else
sub_tmp := '0';
end if;
if counted=N/2-1 then
next_state <= finished;
else
next_state <= operation;
end if;
when operation =>
en_A <= '1';
next_state <= test;
when finished =>
stop <= '1';
next_state <= idle;
end case;
end process;
As you can see, I need to change the value of the sub ONLY in 2 cases (subtract and test), while I don't have to change in the other cases.
The problem is that when I try to synthesize this code it turns out that sub_tmp is a LATCH, but I don't want a latch.
I need to do something like this:
state 1 => set sub to '1' or '0' (depending on another input)
state 2 => do other operations (but sub must remain the value set before) and return to state 1
etc...
To clarify more: in certain states of my FSM (not all of them) I set the value of a variable (let's call it sub_tmp). In other states I don't change its value. Then let's say I have an output PIN called "sub_out". Now, independently of the variable value, I want to output its value to this pin (sub_out <= sub_tmp; or similar).
What am I missing?
What you are missing is the behavior you describe IS a latch. Anything with memory (ie: "in other states I don't change it's value") is either a latch or a register (flip-flop). If you don't want a latch or a register, you need to assign a specific value to the signal in each and every code path, and not let it 'remember' it's previous state.
Fellow SO users,
I'm trying to sample my resistive humidity sensor at a frequency of 5Hz (5 samples a second). I'm using an ADC to read the output. Now, I've been told that you can run the ADC at any frequency but you need to use a 5hz clock to initiate the conversion and read values from the ADC.
The way I'm doing this is by having a process that initiates the conversion by running at 5hz and having a flag, say "start_convert" to '1' on the rising edge of the clock.
PROCESS (CLK_5HZ)
BEGIN
IF (CLK_5HZ'EVENT AND CLK_5HZ = '1') THEN
START_CONVERT <= '1';
END IF;
END PROCESS;
And then I have a state machine for the ADC;
PROCESS (CURR_STATE, INTR)
BEGIN
CASE CURR_STATE IS
WHEN STARTUP =>
WR <= '0';
READ_DATA <= '0';
IF (START_CONVERT = '1') THEN
NEXT_STATE <= CONVERT;
ELSE
NEXT_STATE <= STARTUP;
END IF;
WHEN CONVERT =>
IF (INTR = '0' AND STREAM = '1') THEN
NEXT_STATE <= WAIT500;
ELSIF (INTR = '0' AND STREAM = '0') THEN
NEXT_STATE <= READ1;
ELSE
NEXT_STATE <= CONVERT;
END IF;
WR <= '1';
READ_DATA <= '0';
WHEN WAIT10 =>
IF (COUNTER_WAIT = 10) THEN
NEXT_STATE <= READ1;
ELSE
NEXT_STATE <= WAIT10;
END IF;
COUNTER_WAIT <= COUNTER_WAIT + 1;
WHEN READ1 =>
NEXT_STATE <= CONVERT;
WR <= '1';
READ_DATA <= '1';
WHEN OTHERS =>
NEXT_STATE <= STARTUP;
END CASE;
END PROCESS;
And then I'm using another process at 5hz to detect whenever READ_DATA is 1 so that I read the values from the ADC.
PROCESS (CLK_5HZ, RST)
BEGIN
IF (RST = '1') THEN
Y <= (OTHERS => '0');
ELSIF (CLK_5HZ'EVENT AND CLK_5HZ = '1') THEN
IF (READ_DATA = '1') THEN
Y <= DATA_IN (0) & DATA_IN (1) &
DATA_IN (2) & DATA_IN (3) &
DATA_IN (4) & DATA_IN (5) &
DATA_IN (6) & DATA_IN (7);
END IF;
END IF;
END PROCESS;
Could anyone please advice me whether this is the right approach or not?
EDIT: I'm interfacing the ADC (ADC0804) using a Spartan-3 board.
It's a bit hard to give specific advice, when not knowing the specifics of the device you are trying to interface to.
However, a couple of general comments regarding your code:
Your asynchronous process (the PROCESS (CURR_STATE, INTR)) will generate quite a few latches when synthesizing it, since you are not setting all your signals in all cases. WR and READ_DATA are for instance not being set in your WAIT10 state. This will most probably lead to severe timing issues, so correcting this is something you'd absolutely want to do.
The WAIT10 state in the same process will give you a combinatorial loop, as it runs whenever COUNTER_WAIT is updated. As that state also updates COUNTER_WAIT, it will in theory just keep running, but in practice most synthesizers will just give you an error (I think). You'll need to move the incrementing to a synchronous/clocked process instead, which will also give you control over how long each cycle takes.
Your 5 Hz processes seem to run on a separate clock (CLK_5HZ). I presume that the rest of your system is running at a faster clock? This essentially gives you two (or more) clock domains, that will need special interface logic for interfacing them with each other. A much, much better solution is to run everything on the same (fast) clock, and control slower processes using clock enables. Everything is thus inherently synchronized, and shouldn't give you any nasty timing surprises.