Multidimensional array problem in VHDL?

Question

I'm trying to use a multidimensional array in VHDL and I'm having a lot of trouble getting it to work properly. My issue is that I've got an array of 17, of 16 vectors, of a given size. What I want to do is create 17 registers that are array of 16 * std_logic_vector of 32 bits (which = my b, 512). So, I'm trying to pass in something to input and output on the register instantiation that tells the compiler/synthesizer that I want to pass in something that is 512 bits worth... Similar to in C if I had:

int var[COLS][ROWS][ELEMENTS];
memcpy(&var[3].. // I'm talking about 3rd COL here, passing in memory that is ROWS*ELEMENTS long

(My actual declaration is here:)

type partial_pipeline_registers_type is array (0 to 16, 0 to 15) of std_logic_vector(iw - 1 downto 0); signal h_blk_pipelined_input : partial_pipeline_registers_type;

I tried simply using h_blk_pipelined_input(0) .. up to (16) but this doesn't work. I get the following error, which makes me see that I need to double index in to the array:

ERROR:HDLParsers:821 - (at the register) Wrong index type for h_blk_pipelined_input.

So then I tried what's below, and I get this error:

ERROR:HDLParsers:164 - (at the register code). parse error, unexpected TO, expecting COMMA or CLOSEPAR


  instantiate_h_pipelined_reg : regn
   generic map ( N=> b, init => bzeros )
   port map ( clk => clk , rst => '0', en => '1',
      input => h_blk_pipelined_input((i - 1), 0 to 15),
      output=> h_blk_pipelined_input((i),     0 to 15));
-- Changing 0 to 15 to (0 to 15) has no effect...

I'm using XST, and from their documentation (http://www.xilinx.com/itp/xilinx6/books/data/docs/xst/xst0067_9.html), the above should have worked:

...declaration:

subtype MATRIX15 is array(4 downto 0, 2 downto 0)
   of STD_LOGIC_VECTOR (7 downto 0);

 A multi-dimensional array signal or variable can be completely used:

Just a slice of one row can be specified:

MATRIX15 (4,4 downto 1) <= TAB_B (3 downto 0);

One alternative is that I can create more registers that are 16 times smaller, and instead of trying to do all '0 to 15' at once, I would just do that 15 additional times. However, I think this may lead to inefficiency in synthesis and I don't feel like this is the right solution.

EDIT:

Tried what Ben said,

instantiate_h_m_qa_pipeline_registers: for i in 1 to 16 generate

instantiate_h_pipelined_reg : regn
    generic map ( N=> b, init => bzeros )
    port map ( clk => clk , rst => '0', en => '1',
       input => h_blk_pipelined_input(i - 1),                                            
       output=> h_blk_pipelined_input(i));                                          
end generate instantiate_h_m_qa_pipeline_registers;

The signals are now defined as:

type std_logic_block is array (0 to 15) of std_logic_vector(iw - 1 downto 0) ;
type partial_pipeline_registers_type is array (0 to 16) of std_logic_block;
signal h_blk_pipelined_input : partial_pipeline_registers_type;

And the error I get from XST is:

ERROR:HDLParsers:800 - ((where the register part is)) Type of input is incompatible with type of h_blk_pipelined_input.

I'm able to do everything I was able to do before, using ()() syntax instead of ( , ) so I haven't lost anything going this way, but it still doesn't resolve my problem.

EDIT:

Took it a step further and used a conversion function on the input => and output => arguments for the register instantiation, using a function to convert between the array type and just a std_logic_vector the size that I need. It fixed the 'input => ' part, but..

Actual associated with Formal OUT mode Signal 'output' may not be a type conversion or function call. (LRM 4.3.2.2)

Ben Voigt · Accepted Answer · 2011-02-26 15:20:06Z

3

How about:

TYPE reg512_type IS ARRAY(0 TO 15) OF STD_LOGIC_VECTOR (31 DOWNTO 0);
TYPE partial_pipeline_registers_type IS ARRAY(0 TO 16) OF reg512_type;

The code on that Xilinx site obviously isn't tested. Since ARRAY OF ... are types, not subtypes, their code shouldn't compile.

EDIT: These custom types aren't going to play well with existing IP components, but that's ok, the compiler can infer registers easily enough. Try:

instantiate_h_m_qa_pipeline_registers: FOR i IN 1 TO 16 GENERATE

instantiate_h_pipelined_reg : PROCESS (clk)
BEGIN
    IF RISING_EDGE(clk) THEN
        h_blk_pipelined_input(i) <= h_blk_pipelined_input(i - 1);
    END IF;
END GENERATE instantiate_h_m_qa_pipeline_registers;

edited Feb 26, 2011 at 15:20

answered Feb 26, 2011 at 5:43

Ben Voigt

286k45 gold badges444 silver badges764 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Nektarios Over a year ago

I think we're getting close here.. see my question edits at bottom of question for errors I'm getting now

Ben Voigt Over a year ago

@Nektarios: VHDL is very picky about types. Even types which are compatible, like UNSIGNED and STD_LOGIC_VECTOR, can't be used interchangeably. But you can cast between them (casting in VHDL looks like a C++ constructor-style cast). Or you can change the data type of the component parameters to reg512_type. To do that, you'll have to move the type declarations to a library module so they're visible to multiple components.

Nektarios Over a year ago

so, my register is defined with "input : in std_logic_vector(N-1 downto 0);" .. are you saying its possible that when I instantiate one of these registers and set up " input => h_blk.... " I can somehow recast my h_blk_pipelined_input to be seen as std_logic_vector(size) instead of an array of smaller std_logic_vectors? How would I do that exactly? Thanks

Nektarios · Accepted Answer · 2011-02-26 17:48:32Z

1

To get things to play right, I had to use a function to flatten and unflatten my data interacting with the register instantiation.

The real issue boiled down to me having an array of std_logic_vector() that resulted in a large vector which the register instantiation needed to have.. and then the register was giving me back the vector, which I needed to put in to an array of vectors.

Creating a function and then doing the following is working great.

      input => blk2vec(h_blk_pipelined_input(i - 1)),                        
      vec2blk(output) => h_blk_pipelined_input(i));

answered Feb 26, 2011 at 17:48

Nektarios

10.4k10 gold badges66 silver badges93 bronze badges

2 Comments

Nektarios Over a year ago

Unfortunately, I found that this doesn't synthesize.

Nektarios Over a year ago

But, using another intermediate signal, and then using a vec2blk type function there, does work

Collectives™ on Stack Overflow

Multidimensional array problem in VHDL?

2 Answers 2

3 Comments

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

3 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related