slvcodec is a tool that analyzes VHDL and generates:
- Functions to convert arbitrary VHDL types to and from std_logic_vector.
- Generate testbenches for entities that read inputs from a file, and write outputs to a file.
- Utilities so that unit tests for VHDL code can easily to be written in python.
Here's an example VHDL package.
library ieee;
use ieee.numeric_std.all;
package complex is
constant FIXED_WIDTH: natural := 8;
subtype fixed_t is unsigned(FIXED_WIDTH-1 downto 0);
type complex_t is record
real: fixed_t;
imag: fixed_t;
end record;
type array_of_complex is array(natural range <>) of complex_t;
end package;
The following python script is used to generate a helper package that contains functions to convert the types to and from std_logic_vector.
import os
from slvcodec import filetestbench_generator
thisdir = os.path.dirname(__file__)
def make_slvcodec_package():
complex_pkg_fn = os.path.join(thisdir, 'complex_pkg.vhd')
directory = os.path.join(thisdir, 'generated')
os.mkdir(directory)
filetestbench_generator.add_slvcodec_files(directory, [complex_pkg_fn])
if __name__ == '__main__':
make_slvcodec_package()
Here is what the generated VHDL looks like.
library ieee;
use ieee.std_logic_1164.all;
use ieee.numeric_std.all;
use work.complex.all;
use work.slvcodec.all;
package complex_slvcodec is
function to_slvcodec (constant data: array_of_complex) return std_logic_vector;
function from_slvcodec (constant slv: std_logic_vector) return array_of_complex;
constant fixed_t_slvcodecwidth: natural := fixed_width;
constant complex_t_slvcodecwidth: natural := 2*fixed_width;
function to_slvcodec (constant data: complex_t) return std_logic_vector;
function from_slvcodec (constant slv: std_logic_vector) return complex_t;
end package;
package body complex_slvcodec is
function to_slvcodec (constant data: array_of_complex) return std_logic_vector is
constant W: natural := complex_t_slvcodecwidth;
constant N: natural := data'length;
variable slv: std_logic_vector(N*W-1 downto 0);
begin
for ii in 0 to N-1 loop
slv((ii+1)*W-1 downto ii*W) := to_slvcodec(data(ii));
end loop;
return slv;
end function;
function from_slvcodec (constant slv: std_logic_vector) return array_of_complex is
constant W: natural := complex_t_slvcodecwidth;
constant N: natural := slv'length/W;
variable mapped: std_logic_vector(slv'length-1 downto 0);
variable output: array_of_complex(N-1 downto 0);
begin
mapped := slv;
for ii in 0 to N-1 loop
output(ii) := from_slvcodec(mapped((ii+1)*W-1 downto ii*W));
end loop;
return output;
end function;
function to_slvcodec (constant data: complex_t) return std_logic_vector is
constant W0: natural := 0;
constant W1: natural := W0 + fixed_width;
constant W2: natural := W1 + fixed_width;
variable slv: std_logic_vector(complex_t_slvcodecwidth-1 downto 0);
begin
slv(W1-1 downto W0) := to_slvcodec(data.real);
slv(W2-1 downto W1) := to_slvcodec(data.imag);
return slv;
end function;
function from_slvcodec (constant slv: std_logic_vector) return complex_t is
constant W0: natural := 0;
constant W1: natural := W0 + fixed_width;
constant W2: natural := W1 + fixed_width;
variable data: complex_t;
variable mapped: std_logic_vector(complex_t_slvcodecwidth-1 downto 0);
begin
mapped := slv;
data.real := from_slvcodec(mapped(W1-1 downto W0));
data.imag := from_slvcodec(mapped(W2-1 downto W1));
return data;
end function;
end package body;
Here's an example entity that just returns the magnitude squared of a complex data type that we defined earlier.
library ieee;
use ieee.numeric_std.all;
use work.complex.all;
entity complex_mag2 is
port (
i: in complex_t;
o: out unsigned(FIXED_WIDTH+1-1 downto 0)
);
end entity;
architecture arch of complex_mag2 is
signal real2: signed(FIXED_WIDTH*2-1 downto 0);
signal imag2: signed(FIXED_WIDTH*2-1 downto 0);
signal mag2: unsigned(FIXED_WIDTH*2-1 downto 0);
signal scaled_mag2: unsigned(FIXED_WIDTH+1-1 downto 0);
begin
real2 <= i.real * i.real;
imag2 <= i.imag * i.imag;
mag2 <= unsigned(real2) + unsigned(imag2);
scaled_mag2 <= mag2(FIXED_WIDTH*2-1-1 downto FIXED_WIDTH-2);
o <= scaled_mag2;
end architecture;
We can use slvcodec to generate a testbench that reads input data from a file, and writes output data to another file.
import os
from slvcodec import filetestbench_generator
thisdir = os.path.dirname(__file__)
def make_slvcodec_package():
complex_pkg_fn = os.path.join(thisdir, 'complex_pkg.vhd')
directory = os.path.join(thisdir, 'generated')
os.mkdir(directory)
slvcodec_files = filetestbench_generator.add_slvcodec_files(directory, [complex_pkg_fn])
return slvcodec_files
def make_complex_mag2_testbench():
base_filenames = [
os.path.join(thisdir, 'complex_pkg.vhd'),
os.path.join(thisdir, 'complex_mag2.vhd'),
]
slvcodec_fns = make_slvcodec_package()
with_slvcodec_fns = base_filenames + slvcodec_fns
directory = os.path.join(thisdir, 'generated')
generated_fns, generated_wrapper_fns, resolved = filetestbench_generator.prepare_files(
directory=directory, filenames=with_slvcodec_fns,
top_entity='complex_mag2')
return generated_fns
if __name__ == '__main__':
make_complex_mag2_testbench()
This will generate the following VHDL testbench.
library ieee;
use ieee.std_logic_1164.all;
use work.slvcodec.all;
use ieee.numeric_std.all;
use work.complex.all;
use work.complex_slvcodec.all;
entity complex_mag2_tb is
generic (
CLOCK_PERIOD: time := 10 ns;
RUNNER_CFG: string;
OUTPUT_PATH: string
);
end entity;
architecture arch of complex_mag2_tb is
type t_input is
record
i: complex_t;
end record;
type t_output is
record
o: unsigned((1+fixed_width)-1 downto 0);
end record;
constant t_input_slvcodecwidth: natural := 2*fixed_width;
function to_slvcodec (constant data: t_input) return std_logic_vector;
function from_slvcodec (constant slv: std_logic_vector) return t_input;
function to_slvcodec (constant data: t_input) return std_logic_vector is
constant W0: natural := 0;
constant W1: natural := W0 + 2*fixed_width;
variable slv: std_logic_vector(t_input_slvcodecwidth-1 downto 0);
begin
slv(W1-1 downto W0) := to_slvcodec(data.i);
return slv;
end function;
function from_slvcodec (constant slv: std_logic_vector) return t_input is
constant W0: natural := 0;
constant W1: natural := W0 + 2*fixed_width;
variable data: t_input;
variable mapped: std_logic_vector(t_input_slvcodecwidth-1 downto 0);
begin
mapped := slv;
data.i := from_slvcodec(mapped(W1-1 downto W0));
return data;
end function;
constant t_output_slvcodecwidth: natural := (1+fixed_width);
function to_slvcodec (constant data: t_output) return std_logic_vector;
function from_slvcodec (constant slv: std_logic_vector) return t_output;
function to_slvcodec (constant data: t_output) return std_logic_vector is
constant W0: natural := 0;
constant W1: natural := W0 + (1+fixed_width);
variable slv: std_logic_vector(t_output_slvcodecwidth-1 downto 0);
begin
slv(W1-1 downto W0) := to_slvcodec(data.o);
return slv;
end function;
function from_slvcodec (constant slv: std_logic_vector) return t_output is
constant W0: natural := 0;
constant W1: natural := W0 + (1+fixed_width);
variable data: t_output;
variable mapped: std_logic_vector(t_output_slvcodecwidth-1 downto 0);
begin
mapped := slv;
data.o := from_slvcodec(mapped(W1-1 downto W0));
return data;
end function;
signal input_data: t_input;
signal output_data: t_output;
signal input_slv: std_logic_vector(t_input_slvcodecwidth-1 downto 0);
signal output_slv: std_logic_vector(t_output_slvcodecwidth-1 downto 0);
signal clk: std_logic;
signal read_clk: std_logic;
signal write_clk: std_logic;
begin
input_data <= from_slvcodec(input_slv);
output_slv <= to_slvcodec(output_data);
file_reader: entity work.ReadFile
generic map(FILENAME => OUTPUT_PATH & "/indata.dat",
PASSED_RUNNER_CFG => RUNNER_CFG,
WIDTH => t_input_slvcodecwidth)
port map(clk => read_clk,
out_data => input_slv);
file_writer: entity work.WriteFile
generic map(FILENAME => OUTPUT_PATH & "/outdata.dat",
WIDTH => t_output_slvcodecwidth)
port map(clk => write_clk,
in_data => output_slv);
clock_generator: entity work.ClockGenerator
generic map(CLOCK_PERIOD => CLOCK_PERIOD,
CLOCK_OFFSET => 0 ns
)
port map(clk => clk);
read_clock_generator: entity work.ClockGenerator
generic map(CLOCK_PERIOD => CLOCK_PERIOD,
CLOCK_OFFSET => CLOCK_PERIOD/10
)
port map(clk => read_clk);
write_clock_generator: entity work.ClockGenerator
generic map(CLOCK_PERIOD => CLOCK_PERIOD,
CLOCK_OFFSET => 4*CLOCK_PERIOD/10
)
port map(clk => write_clk);
dut: entity work.complex_mag2
port map(
i => input_data.i,
o => output_data.o
);
end architecture;
But generating a test bench that just reads and writes the input and output data to and from files isn't particularly useful unless we have a way of generating the input data, and checking the output data. Slvcodec include tools to do this with python.
We define a python class with a make_input_data
method that returns an iterable of
dictionaries specifying the input data, and a check_output_data
method that receives
a list of input_data dictionaries and a list of output data dictionaries, that raises an
exeception is the output data is incorrect.
class ComplexMag2Test:
def __init__(self, resolved, generics, top_params):
# Here we're taking advantage of the fact that when the test is intialized it
# has access to the parsed VHDL. We use that to get the value of the constant
# FIXED_WIDTH that is defined in complex_pkg.vhd.
self.fixed_width = resolved['packages']['complex'].constants['fixed_width'].value()
self.max_fixed = pow(2, self.fixed_width-1)-1
self.min_fixed = -pow(2, self.fixed_width-1)
self.n_data = 100
def fixed_to_float(self, f):
r = f / pow(2, self.fixed_width-2)
return r
def make_input_data(self, seed=None, n_data=3000):
input_data = [{
'i': {'real': random.randint(self.min_fixed, self.max_fixed),
'imag': random.randint(self.min_fixed, self.max_fixed)},
} for i in range(self.n_data)]
return input_data
def check_output_data(self, input_data, output_data):
inputs = [self.fixed_to_float(d['i']['real']) + self.fixed_to_float(d['i']['imag']) * 1j
for d in input_data]
input_float_mag2s = [abs(v)*abs(v) for v in inputs]
outputs = [self.fixed_to_float(d['o']) for d in output_data]
differences = [abs(expected - actual) for expected, actual in zip(input_float_mag2s, outputs)]
allowed_error = 1/pow(2, self.fixed_width-2)
assert all([d < allowed_error for d in differences])
We then use slvcodec.test_utils.register_test_with_vunit
to generate an appropriate testbench and input
data file, and register the produced test with vunit. VUnit can then be run as normal.
from slvcodec import test_utils, config
import os
if __name__ == '__main__':
random.seed(0)
# Initialize vunit with command line parameters.
vu = config.setup_vunit()
# Set up logging.
config.setup_logging(vu.log_level)
# Get filenames for test
this_dir = os.path.dirname(os.path.realpath(__file__))
filenames = [
os.path.join(this_dir, 'complex_pkg.vhd'),
os.path.join(this_dir, 'complex_mag2.vhd'),
]
# Register the test with VUnit.
test_output_directory = os.path.join(this_dir, 'generated')
test_utils.register_test_with_vunit(
vu=vu,
directory=test_output_directory,
filenames=filenames,
top_entity='complex_mag2',
all_generics=[{}],
test_class=ComplexMag2Test,
top_params={},
)
# Run the tests with VUnit
vu.set_sim_option('disable_ieee_warnings', True)
vu.main()