Overflow in exponentiation

[erick166]

Description
I get an overflow error and I wonder if it's a GHDL or the code problem.

How to reproduce?
The testbench below has 3 scenarios:
First, using the attribute to print the maximum integer positive value which works okay.
Then, using exponentiation, which works for the to_string() procedure but not for the other 2 cases, which give the following error:

ghdl:error: overflow in exponentiation
ghdl:error: simulation failed
Error: Program ended with exit code 1

Finally, using exponentiation but with a hardcoded value, which gives the following errors:

example_tb.vhd:24:46:warning: arithmetic overflow in static expression [-Wruntime-error]
report "Using 'image:" & integer'image(2 ** 31 - 1);
^
example_tb.vhd:25:21:warning: arithmetic overflow in static expression [-Wruntime-error]
v_integer := (2 ** 31) - 1;       
^      

library ieee;
use ieee.std_logic_1164.all;
use ieee.numeric_std.all;

entity example_tb is
end entity;

architecture rtl of example_tb is
begin

  p_main : process
    variable v_exponent : integer := 31;
    variable v_integer  : integer;
  begin
    report "Using to_string:" & to_string(integer'right);
    report "Using 'image:" & integer'image(integer'right);
    v_integer := integer'right;

    report "Using to_string:" & to_string(2 ** v_exponent - 1);
    report "Using 'image:" & integer'image(2 ** v_exponent - 1);
    v_integer := (2 ** v_exponent) - 1;

    --report "Using to_string:" & to_string(2 ** 31 - 1);
    --report "Using 'image:" & integer'image(2 ** 31 - 1);
    --v_integer := (2 ** 31) - 1;

    wait;
  end process p_main;

end architecture rtl;

Context

• OS: Windows 11
• GHDL 5.0.0-dev (4.0.0.r24.g14b411c57)
• UVVM v2 2024.03.08

[tgingold]

I think this is correct. 2**31 is beyond the bounds of integer.

[erick166]

But since I'm actually using (2 ** 31 - 1) perhaps it should be okay? I just wonder because if it's used inside the to_string() function it works fine.

[tgingold]

So 2**31 is out of the integer range, that's why you get overflow in most cases.

For to_integer(2**v_exponent - 1), there are many possible types, and according to vhdl it should be universal_integer, which is a 64b type in ghdl. That's why there is no overflow in that case.

[erick166]

Right, I just thought since I'm actually using (2 ** 31 - 1) and not 2 ** 31 as the integer value perhaps it could work, since it works when using it inside the to_string() function, but if it's not the case then I'll just use a workaround with integer'right.

[nselvara]

I guess the ** operator is evaluated first as it has higher precedence which then leads to overflow of the value.
You can use integer'high which should return you always the highest value.