Ghdl: Bound Check Failure

[ghost]

Description
When I run my simulation, GHDL reports a "bound check failure". I checked the code-line that is reported and to me it looks correct. The line assigns the output of a function to a variable. Both are standard logic vector and have the same width.
Of course there could be a bug inside the function but I would assume that GHDL would report the code-line within the function in that case. I also know that the code is properly working from modelsim simulations.

Context

C:\Program Files (x86)\GHDL\bin\ghdl.EXE:error: bound check failure at D:/git/Libraries/VHDL/psi_fix/testbench/psi_fix_lin_approx_tb/sin18b/psi_fix_lin_approx_sin18b_tb.vhd:77
  from: process psi_fix.psi_fix_lin_approx_sin18b_tb(sim).P0 at psi_fix_lin_approx_sin18b_tb.vhd:77
  from: [unknown caller]
  from: process psi_fix.psi_fix_lin_approx_calc(rtl).p_comb at psi_fix_lin_approx_calc.vhd:131

• OS: Win10 - 64-bit
• Version:
  FailSin18b.zip
  GHDL release: 0.35 (tarball) [Dunoon edition]

How to reproduce?
I will attach a ZIP file containing the error case.

[diogratia]

ghdl -r --std=08 psi_fix_lin_approx_sin18b_tb
./psi_fix_lin_approx_sin18b_tb:error: bound check failure at psi_fix_lin_approx_sin18b_tb.vhdl:76
./psi_fix_lin_approx_sin18b_tb:error: simulation failed

Where line 76 is:

InData <= std_logic_vector(to_unsigned(SigIn(0), InData'length));

Replace that with a process:

    -- InData <= std_logic_vector(to_unsigned(SigIn(0), InData'length));
    process (SigIn)
        variable  intermed: unsigned(InData'range);
    begin
        report "InData'length = " & integer'image(InData'length);
        report "intermed'range = (" & integer'image(intermed'left) &
        " downto " & integer'image(intermed'right) & ")";
        intermed := to_unsigned(SigIn(0), InData'length);  -- this is line 83
        InData <= std_logic_vector(intermed);
    end process;

And produces:

ghdl -a --std=08 psi_fix_lin_approx_sin18b_tb.vhdl
ghdl -e --std=08 psi_fix_lin_approx_sin18b_tb          
ghdl -r --std=08 psi_fix_lin_approx_sin18b_tb
psi_fix_lin_approx_sin18b_tb.vhdl:80:9:@0ms:(report note): InData'length = 20
psi_fix_lin_approx_sin18b_tb.vhdl:81:9:@0ms:(report note): intermed'range = (19 downto 0)
./psi_fix_lin_approx_sin18b_tb:error: bound check failure at psi_fix_lin_approx_sin18b_tb.vhdl:83
./psi_fix_lin_approx_sin18b_tb:error: simulation failed

Where we can see the variable assignment failed.

I'll cut to the chase. The to_unsigned conversion failed because SigIn(0) (type integer) is not contained in the natural range of arg in package declaration numeric_std:

  -- Id: D.3
  function TO_UNSIGNED (ARG, SIZE : NATURAL) return UNRESOLVED_UNSIGNED;
  -- Result subtype: UNRESOLVED_UNSIGNED(SIZE-1 downto 0)
  -- Result: Converts a nonnegative INTEGER to an UNRESOLVED_UNSIGNED vector with
  --         the specified SIZE.

SigIn(0) has a default value of integer'left which is not a natural range value, it's instead the maximum negative integer value.

---

Restoring the assignment to InData as original, providing a default value for SigIn:

    signal SigIn                    : TextfileData_t(0 to 0) := (0 => 0);

and dealing with the generic:

         StimuliDir_g       : string        := "."

And your testbench runs to completion:

ghdl -a --std=08 psi_fix_lin_approx_sin18b_tb.vhdl
ghdl -e --std=08 psi_fix_lin_approx_sin18b_tb
ghdl -r --std=08 psi_fix_lin_approx_sin18b_tb
../../src/ieee2008/numeric_std-body.vhdl:3060:7:@0ms:(assertion warning): NUMERIC_STD.TO_INTEGER: metavalue detected, returning 0
../../src/ieee2008/numeric_std-body.vhdl:3034:7:@5ns:(assertion warning): NUMERIC_STD.TO_INTEGER: metavalue detected, returning 0
../../src/ieee2008/numeric_std-body.vhdl:3060:7:@5ns:(assertion warning): NUMERIC_STD.TO_INTEGER: metavalue detected, returning 0
../../src/ieee2008/numeric_std-body.vhdl:3034:7:@15ns:(assertion warning): NUMERIC_STD.TO_INTEGER: metavalue detected, returning 0

Where we see there are other occurrences of missing initialization.

[ghost]

Thank you for the help. This solved the issue.

Wouldn't it be very useful if the error message provided a bit more information such as:

• Simulation time the error occurs
• Expected range for the bound-check that failed
• Actual value for the bound-check that failed

This would allow users to find their errors easier (and hence reduce the amount of support work as required for this case... sorry :-|).

[abyszuk]

I've just run into very similar issue.
It would be very helpful if GHDL could output more information just like ghost proposed

[tgingold]

Do you have a test case ? I think the initial case has now more information.