Can Vivado handle user defined physical types?
Asked Answered
K

2

0

I wrote some cross platform VHDL libraries for Xilinx XST, iSim, Altera Quartus II, Mentor Graphics QuestaSim and GHDL. Now I wanted to port my ISE 14.7 project, which uses these libraries to Vivado 2014.4, but one library seems to have fatal problems.

My library physical defines several new user defined physical types like: FREQUENCY and BAUD; conversion functions and report functions.

One main use case is the calculation of delay or counter cycles for a given delay and system frequency. So for example a delay of 125 ns requires 12 or 13 delay cycles at 100 MHz (it depends on the rounding mode at .5).

I get several infos and warnings from Vivado Synth (some are results of an assert statement, see minimal example below):

[Synth 8-638] synthesizing module 'Top_PhysicalTest' ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":410]
[Synth 8-63] RTL assertion: "to_time: f= 2147483647.1000 THz  return 2147483647.1000 sec" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":277]
[Synth 8-63] RTL assertion: "res_real: 0.000000" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":321]
[Synth 8-63] RTL assertion: "TimingToCycles: 
  Timing: 2147483647.1000 sec
  Clock_Period: 2147483647.1000 sec
  RoundingStyle: TO_NEAREST
  res_real = 2147483647.1000
  => 0" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":323]
[Synth 8-26] 'image of non-integer, non-enum type not implemented ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":422]
[Synth 8-63] RTL assertion: "CLOCK_FREQ: <complex-type>" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":422]
[Synth 8-63] RTL assertion: "CLOCK_FREQ: 2147483647.1000 THz" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":423]
[Synth 8-26] 'image of non-integer, non-enum type not implemented ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":424]
[Synth 8-63] RTL assertion: "DELAY: <complex-type>" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":424]
[Synth 8-63] RTL assertion: "DELAY: 2147483647.1000 sec" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":425]
[Synth 8-63] RTL assertion: "CYCLES: 0" ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":426]
[Synth 8-256] done synthesizing module 'Top_PhysicalTest' (1#1) ["D:/Temp/PhysicalTest_Vivado2014.4/vhdl/Top_PhysicalTest.vhd":410]
[Synth 8-3330] design Top_PhysicalTest has an empty top module
[Synth 8-3331] design Top_PhysicalTest has unconnected port Clock
[Synth 8-3330] design Top_PhysicalTest has an empty top module
[Synth 8-3331] design Top_PhysicalTest has unconnected port Clock
[Project 1-571] Translating synthesized netlist

My scenario is a bit complex, so the minimal example look not as minimal as it should look. I did not inline every function to prevent copy/replace errors and I did not remove the debug and assert/report routines.

Overview:

  1. Package utils: common types, enums and functions
  2. Package strings: string manipulation and conversion functions
  3. Package physical: the new type and it's functions
  4. Entity: a single top-level entity implementing a simple delay element / shift-register

Minimal example:

-- EMACS settings: -*-  tab-width: 2; indent-tabs-mode: t -*-
-- vim: tabstop=2:shiftwidth=2:noexpandtab
-- kate: tab-width 2; replace-tabs off; indent-width 2;
-- 
-- ============================================================================
-- Package:                 Common functions and types
--
-- Authors:                 Thomas B. Preusser
--                          Martin Zabel
--                          Patrick Lehmann
--
-- License:
-- ============================================================================
-- Copyright 2007-2015 Technische Universitaet Dresden - Germany
--                                       Chair for VLSI-Design, Diagnostics and Architecture
-- 
-- Licensed under the Apache License, Version 2.0 (the "License");
-- you may not use this file except in compliance with the License.
-- You may obtain a copy of the License at
-- 
--      http://www.apache.org/licenses/LICENSE-2.0
-- 
-- Unless required by applicable law or agreed to in writing, software
-- distributed under the License is distributed on an "AS IS" BASIS,
-- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-- See the License for the specific language governing permissions and
-- limitations under the License.
-- ============================================================================

library IEEE;
use     IEEE.STD_LOGIC_1164.all;
use     IEEE.NUMERIC_STD.all;

package utils is
    -- rounding style
    type T_ROUNDING_STYLE   is (ROUND_TO_NEAREST, ROUND_TO_ZERO, ROUND_TO_INF, ROUND_UP, ROUND_DOWN);

    function ite(cond : BOOLEAN; value1 : STRING; value2 : STRING) return STRING;
    function imin(arg1 : integer; arg2 : integer) return integer;
    function imax(arg1 : integer; arg2 : integer) return integer;
    function log2ceil(arg : positive) return natural;
    function log2ceilnz(arg : positive) return positive;
end package utils;

package body utils is
    function ite(cond : BOOLEAN; value1 : STRING; value2 : STRING) return STRING is
    begin
        if cond then
            return value1;
        else
            return value2;
        end if;
    end function;

    function imin(arg1 : integer; arg2 : integer) return integer is
    begin
        if arg1 < arg2 then return arg1; end if;
        return arg2;
    end function;

    function imax(arg1 : integer; arg2 : integer) return integer is
    begin
        if arg1 > arg2 then return arg1; end if;
        return arg2;
    end function;

    function log2ceil(arg : positive) return natural is
        variable tmp : positive     := 1;
        variable log : natural      := 0;
    begin
        if arg = 1 then return 0; end if;
        while arg > tmp loop
            tmp := tmp * 2;
            log := log + 1;
        end loop;
        return log;
    end function;

    function log2ceilnz(arg : positive) return positive is
    begin
        return imax(1, log2ceil(arg));
    end function;
end utils;

-- ============================================================================
-- Package:                 String related functions and types
--
-- Authors:                 Thomas B. Preusser
--                          Martin Zabel
--                          Patrick Lehmann
--
-- =============================================================================
library IEEE;
use     IEEE.STD_LOGIC_1164.all;
use     IEEE.NUMERIC_STD.all;
use     IEEE.MATH_REAL.all;

use     work.utils.all;

package strings is
    function raw_format_nat_dec(value : NATURAL)                return STRING;
    function str_format(value : REAL; precision : NATURAL := 3) return STRING;
    FUNCTION resize(str : STRING; size : POSITIVE; FillChar : CHARACTER := NUL) RETURN STRING;
    function str_length(str : STRING)                           return NATURAL;
    function str_trim(str : STRING)                             return STRING;
    function str_substr(str : STRING; start : INTEGER := 0; length : INTEGER := 0) return STRING;
end package strings;

package body strings is
    -- raw_format_* functions
    function raw_format_nat_dec(value : NATURAL) return STRING is
    begin
        return INTEGER'image(value);
    end function;

    -- str_format_* functions
    function str_format(value : REAL; precision : NATURAL := 3) return STRING is
        constant s      : REAL          := sign(value);
        constant int    : INTEGER       := integer((value * s) - 0.5);                                                                      -- force ROUND_DOWN
        constant frac   : INTEGER       := integer((((value * s) - real(int)) * 10.0**precision) - 0.5);    -- force ROUND_DOWN
        constant res    : STRING        := raw_format_nat_dec(int) & "." & raw_format_nat_dec(frac);
    begin
--      assert (not MY_VERBOSE)
--          report "str_format:" & CR &
--                       "  value:" & REAL'image(value) & CR &
--                       "  int = " & INTEGER'image(int) & CR &
--                       "  frac = " & INTEGER'image(frac)
--          severity note;
        return ite((s   < 0.0), "-" & res, res);
    end function;

    -- resize
    FUNCTION resize(str : STRING; size : POSITIVE; FillChar : CHARACTER := NUL) RETURN STRING IS
        CONSTANT MaxLength  : NATURAL               := imin(size, str'length);
        VARIABLE Result     : STRING(1 TO size)     := (OTHERS => FillChar);
    BEGIN
        if (MaxLength > 0) then
            Result(1 TO MaxLength) := str(str'low TO str'low + MaxLength - 1);
        end if;
        RETURN Result;
    END FUNCTION;

    -- String functions
    FUNCTION str_length(str : STRING) RETURN NATURAL IS
        VARIABLE l  : NATURAL       := 0;
    BEGIN
        FOR I IN str'range LOOP
            IF (str(I) = NUL) THEN
                RETURN l;
            ELSE
                l := l + 1;
            END IF;
        END LOOP;
        RETURN str'length;
    END FUNCTION;

    function str_trim(str : STRING) return STRING is
        constant len    : NATURAL   := str_length(str);
    begin
        if (len = 0) then
            return "";
        else
            return resize(str, len);
        end if;
    end function;

    function str_substr(str : STRING; start : INTEGER := 0; length : INTEGER := 0) return STRING is
        variable StartOfString      : positive;
        variable EndOfString        : positive;
    begin
        if (start < 0) then         -- start is negative -> start substring at right string boundary
            StartOfString       := str'high + start + 1;
        elsif (start = 0) then  -- start is zero -> start substring at left string boundary
            StartOfString       := str'low;
        else                                        -- start is positive -> start substring at left string boundary + offset
            StartOfString       := start;
        end if;

        if (length < 0) then        -- length is negative -> end substring at length'th character before right string boundary
            EndOfString         := str'high + length;
        elsif (length = 0) then -- length is zero -> end substring at right string boundary
            EndOfString         := str'high;
        else                                        -- length is positive -> end substring at StartOfString + length
            EndOfString         := StartOfString + length - 1;
        end if;

        if (StartOfString < str'low) then   report "StartOfString is out of str's range. (str=" & str & ")" severity error;     end if;
        if (EndOfString < str'high) then    report "EndOfString is out of str's range. (str=" & str & ")" severity error;           end if;

        return str(StartOfString to EndOfString);
    end function;
end strings;


-- ============================================================================
-- Package:                 This VHDL package declares new physical types and their
--                                  conversion functions.
--
-- Authors:                 Patrick Lehmann
-- 
-- ============================================================================
library IEEE;
use     IEEE.MATH_REAL.all;

use     work.utils.all;
use     work.strings.all;

package physical is
    type FREQ is range 0 to INTEGER'high units
        Hz;
        kHz = 1000 Hz;
        MHz = 1000 kHz;
        GHz = 1000 MHz;
        THz = 1000 GHz;
    end units;

    constant C_PHYSICAL_REPORT_TIMING_DEVIATION     : BOOLEAN       := TRUE;
    function to_time(f : FREQ)  return TIME;
    function to_real(t : TIME;          scale : TIME)       return REAL;
    function to_real(f : FREQ;          scale : FREQ)       return REAL;
    function TimingToCycles(Timing : TIME; Clock_Period : TIME; RoundingStyle : T_ROUNDING_STYLE := ROUND_TO_NEAREST) return NATURAL;
    function TimingToCycles(Timing : TIME; Clock_Frequency  : FREQ; RoundingStyle : T_ROUNDING_STYLE := ROUND_TO_NEAREST) return NATURAL;
    function to_string(t : TIME; precision : NATURAL := 3)  return STRING;
    function to_string(f : FREQ; precision : NATURAL := 3)  return STRING;
end physical;


package body physical is
    -- iSim 14.7 does not support fs in simulation by default (fs values are converted to 0 ps)
    --  activate fs support by overriding the time precision
    --  fuse[.exe] [...] -timeprecision_vhdl 1fs [...]
    function MinimalTimeResolutionInSimulation return TIME is
    begin
        if      (1 fs > 0 sec) then return 1 fs;
        elsif   (1 ps > 0 sec) then return 1 ps;
        elsif   (1 ns > 0 sec) then return 1 ns;
        elsif   (1 us > 0 sec) then return 1 us;
        elsif   (1 ms > 0 sec) then return 1 ms;
        else                                            return 1 sec;
        end if;
    end function;

    -- real division for physical types
    function div(a : TIME; b : TIME) return REAL is
        constant MTRIS  : TIME      := MinimalTimeResolutionInSimulation;
    begin
        if  (a < 1 us) then
            return real(a / MTRIS) / real(b / MTRIS);
        elsif (a < 1 ms) then
            return real(a / (1000 * MTRIS)) / real(b / MTRIS) * 1000.0;
        elsif (a < 1 sec) then
            return real(a / (1000000 * MTRIS)) / real(b / MTRIS) * 1000000.0;
        else
            return real(a / (1000000000 * MTRIS)) / real(b / MTRIS) * 1000000000.0;
        end if;
    end function;

    function div(a : FREQ; b : FREQ) return REAL is
    begin
        return real(a / 1 Hz) / real(b / 1 Hz);
    end function;


    -- conversion functions
    function to_time(f : FREQ) return TIME is
        variable res : TIME;
    begin
        if    (f < 1.0 kHz) then res := div(1.0  Hz, f) * 1.0 sec;
        elsif (f < 1.0 MHz) then res := div(1.0 kHz, f) * 1.0 ms;
        elsif (f < 1.0 GHz) then res := div(1.0 MHz, f) * 1.0 us;
        elsif (f < 1.0 THz) then res := div(1.0 GHz, f) * 1.0 ns;
        else                     res := div(1.0 THz, f) * 1.0 ps;
        end if;

        assert FALSE report "to_time: f= " & to_string(f) & "  return " & to_string(res) severity note;
        return res;
    end function;

    -- convert physical types (TIME, FREQ) to standard type (REAL)
    function to_real(t : TIME; scale : TIME) return REAL is
    begin
        if      (scale = 1.0    fs) then    return div(t, 1.0    fs);
        elsif   (scale = 1.0    ps) then    return div(t, 1.0    ps);
        elsif   (scale = 1.0    ns) then    return div(t, 1.0    ns);
        elsif   (scale = 1.0    us) then    return div(t, 1.0    us);
        elsif   (scale = 1.0    ms) then    return div(t, 1.0    ms);
        elsif   (scale = 1.0 sec) then  return div(t, 1.0 sec);
        else    report "to_real: scale must have a value of '1.0 <unit>'" severity failure;
        end if;
    end;

    function to_real(f : FREQ; scale : FREQ) return REAL is
    begin
        if      (scale = 1.0    Hz) then    return div(f, 1.0    Hz);
        elsif   (scale = 1.0 kHz) then  return div(f, 1.0 kHz);
        elsif   (scale = 1.0 MHz) then  return div(f, 1.0 MHz);
        elsif   (scale = 1.0 GHz) then  return div(f, 1.0 GHz);
        elsif   (scale = 1.0 THz) then  return div(f, 1.0 THz);
        else    report "to_real: scale must have a value of '1.0 <unit>'" severity failure;
        end if;
    end;

    -- calculate needed counter cycles to achieve a given 1. timing/delay and 2. frequency/period
    -- ===========================================================================
    --  @param Timing           A given timing or delay, which should be achived
    --  @param Clock_Period     The period of the circuits clock
    --  @RoundingStyle          Default = round to nearest; other choises: ROUND_UP, ROUND_DOWN
    function TimingToCycles(Timing : TIME; Clock_Period : TIME; RoundingStyle : T_ROUNDING_STYLE := ROUND_TO_NEAREST) return NATURAL is
        variable res_real   : REAL;
        variable res_nat    : NATURAL;
    begin
        res_real := div(Timing, Clock_Period);  
        case RoundingStyle is
            when ROUND_TO_NEAREST =>    res_nat := natural(round(res_real));
            when ROUND_UP =>            res_nat := natural(res_real + 0.5);
            when ROUND_DOWN =>          res_nat := natural(res_real);
            when others =>  report "RoundingStyle '" & T_ROUNDING_STYLE'image(RoundingStyle) & "' not supported." severity failure;
        end case;
        report "res_real: " & REAL'image(res_real) severity note;

        assert FALSE
            report "TimingToCycles: " &     CR &
                         "  Timing: " &             to_string(Timing) & CR &
                         "  Clock_Period: " &       to_string(Clock_Period) & CR &
                         "  RoundingStyle: " &  str_substr(T_ROUNDING_STYLE'image(RoundingStyle), 7) & CR &
                         "  res_real = " &          str_format(res_real, 3) & CR &
                         "  => " &                  INTEGER'image(res_nat)
            severity note;

        return res_nat;
    end;

    function TimingToCycles(Timing : TIME; Clock_Frequency  : FREQ; RoundingStyle : T_ROUNDING_STYLE := ROUND_TO_NEAREST) return NATURAL is
    begin
        return TimingToCycles(Timing, to_time(Clock_Frequency), RoundingStyle);
    end function;

    -- convert and format physical types to STRING
    function to_string(t : TIME; precision : NATURAL := 3) return STRING is
        variable unit       : STRING(1 to 3)    := (others => NUL);
        variable value  : REAL;
    begin
        if (t < 1.0 ps) then
            unit(1 to 2)    := "fs";
            value           := to_real(t, 1.0 fs);
        elsif (t < 1.0 ns) then
            unit(1 to 2)    := "ps";
            value           := to_real(t, 1.0 ps);
        elsif (t < 1.0 us) then
            unit(1 to 2)    := "ns";
            value           := to_real(t, 1.0 ns);
        elsif (t < 1.0 ms) then
            unit(1 to 2)    := "us";
            value           := to_real(t, 1.0 us);
        elsif (t < 1.0 sec) then
            unit(1 to 2)    := "ms";
            value           := to_real(t, 1.0 ms);
        else
            unit            := "sec";
            value           := to_real(t, 1.0 sec);
        end if;

        return str_format(value, precision) & " " & str_trim(unit);
    end function;

    function to_string(f : FREQ; precision : NATURAL := 3) return STRING is
        variable unit       : STRING(1 to 3)    := (others => NUL);
        variable value  : REAL;
    begin
        if (f < 1.0 kHz) then
            unit(1 to 2)    := "Hz";
            value           := to_real(f, 1.0 Hz);
        elsif (f < 1.0 MHz) then
            unit            := "kHz";
            value           := to_real(f, 1.0 kHz);
        elsif (f < 1.0 GHz) then
            unit            := "MHz";
            value           := to_real(f, 1.0 MHz);
        elsif (f < 1.0 THz) then
            unit            := "GHz";
            value           := to_real(f, 1.0 GHz);
        else
            unit            := "THz";
            value           := to_real(f, 1.0 THz);
        end if;

        return str_format(value, precision) & " " & str_trim(unit);
    end function;
end package body;


library IEEE;
use     IEEE.STD_LOGIC_1164.ALL;
use     IEEE.NUMERIC_STD.ALL;

use     work.utils.all;
use     work.strings.all;
use     work.physical.all;

entity Top_PhysicalTest is
    Port (
        Clock   : in    STD_LOGIC;
        Input   : in    STD_LOGIC;
        Output  : out   STD_LOGIC
    );
end;

architecture rtl of Top_PhysicalTest is
    -- configuration
    constant CLOCK_FREQ                 : FREQ          := 100 MHz;
    constant SHIFTER_DELAY              : TIME          := 125 ns;
    -- calculations
    constant SHIFTER_DELAY_CYCLES       : NATURAL       := TimingToCycles(SHIFTER_DELAY, CLOCK_FREQ);
    constant SHIFTER_BITS               : NATURAL       := SHIFTER_DELAY_CYCLES + 2; -- to prevent an underrun, while Vivado has a bug

    signal Shifter_nxt                  : STD_LOGIC_VECTOR(SHIFTER_BITS - 1 downto 0);
    signal Shifter_d                    : STD_LOGIC_VECTOR(SHIFTER_BITS - 2 downto 0)   := (others => '0');

begin
    assert false report "CLOCK_FREQ: " & FREQ'image(CLOCK_FREQ)                             severity note;
    assert false report "CLOCK_FREQ: " & to_string(CLOCK_FREQ)                            severity note;
    assert false report "DELAY: "      & TIME'image(SHIFTER_DELAY)                      severity note;
    assert false report "DELAY: "      & to_string(SHIFTER_DELAY)                       severity note;
    assert false report "CYCLES: "     & INTEGER'image(SHIFTER_DELAY_CYCLES)  severity note;

    Shifter_nxt <= Shifter_d & Input;
    Shifter_d   <= Shifter_nxt(Shifter_d'range) when rising_edge(Clock);
    Output      <= Shifter_nxt(SHIFTER_DELAY_CYCLES);
end;

UDPT ::= user defined physical type

My observations:

  • synthesis runs through without any error but causes false calculations
  • 'image is not implemented -> How should one debug this?
  • seems to hold a value of INTEGER'high

Questions:

  • How can I fix this for Vivado?
  • Can anyone confirm this behavior?
  • Which VHDL standard (87, 93, ...) introduced physical types?

Notice: Move my question from CR to SO by hand.

Edit 1:

I stripped the errors down to:

  • decimal literals in physical types are false handled -> result is 0 in every case
  • attribute 'image(..) on physical types is not implemented any more
  • operations with physical types, incl. comparisons results in false values
  • value ranges of physical types are not monotone

The issues are reported in the Xilinx forum.

Kerchief answered 6/1, 2015 at 5:11 Comment(2)
Physical types can be found in VHDL 1987 on page 3-5.Misdeem
One more surprise to be prepared for in physical types for synthesis... electronics.stackexchange.com/questions/124264/…Cynth
A
1

THz is out of range for the minimum range of type Time (see IEEE Std 1076-2008, 16.3 Package standard, "type INTEGER is range implementation_defined;", 5.2.3 Integer types "...An implementation may restrict the bounds of the range constraint of integer types other than type universal_integer. However, an implementation shall allow the declaration of any integer type whose range is wholly contained within the bounds –2147483647 and +2147483647 inclusive.", type FREQ).

Counting from the top, line:pos 272:65, 298:55, 384:51.

I'm not surprised at your results.

  • How can I fix this for Vivado?

Prescalers. Use two integers, one for a fraction part of some integer unit. Alternatively use a bit array type, which shifts the problem to one of string conversion.

It seems unlikely you'll get Xilinx to support 64 bit integers for ranges as in the type declaration:

type FREQ is range 0 to INTEGER'high units
    Hz;
    kHz = 1000 Hz;
    MHz = 1000 kHz;
    GHz = 1000 MHz;
    THz = 1000 GHz;
end units;
  • Can anyone confirm this behavior?

The arithmetic works out, no Vivaldo. It says 32 bit integers are supported.

  • Which VHDL standard (87, 93, ...) introduced physical types?

As Bill Lynch notes physical types are found in IEEE Std 1076-1987, 3.1.3 Physical types, Page 3-5.

This demonstrates the advantages of using a VHDL analyzer/simulator to validate a design specification before synthesis:

ghdl -a top_physicaltest.vhdl
top_physicaltest.vhdl:272:65: static constant violates bounds
top_physicaltest.vhdl:298:55: static constant violates bounds
top_physicaltest.vhdl:384:51: static constant violates bounds


Using ghdl-0.31 I increased the HIGH bound of FREQ:

    type FREQ is range 0 to 2**61 units

This works because ghdl's universal integer is 64 bit and physical types can have a universal integer range.

(There's a bug in ghdl, should work with the integer equivalent of 2**63 -1, Tristan's fixed it I think for ghdl-0.33. I happended to know 2**61 is a safe bound).

That analyzes. You could note that the Real Range limits the accuracy of any scaling your perform. Without checking I couldn't tell you if ghdl has a 64 bit universal Real to match it's 64 bit universal integer.

So then analyze, elaborate and run Top_PhysicalTest (the test bench).

ghdl -r top_physicaltest
./top_physicaltest:error: bound check failure at physical.vhdl:117
ghdl: compilation error

    -- raw_format_* functions
    function raw_format_nat_dec(value : NATURAL) return STRING is
    begin
        return INTEGER'image(value);
    end function;

    -- str_format_* functions
    function str_format(value : REAL; precision : NATURAL := 3) return STRING is
        constant s      : REAL          := sign(value);
        constant int    : INTEGER       := integer((value * s) - 0.5);                                                                      -- force ROUND_DOWN
        constant frac   : INTEGER       := integer((((value * s) - real(int)) * 10.0**precision) - 0.5);    -- force ROUND_DOWN
        constant res    : STRING        := raw_format_nat_dec(int) & "." & raw_format_nat_dec(frac); -- LINE 117
    begin
--      assert (not MY_VERBOSE)
--          report "str_format:" & CR &
--                       "  value:" & REAL'image(value) & CR &
--                       "  int = " & INTEGER'image(int) & CR &
--                       "  frac = " & INTEGER'image(frac)
--          severity note;
        return ite((s   < 0.0), "-" & res, res);
    end function;

Where Line 117 is the declaration for constant res in the package body for package strings.

And from std.standard:

subtype NATURAL is INTEGER range 0 to INTEGER'HIGH;

Type integer is 32 bit precision in ghdl (and universally in synthesis tools).

And that looks like a 'IMAGE error I may or may not have notified Tristan of (Yup!, see #31 Type conversion subtype constraint check not performed, it's fixed for 0.33, too (gee I'm going to skip 0.32 I think)).

So I tried with nvc, which also has 64 bit universal integers, and I had previously submitted some errors to do with expressions for:

nvc -a physical.vhdl
nvc -e Top_PhysicalTest
** Fatal: expression cannot be folded to an integer constant
File physical.vhdl, Line 412

And Line 412 is in Top_PhysicalTest:

    constant CLOCK_FREQ                 : FREQ          := 100 MHz;
    constant SHIFTER_DELAY              : TIME          := 125 ns;
    -- calculations
    constant SHIFTER_DELAY_CYCLES       : NATURAL       := TimingToCycles(SHIFTER_DELAY, CLOCK_FREQ);
    constant SHIFTER_BITS               : NATURAL       := SHIFTER_DELAY_CYCLES + 2; -- to prevent an underrun, while Vivado has a bug

    constant SHIFTER_BITS               : NATURAL       := SHIFTER_DELAY_CYCLES + 2; -- to prevent an underrun, while Vivado has a bug

    signal Shifter_nxt                  : STD_LOGIC_VECTOR(SHIFTER_BITS - 1 downto 0);

The declaration for signal Shifter_nxt.

(And which looks like a different type of error).

Accost answered 6/1, 2015 at 7:35 Comment(6)
within the bounds –2147483647 and +2147483647 does this mean VHDL uses one's complement instead of two's complement to store negative numbers? Or is this just a typo? I used INTEGER'high, to be prepared, if future tools support 64 bit integers ;)Kerchief
I changed <UDPT>'image(...) to INTEGER'image(<UDPT>'pos(...)) for testing purposes. I noticed that passing the parameter 1.0 MHz to div(freq, freq) reports 0 for the first parameter. When I change parameter a to 1 MHz div reports 1,000,000. VHDL-1987 allows decimal literals in physical literals... strange.Kerchief
VHDL doesn't mandate how your computer stores negative numbers. So I presume the omission of -2^31 from the guaranteed range was a way to be on the safe side. (Numeric_Std does mandate 2's comp, but nobody forces you to use it. It presumably leads to slow simulations on sign+magnitude or ones comp machines!)Cynth
VHDL is compatible with big iron, BCD (separate sign bit) ALUs or even floating point only with large enough mantissa. Those are largely lost in history while range and array indexes are still tied to universal integer inherited from CONLAN's univ@, a base type. The only place you find actual bits is in bit types and their array types. It's about the arithmetic and not the implementation. Even Ada with it's modular types is about emulating bits (and happens to be very fast on hardware with bits comprising a base type).Accost
@DavidKoontz As far as I can tell now, there seem to be errors in (1) the parser when reading decimal literals: 1 ns is ok, but 1.0 ns is just 0 fs; 100 MHz is ok, but 100.1 MHz is 2^31 (YES its' INTEGER'high + 1 !). (2) mult and div operators on UDPT cause strange results: 0.01 * 1 us -> 1.2 billion. (3) errors in type conversion: constant res : integer := natural(-0.5); -> -1 ... My steps are: (a) rewrite the entire code to integer literals in phy literals and (b) use FREQ'pos for all calculations. After that, I'll extract this errors into a new test so I can proof them standalone.Kerchief
Didn't see notification for comment.Accost
K
0

I seems like Xilinx Vivado 2019.2 is going to fix physical types and especially TIME.

From Vivado 2019.2 release notes:
enter image description here

Kerchief answered 17/11, 2019 at 10:54 Comment(0)

© 2022 - 2024 — McMap. All rights reserved.