Re: enhancement/409: lists in part-selects

From: Michael McNamara (mac@verisity.com)
Date: Wed Jul 30 2003 - 18:10:02 PDT

  • Next message: Steven Sharp: "Re: enhancement/409: lists in part-selects"

    Precedence: bulk

    The following reply was made to PR enhancement/409; it has been noted by GNATS.

    From: Michael McNamara <mac@verisity.com>
    To: Steven Sharp <sharp@cadence.com>
    Cc: etf-bugs@boyd.com
    Subject: Re: enhancement/409: lists in part-selects
    Date: Wed, 30 Jul 2003 17:13:50 -0700

     Steven Sharp writes:
    > Precedence: bulk
    >
    > The following reply was made to PR enhancement/409; it has been noted by GNATS.
    >
    > From: Steven Sharp <sharp@cadence.com>
    > To: sharp@cadence.com, Shalom.Bresticker@motorola.com
    > Cc: etf-bugs@boyd.com
    > Subject: Re: enhancement/409: lists in part-selects
    > Date: Wed, 30 Jul 2003 20:35:27 -0400 (EDT)
    >
    > >You'd be surprised.
    > >
    > >If often comes up in contexts like
    > >
    > >"if ((a[3:5] == 3'b101) & !a[7] & (a[9:11] == 3'b110) & a[13])".
    >
    >
    > I assume you are suggesting recoding this as
    >
    > if (a[3:5,7,9:11,13] == 8'b10101101)
    >
    > or
    >
    > if (a[3:5,7,9:11,13] == 8'hab)
    >
    >
    > This can already be done with bit masks. Assuming a[0:15],
    >
    > if (a & 16'b0001110101110100 == 16'b0001010001100100)
    >
    > or
    >
    > if (a & 16'h1d74 == 16'h1464)
    >
    > This is similar in compactness (regardless of the length of the signal
    > name). It should execute faster, since it uses bitwise operations
    > that can be done in parallel, rather than extracting and reassembling
    > bit fields sequentially. Most C programmers should already be familiar
    > with the technique of using bit masks to extract only a desired subset of
    > bits for testing. A bit mask can be a parameter or can be derived from
    > multiple partial masks by operators, while a list of ranges and bits
    > cannot. This makes it easier to parameterize the code if desired.
    > And bit masks can be done without any language additions.
    >
    > Some designers may not know how to do this with the existing language
    > features, but it is no harder than learning how to do it a new way,
    > and it actually works better.
    >
    >
    >
    >
    > Steven Sharp
    > sharp@cadence.com
    >
     
     I agree with Steve, but suggest he meant to type
     
       if ( (a & 16'b0001110101110100) == 16'b0001010001100100) begin
     
     as in Verilog (and C) the == operator binds tighter than the &
     operator, so what Steve typed would be interpreted as if it were
     
       if ( a & (16'b0001110101110100 == 16'b0001010001100100))
     

    This archive was generated by hypermail 2.1.4 : Wed Jul 30 2003 - 18:10:48 PDT and
    sponsored by Boyd Technology, Inc.