Subject: RE: shortreal <-> bit conversion proposal
From: Jay Lawrence (lawrence@cadence.com)
Date: Tue Jan 07 2003 - 05:30:22 PST
Gord,
This looks fine if we want a separate task for this purpose.
Was the alternative considered that $realtobits() could just be modified
to accept either a 32 or 64-bit real value and return the corresponding
width vector? There is some precendent to have $tasks with variable type
args (i.e. $display). I think the proposal on $cast is proposing to do
the same thing.
It may be handy if I'm modeling with a real and then determine a
shortreal provides acceptable precision to not have to change the $tasks
I use.
Just a thought,
Jay
> -----Original Message-----
> From: Gordon Vreugdenhil [mailto:gvreugde@synopsys.com]
> Sent: Monday, January 06, 2003 5:09 PM
> To: SV_BC List
> Subject: shortreal <-> bit conversion proposal
>
>
> Proposal:
>
> ADD:
>
> Section 16.7 "shortreal conversions"
>
> Verilog 2001 defines a "real" type and functions $realtobits and
> $bitstoreal to permit exact bit pattern transfers between a
> real and a
> 64 bit vector. SystemVerilog adds the "shortreal" type and in a
> parallel manner, $shortrealtobits and $bitstoshortreal are
> defined to
> permit exact bit transfers between a shortreal and a 32 bit vector.
>
> [31:0] $shortrealtobits(shortreal_val) ;
> shortreal $bitstoshortreal(bit_val) ;
>
> $shortrealtobits converts from a shortreal number to the 32-bit
> representation (vector) of that shortreal number. $bitstoshortreal
> is the reverse of $shortrealtobits; it converts from the
> bit pattern to
> a shortreal number.
>
>
>
> Note: It would probably be better to have the above section become
> 16.3 and shift 16.3-16.6 down to 16.4-16.7. If so, change
> "16.7" in the following proposal to "16.3".
>
> : The above text is a simply a slightly modified version of
> the IEEE 2001 text for $bitstoreal and $realtobits.
>
>
>
> Proposal (dependent on the above being accepted):
>
> In Section 3.8:
>
> REPLACE:
>
> When a shortreal is converted to an int, its value is rounded as
> in Verilog. So the conversion can lose information. When a
> shortreal is
> converted to 32 bits, its bit pattern is preserved, which
> means it can be
> converted back to the same value without any loss of
> information. This
> technique can also be used for structures, where the $bits
> attribute gives
> the size of a structure in bits (the $bits system function is
> discussed in
> section 16.2):
>
> WITH:
>
> When a shortreal is converted to an int or to 32 bits, its value is
> rounded as in Verilog. So the conversion can lose
> information. To convert
> a shortreal to its underlying bit representation without a loss of
> information,
> use $shortrealtobits as defined in Section 16.7. To convert from the
> bit representation of a shortreal value into a shortreal, use
> $bitstoshortreal as defined in Section 16.7.
>
> Structures can be converted to bits preserving the bit
> pattern, which means
> they can be converted back to the same value without any loss of
> information. The following example demonstrates this
> conversion. In the
> example, the $bits attribute gives the size of a structure in
> bits (the
> $bits system function is discussed in section 16.2):
>
>
> --
> ----------------------------------------------------------------------
> Gord Vreugdenhil gvreugde@synopsys.com
> Staff Engineer, VCS (Verification Tech. Group) (503) 547-6054
> Synopsys Inc., Beaverton OR
>
This archive was generated by hypermail 2b28 : Tue Jan 07 2003 - 05:31:02 PST