| Add Proposal | Add Analysis | Edit Class, Environment, or Release |
| Number | 274
|
| Category | errata
|
| Synopsis | LRM does not specify how Infinity and NaNs to be displayed for
|
| State | open
|
| Class | errata-discuss
|
| Arrival-Date | Jan 26 2003
|
| Originator | Shalom Bresticker <Shalom.Bresticker@motorola.com>
|
| Release | 2001b: 17.1
|
| Environment |
|
| Description |
-- Shalom Bresticker Shalom.Bresticker@motorola.com Design & Reuse Methodology Tel: +972 9 9522268 Motorola Semiconductor Israel, Ltd. Fax: +972 9 9522890 POB 2208, Herzlia 46120, ISRAEL Cell: +972 50 441478 |
| Fix |
|
| Audit-Trail |
From: Steven Sharp <sharp@cadence.com> To: etf-bugs@boyd.com, Shalom.Bresticker@motorola.com Cc: Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Thu, 21 Aug 2003 19:39:37 -0400 (EDT) I suspect that all tools just use the C stdio library to format their real numbers for displaying. We certainly do. The C standard does not specify how these values should be displayed (as should be expected, since the C standard does not assume IEEE floating point). This could result in slight format differences between platforms. I have no interest in reinventing the C stdio formatting capabilities, especially since those routines may have been optimized using specialized knowledge of floating point algorithms. So I would oppose specifying anything that might prevent using those routines. Steven Sharp sharp@cadence.com From: Shalom.Bresticker@freescale.com To: Steven Sharp <sharp@cadence.com> Cc: etf-bugs@boyd.com Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Wed, 19 May 2004 10:58:34 +0300 (IDT) Hi, I'm catching up on mail that I've left in my Inbox many months ago. Not that I don't have work to do, just a little bored... Anyway... But the IEEE Verilog standard DOES specify use of IEEE Std 754. By chance, I was reading an old interview with Prof. William Kahan, one of the key developers of 754 (an extremely interesting person, see the publications at the end of his home page, http://www.cs.berkeley.edu/~wkahan/). He pointed out that 754 is a software standard as well as a hardware standard. I want and expect to get consistent results from my Verilog programs on all platforms which I can easily run a diff on. The C standard does in fact specify output formats of real numbers, including how to handle Infinity and Nan. I would accept using the same specification as the C standard. It might even be possible to reference the appropriate section of the C standard, just as we reference 754. The C standard does allow a little flexibility. In fact, maybe we should consider that large parts of the I/O routine definitions could be replaced by references to the appropriate places in the C standard. But as long as the Verilog LRM is stand-alone, it must specify how Infinity and NaN are to be displayed. On Thu, 21 Aug 2003, Steven Sharp wrote: > I suspect that all tools just use the C stdio library to format their real > numbers for displaying. We certainly do. The C standard does not specify > how these values should be displayed (as should be expected, since the C > standard does not assume IEEE floating point). This could result in slight > format differences between platforms. I have no interest in reinventing > the C stdio formatting capabilities, especially since those routines may > have been optimized using specialized knowledge of floating point algorithms. > > So I would oppose specifying anything that might prevent using those routines. From: "Vreugdenhil, Gordon" <gordon_vreugdenhil@mentorg.com> To: etf-bugs@boyd.com Cc: Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Wed, 19 May 2004 07:11:42 -0700 Shalom, I agree that we should essentially adopt C rules for reals. Note, however, that there is variance between C libs in the capitalization of NaN and Inf -- some vendors use "NaN" and "Inf" while others use "nan" and "inf". I don't think that the standards specify that aspect of the behavior. Gord. -- -------------------------------------------------------------------- Gordon Vreugdenhil, Staff Engineer 503-685-0808 Model Technology (Mentor Graphics) gordonv@model.com From: Shalom.Bresticker@freescale.com To: "Vreugdenhil, Gordon" <gordon_vreugdenhil@mentorg.com> Cc: etf-bugs@boyd.com Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Wed, 19 May 2004 17:32:02 +0300 (IDT) The draft C standard that I have (from Aug 3, 1998) says about the f format: A double argument representing an infinity is converted in one of the styles [-]inf or [-]infinity which style is implementation-defined. A double argument representing a NaN is converted in one of the styles [-]nan or [-]nan(n-char-sequence) which style, and the meaning of any n-char-sequence, is implementation-defined. The F conversion specifier produces INF, INFINITY, or NAN instead of inf, infinity, or nan, respectively. Of course, we could be more flexible about capatilization. Shalom > Shalom, I agree that we should essentially adopt C rules for > reals. Note, however, that there is variance between C libs > in the capitalization of NaN and Inf -- some vendors use > "NaN" and "Inf" while others use "nan" and "inf". I don't think > that the standards specify that aspect of the behavior. -- Shalom Bresticker Shalom.Bresticker @freescale.com Design & Reuse Methodology Tel: +972 9 9522268 Freescale Semiconductor Israel, Ltd. Fax: +972 9 9522890 POB 2208, Herzlia 46120, ISRAEL Cell: +972 50 5441478 [ ]Freescale Internal Use Only [ ]Freescale Confidential Proprietary From: Steven Sharp <sharp@cadence.com> To: etf-bugs@boyd.com, Shalom.Bresticker@freescale.com Cc: Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Wed, 19 May 2004 13:17:07 -0400 (EDT) I suspect that all existing implementations just call C sprintf() to format real numbers for display. I can't imagine why they would do anything else. As a practical matter, the standard shouldn't specify anything that conflicts with that. Steven Sharp sharp@cadence.com From: Shalom.Bresticker@freescale.com To: Steven Sharp <sharp@cadence.com> Cc: etf-bugs@boyd.com Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Wed, 19 May 2004 21:45:02 +0300 (IDT) > I suspect that all existing implementations just call C sprintf() to format > real numbers for display. I can't imagine why they would do anything else. > As a practical matter, the standard shouldn't specify anything that conflicts > with that. The point is (one point, anyway) that the user has a right to know in advance what his program will print in such a case. This came up when we were discussing the power operator. It never occurred to me, for example, that I might get an output with non-numeric characters. If I were writing a program or script to read that output, I would have to take that into account. No one is suggesting to conflict with C. Shalom From: Steven Sharp <sharp@cadence.com> To: sharp@cadence.com, Shalom.Bresticker@freescale.com Cc: etf-bugs@boyd.com Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Wed, 19 May 2004 14:52:58 -0400 (EDT) I notice that the LRM doesn't fully describe the capabilities of the real number formats either. It just says they "have the full formatting capabilities available in the C language." Steven Sharp sharp@cadence.com From: Shalom Bresticker <Shalom.Bresticker@freescale.com> To: Steven Sharp <sharp@cadence.com> Cc: Shalom Bresticker <Shalom.Bresticker@motorola.com>, etf-bugs@boyd.com Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Thu, 20 May 2004 07:39:00 +0300 Hmmm. That is itself a statement which needs to be verified. I.e., does what is explicitly specified in the Verilog LRM contradict anything from C? Shalom Steven Sharp wrote: > I notice that the LRM doesn't fully describe the capabilities of the real > number formats either. It just says they "have the full formatting > capabilities available in the C language." > > Steven Sharp > sharp@cadence.com -- Shalom Bresticker Shalom.Bresticker @freescale.com Design & Reuse Methodology Tel: +972 9 9522268 Freescale Semiconductor Israel, Ltd. Fax: +972 9 9522890 POB 2208, Herzlia 46120, ISRAEL Cell: +972 50 5441478 [ ]Freescale Internal Use Only [ ]Freescale Confidential Proprietary From: Shalom.Bresticker@freescale.com To: Steven Sharp <sharp@cadence.com> Cc: etf-bugs@boyd.com Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Thu, 20 May 2004 22:15:15 +0300 (IDT) On Thu, 20 May 2004, Steven Sharp wrote: > > Hmmm. That is itself a statement which needs to be verified. > > I.e., does what is explicitly specified in the Verilog LRM contradict anything > > from C? > > The statement is only referring to the real formats (%e, %f, %g). That > statement and a simple example are the only things specified for them in > the LRM. I know. > This statement also doesn't specify a C standard version, which could be > an issue if the formatting capabilities for reals have changed in C. That's exactly the point. I have a draft version of the ISO standard, I don't have the final version. But this draft does not seem to be 100% identical to what the Verilog LRM says in some of the details. Shalom -- Shalom Bresticker Shalom.Bresticker @freescale.com Design & Reuse Methodology Tel: +972 9 9522268 Freescale Semiconductor Israel, Ltd. Fax: +972 9 9522890 POB 2208, Herzlia 46120, ISRAEL Cell: +972 50 5441478 [ ]Freescale Internal Use Only [ ]Freescale Confidential Proprietary From: Steven Sharp <sharp@cadence.com> To: etf-bugs@boyd.com, Shalom.Bresticker@freescale.com Cc: Subject: Re: errata/274: LRM does not specify how Infinity and NaNs to be displayed for real numbers Date: Thu, 20 May 2004 14:11:54 -0400 (EDT) > Hmmm. That is itself a statement which needs to be verified. > I.e., does what is explicitly specified in the Verilog LRM contradict anything > from C? The statement is only referring to the real formats (%e, %f, %g). That statement and a simple example are the only things specified for them in the LRM. This statement also doesn't specify a C standard version, which could be an issue if the formatting capabilities for reals have changed in C. Steven Sharp sharp@cadence.com |
| Unformatted |
real numbers |
Hosted by Boyd Technology