| Number | 8
|
| Category | errata
|
| Synopsis | Incorrect zero extend rules for relational and equality operators
|
| State | lrmdraft
|
| Class | errata-simple
|
| Arrival-Date | Jul 31 2001
|
| Originator | Paul Graham <pgraham@cadence.com>
|
| Release | 2001a 2001b, 4.1.7
|
| Environment |
|
| Description |
Section 4.1.7 says that when a relational operator has two operands of different sizes, then the smaller operand is zero-extended. This makes no intuitive sense for signed operands, and it conflicts with the general rules for determining sign given at the end of chapter 4. It also conflicts with the behavior of Verilog-XL and NC-Verilog. Likewise, 4.1.8 says that the smaller operand to an equality operator is zero-extended, which I also believe is wrong. Paul |
| Fix |
Submitted to IEEE in October 2001: Changes incorporated into 2001a: Passed 10/7/02: BTF-PR8: Section 4.1.7 - 3rd & 2nd to last paragraphs REPLACE: When two operands of unequal bit lengths are used, the smaller operand shall be zero filled on the most significant bit side to extend to the size of the larger operand. WITH: When two operands of unequal bit lengths are used and one or both of the operands is unsigned, the smaller operand shall be zero filled on the most significant bit side to extend to the size of the larger operand. REPLACE: When both operands of a relational expression are signed integral operands (an integer, or a unsized, unbased integer) then the expression shall be interpreted as a comparison between signed values. WITH: When both operands of a relational expression are signed integral operands (an integer, a signed reg data type, or an unsized, unbased integer) then the expression shall be interpreted as a comparison between signed values. Additional changes passed for 2001b 10/7/2002: REPLACE: When both operands of a relational expression are signed integral operands (an integer, a signed reg data type, or an unsized, unbased integer) then the expression shall be interpreted as a comparison between signed values. WITH: When both operands of a relational expression are signed operands, then the expression shall be interpreted as a comparison between signed values. |
| Audit-Trail |
Reply from Shalom: Date: Fri, 27 Sep 2002 12:35:38 +0300 (IDT) From: Shalom.Bresticker@motorola.com To: etf-bugs@boyd.com Cc: 1364@accellera.org Subject: /8: Incorrect zero extend rules for relational and equality operators Note: errata/8 was split into 2 parts. errata/8 now only relates to 4.1.7, errata/106 relates to 4.1.8. I request to return errata/8 to the ETF for reconsideration. The proposed fix has several problems: Part of the proposed change is: REPLACE: When both operands of a relational expression are signed integral operands (an integer, or a unsized, unbased integer) then the expression shall be interpreted as a comparison between signed values. WITH: When both operands of a relational expression are signed integral operands (an integer, a signed reg data type, or an unsized, unbased integer) then the expression shall be interpreted as a comparison between signed values. First, in the replacement text, "signed integral operands" was supposed to be just "signed operands". See the original Cliff proposal. However, the word "integral" was originally there to distinguish this case from the case of real operands. Second, this formulation causes confusion between "an integer" and "an unsized, unbased integer". It also leaves out the case of signed net data types, and explicitly signed integer constants. So if we want to list all the possible signed operand types, we have: integer data type signed reg data type signed net data type (explicitly) signed integer constant unsized, unbased integer constant I have some difficulty to formulate it nicely, but the current formulation is not good enough. Shalom From: Shalom.Bresticker@motorola.com To: etf-bugs@boyd.com Cc: 1364@accellera.org Subject: Re: errata/8: Incorrect zero extend rules for relational and equality operators Date: Fri, 4 Oct 2002 11:41:38 +0300 (IDT) >Category: errata >Confidential: no >Originator: Shalom.Bresticker@motorola.com >Class: TBD >Description: I propose to reword this simply as follows: > When both operands of a relational expression are > signed operands, > then the expression > shall be interpreted as a comparison between > signed values. Reason: It is unneeded and difficult to specify here all the cases of signed non-real operands. The following sentence, which begins, "When either operand of a relational expression is a real operand ...", does not contradict this first reworded sentence. Shalom On Fri, 27 Sep 2002, Shalom Bresticker wrote: > WITH: When both operands of a relational expression are > signed integral operands (an integer, a signed reg > data type, or an unsized, unbased integer) then the expression > shall be interpreted as a comparison between > signed values. |
| Unformatted |
|
Hosted by Boyd Technology