You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@arrow.apache.org by Hirokazu Suzuki <he...@gmail.com> on 2023/04/09 00:03:34 UTC
[DISCUSSION][C++] Re-proposal of 'divmod' in Compute Functions
Hello all,
This is a (re)proposal for a divmod kernel for the compute function.
This proposes the following two combinations of quotient and remainder for
'divmod'.
- [floor-quotient, modulo]
- [trunc-quotient, remainder]
The first proposal for divmod was made in ARROW-12755 [1] and has been
discussed in GH-11116 [2], but has not yet been completed.
On the other hand, the following relationships hold for the dividend,
quotient, divisor,
and modulo,
quotient ∈ Ｚ
dividend = quotient * divisor + modulo,
|modulo| < |divisor|
Since the sign of modulo is undefined in this expression, each language has
a different
behavior regarding the sign of the remainder when a negative number is
involved.
For example, in C, C++, Go, Java, Rust, etc., the remainder will match the
sign
of the divisor. This is equivalent to performing a "truncation division to
zero: trunc"
when finding the quotient.
On the other hand, R yields a remainder that matches the sign of the
divisor,
which is equivalent to performing "truncated division to infinity: floor"
when finding
the quotient.
In addition, some languages provide both of these, such as Julia and Ruby.
The following table lists the signs and operators for integer remainders.
Table: Language, operator and operands
which sign is corresponding to the sign of remainder
|Language |dividend |divisor |
|:--------|:-----------|:--------------|
| C | %, div | |
| C++ | %, div | |
| C# | % | |
| Go | % | |
| Java | % | Math.floorMod |
| Julia | rem | mod |
| MATLAB | rem | mod |
| Python | math.fmod | % |
| R | | %% |
| Ruby | remainder | %, modulo |
| Rust | % | |
| Scala | % | |
| Scheme | remainder | modulo |
| SQL | mod | |
（The remainder of floating point numbers is omitted.）
Since both styles exist, it is necessary to name and implement compute
functions
in a way avoiding confusion that may be used by many languages.
So I propose to create two different divmods sets of quotient rounding
(trunc or floor) and remainders.
- trunc_divmod ≡ [trunc_quotient, remainder]
- floor_divmod ≡ [floor_quotient, modulo]
It is inconvenient without the kernel, at least for positive number.
So I put this proposal here to step the work forward.
I am not an expert in this field and there may be some misunderstandings.
I would appreciate it if you could point out any mistakes.
[1] https://issues.apache.org/jira/browse/ARROW-12755)
[2] https://github.com/apache/arrow/pull/11116)
Best regards,
Hirokazu SUZUKI
Re: [DISCUSSION][C++] Re-proposal of 'divmod' in Compute Functions
Posted by Hirokazu Suzuki <he...@gmail.com>.
@kou Thank you for the comment. I will create issue about this.
2023年5月10日(水) 9:43 Sutou Kouhei <ko...@clear-code.com>:
> Hi,
>
> There is no objection for this proposal.
> Could you write this proposal to
> https://github.com/apache/arrow/issues/28497 too?
>
> Then we can move forward (implement) this proposal.
>
> Thanks,
> --
> kou
>
> In <CA...@mail.gmail.com>
> "[DISCUSSION][C++] Re-proposal of 'divmod' in Compute Functions" on Sun,
> 9 Apr 2023 09:03:34 +0900,
> Hirokazu Suzuki <he...@gmail.com> wrote:
>
> > Hello all,
> >
> > This is a (re)proposal for a divmod kernel for the compute function.
> > This proposes the following two combinations of quotient and remainder
> for
> > 'divmod'.
> >
> > - [floor-quotient, modulo]
> > - [trunc-quotient, remainder]
> >
> > The first proposal for divmod was made in ARROW-12755 [1] and has been
> > discussed in GH-11116 [2], but has not yet been completed.
> >
> > On the other hand, the following relationships hold for the dividend,
> > quotient, divisor,
> > and modulo,
> >
> > quotient ∈ Ｚ
> > dividend = quotient * divisor + modulo,
> > |modulo| < |divisor|
> >
> > Since the sign of modulo is undefined in this expression, each language
> has
> > a different
> > behavior regarding the sign of the remainder when a negative number is
> > involved.
> >
> > For example, in C, C++, Go, Java, Rust, etc., the remainder will match
> the
> > sign
> > of the divisor. This is equivalent to performing a "truncation division
> to
> > zero: trunc"
> > when finding the quotient.
> >
> > On the other hand, R yields a remainder that matches the sign of the
> > divisor,
> > which is equivalent to performing "truncated division to infinity: floor"
> > when finding
> > the quotient.
> >
> > In addition, some languages provide both of these, such as Julia and
> Ruby.
> >
> > The following table lists the signs and operators for integer remainders.
> >
> > Table: Language, operator and operands
> > which sign is corresponding to the sign of remainder
> >
> > |Language |dividend |divisor |
> > |:--------|:-----------|:--------------|
> > | C | %, div | |
> > | C++ | %, div | |
> > | C# | % | |
> > | Go | % | |
> > | Java | % | Math.floorMod |
> > | Julia | rem | mod |
> > | MATLAB | rem | mod |
> > | Python | math.fmod | % |
> > | R | | %% |
> > | Ruby | remainder | %, modulo |
> > | Rust | % | |
> > | Scala | % | |
> > | Scheme | remainder | modulo |
> > | SQL | mod | |
> >
> > （The remainder of floating point numbers is omitted.）
> >
> > Since both styles exist, it is necessary to name and implement compute
> > functions
> > in a way avoiding confusion that may be used by many languages.
> >
> > So I propose to create two different divmods sets of quotient rounding
> > (trunc or floor) and remainders.
> >
> > - trunc_divmod ≡ [trunc_quotient, remainder]
> > - floor_divmod ≡ [floor_quotient, modulo]
> >
> > It is inconvenient without the kernel, at least for positive number.
> > So I put this proposal here to step the work forward.
> >
> > I am not an expert in this field and there may be some misunderstandings.
> > I would appreciate it if you could point out any mistakes.
> >
> > [1] https://issues.apache.org/jira/browse/ARROW-12755)
> > [2] https://github.com/apache/arrow/pull/11116)
> >
> > Best regards,
> > Hirokazu SUZUKI
>
Re: [DISCUSSION][C++] Re-proposal of 'divmod' in Compute Functions
Posted by Sutou Kouhei <ko...@clear-code.com>.
Hi,
There is no objection for this proposal.
Could you write this proposal to
https://github.com/apache/arrow/issues/28497 too?
Then we can move forward (implement) this proposal.
Thanks,
--
kou
In <CA...@mail.gmail.com>
"[DISCUSSION][C++] Re-proposal of 'divmod' in Compute Functions" on Sun, 9 Apr 2023 09:03:34 +0900,
Hirokazu Suzuki <he...@gmail.com> wrote:
> Hello all,
>
> This is a (re)proposal for a divmod kernel for the compute function.
> This proposes the following two combinations of quotient and remainder for
> 'divmod'.
>
> - [floor-quotient, modulo]
> - [trunc-quotient, remainder]
>
> The first proposal for divmod was made in ARROW-12755 [1] and has been
> discussed in GH-11116 [2], but has not yet been completed.
>
> On the other hand, the following relationships hold for the dividend,
> quotient, divisor,
> and modulo,
>
> quotient ∈ Ｚ
> dividend = quotient * divisor + modulo,
> |modulo| < |divisor|
>
> Since the sign of modulo is undefined in this expression, each language has
> a different
> behavior regarding the sign of the remainder when a negative number is
> involved.
>
> For example, in C, C++, Go, Java, Rust, etc., the remainder will match the
> sign
> of the divisor. This is equivalent to performing a "truncation division to
> zero: trunc"
> when finding the quotient.
>
> On the other hand, R yields a remainder that matches the sign of the
> divisor,
> which is equivalent to performing "truncated division to infinity: floor"
> when finding
> the quotient.
>
> In addition, some languages provide both of these, such as Julia and Ruby.
>
> The following table lists the signs and operators for integer remainders.
>
> Table: Language, operator and operands
> which sign is corresponding to the sign of remainder
>
> |Language |dividend |divisor |
> |:--------|:-----------|:--------------|
> | C | %, div | |
> | C++ | %, div | |
> | C# | % | |
> | Go | % | |
> | Java | % | Math.floorMod |
> | Julia | rem | mod |
> | MATLAB | rem | mod |
> | Python | math.fmod | % |
> | R | | %% |
> | Ruby | remainder | %, modulo |
> | Rust | % | |
> | Scala | % | |
> | Scheme | remainder | modulo |
> | SQL | mod | |
>
> （The remainder of floating point numbers is omitted.）
>
> Since both styles exist, it is necessary to name and implement compute
> functions
> in a way avoiding confusion that may be used by many languages.
>
> So I propose to create two different divmods sets of quotient rounding
> (trunc or floor) and remainders.
>
> - trunc_divmod ≡ [trunc_quotient, remainder]
> - floor_divmod ≡ [floor_quotient, modulo]
>
> It is inconvenient without the kernel, at least for positive number.
> So I put this proposal here to step the work forward.
>
> I am not an expert in this field and there may be some misunderstandings.
> I would appreciate it if you could point out any mistakes.
>
> [1] https://issues.apache.org/jira/browse/ARROW-12755)
> [2] https://github.com/apache/arrow/pull/11116)
>
> Best regards,
> Hirokazu SUZUKI