Interest for a Fixed Point Arithmetics library
Hello everybody. I am working on a project that needs and can use fixed point numbers, so I started searching for a library implementing fixed point numbers, but I didn't find anything. So I was wondering if people would need fixed point arithmetics, and here I am. The aim of this FixedPointNumber is to fix the point at compile-time using template, and having another template argument for the underlying type. So we have : template<typename T, size_t N> class FixedPointNumber { T value; public: //interface }; The N last bits of value is the decimal part : so to convert a FixedPointNumber to an int, we need only to use a bitshift operator (and same for construction from int). sizeof(FixedPointNumber<T,N>) == sizeof(T) , so there is no overhead in this implementation. Arithmetic and relational operators will be implemented. Multiplication needs one int multiplication and one bit shift. Division needs one int division and one bit shift. There is only one value for zero. Sign can be extracted with Most Significant Bit. I have already a basic implementation but if it interests more people I will do a proper implementation. Thank you for reading !
-----Original Message----- From: Boost [mailto:boost-bounces@lists.boost.org] On Behalf Of Julien Vernay via Boost Sent: 30 December 2017 18:30 To: boost@lists.boost.org Cc: Julien Vernay Subject: [boost] Interest for a Fixed Point Arithmetics library
Hello everybody. I am working on a project that needs and can use fixed point numbers, so I started searching for a library implementing fixed point numbers, but I didn't find anything. So I was wondering if people would need fixed point arithmetics, and here I am.
The aim of this FixedPointNumber is to fix the point at compile-time using template, and having another template argument for the underlying type. So we have :
template<typename T, size_t N> class FixedPointNumber { T value; public: //interface };
The N last bits of value is the decimal part : so to convert a FixedPointNumber to an int, we need only to use a bitshift operator (and same for construction from int). sizeof(FixedPointNumber<T,N>) == sizeof(T) , so there is no overhead in this implementation. Arithmetic and relational operators will be implemented. Multiplication needs one int multiplication and one bit shift. Division needs one int division and one bit shift. There is only one value for zero. Sign can be extracted with Most Significant Bit.
I have already a basic implementation but if it interests more people I will do a proper implementation.
An implementation of fixed-point has been written (and is in real-life use) by Chris Kormanyos see https://github.com/BoostGSoC15/fixed_point/ this uses a specified (fixed at compile-time) fixed-point format. Big formats are possible and it works seamlessly with many Boost.Math functions, often using constexpr for compile-time computations). It did not receive much attention when tentatively proposed for Boost. C++ standards groups are also considering a proposal for an 'elastic' (at run-time) fixed-point library by John Macfarlane http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2017/p0554r0.html I suggest you should study these different approaches (not necessarily mutually exclusive)? Paul --- Paul A. Bristow Prizet Farmhouse Kendal UK LA8 8AB +44 (0) 1539 561830
On December 31, 2017 6:50:01 PM "Paul A. Bristow via Boost" <boost@lists.boost.org> wrote:
-----Original Message----- From: Boost [mailto:boost-bounces@lists.boost.org] On Behalf Of Julien Vernay via Boost Sent: 30 December 2017 18:30 To: boost@lists.boost.org Cc: Julien Vernay Subject: [boost] Interest for a Fixed Point Arithmetics library
Hello everybody. I am working on a project that needs and can use fixed point numbers, so I started searching for a library implementing fixed point numbers, but I didn't find anything. So I was wondering if people would need fixed point arithmetics, and here I am.
The aim of this FixedPointNumber is to fix the point at compile-time using template, and having another template argument for the underlying type. So we have :
template<typename T, size_t N> class FixedPointNumber { T value; public: //interface };
The N last bits of value is the decimal part : so to convert a FixedPointNumber to an int, we need only to use a bitshift operator (and same for construction from int). sizeof(FixedPointNumber<T,N>) == sizeof(T) , so there is no overhead in this implementation. Arithmetic and relational operators will be implemented. Multiplication needs one int multiplication and one bit shift. Division needs one int division and one bit shift. There is only one value for zero. Sign can be extracted with Most Significant Bit.
I have already a basic implementation but if it interests more people I will do a proper implementation.
An implementation of fixed-point has been written (and is in real-life use) by Chris Kormanyos
see https://github.com/BoostGSoC15/fixed_point/
this uses a specified (fixed at compile-time) fixed-point format. Big formats are possible and it works seamlessly with many Boost.Math functions, often using constexpr for compile-time computations).
It did not receive much attention when tentatively proposed for Boost.
Is this proposal stalled? I vaguely remember a fixed-point proposal but I don't remember the details. I'm interested in a fast compile-time configured fixed-point type as I had to write one for one project.
-----Original Message----- From: Boost [mailto:boost-bounces@lists.boost.org] On Behalf Of Andrey Semashev via Boost Sent: 31 December 2017 16:14 To: boost@lists.boost.org Cc: Andrey Semashev Subject: Re: [boost] Interest for a Fixed Point Arithmetics library
On December 31, 2017 6:50:01 PM "Paul A. Bristow via Boost" <boost@lists.boost.org> wrote:
-----Original Message----- From: Boost [mailto:boost-bounces@lists.boost.org] On Behalf Of Julien Vernay via Boost Sent: 30 December 2017 18:30 To: boost@lists.boost.org Cc: Julien Vernay Subject: [boost] Interest for a Fixed Point Arithmetics library
Hello everybody. I am working on a project that needs and can use fixed point numbers, so I started searching for a library implementing fixed point numbers, but I didn't find anything. So I was wondering if people would need fixed point arithmetics, and here I am.
The aim of this FixedPointNumber is to fix the point at compile-time using template, and having another template argument for the underlying type. So we have :
template<typename T, size_t N> class FixedPointNumber { T value; public: //interface };
The N last bits of value is the decimal part : so to convert a FixedPointNumber to an int, we need only to use a bitshift operator (and same for construction from int). sizeof(FixedPointNumber<T,N>) == sizeof(T) , so there is no overhead in this implementation. Arithmetic and relational operators will be implemented. Multiplication needs one int multiplication and one bit shift. Division needs one int division and one bit shift. There is only one value for zero. Sign can be extracted with Most Significant Bit.
I have already a basic implementation but if it interests more people I will do a proper implementation.
An implementation of fixed-point has been written (and is in real-life use) by Chris Kormanyos
see https://github.com/BoostGSoC15/fixed_point/
this uses a specified (fixed at compile-time) fixed-point format. Big formats are possible and it works seamlessly with many Boost.Math functions, often using constexpr for compile-time computations).
It did not receive much attention when tentatively proposed for Boost.
Is this proposal stalled? I vaguely remember a fixed-point proposal but I don't remember the details.
I'm interested in a fast compile-time configured fixed-point type as I had to write one for one project.
It is working and in 'real-life' use 'in anger', and reasonably documented with examples of usage. But nobody in Boost seemed to be at all interested, so it wasn't proposed formally. Paul --- Paul A. Bristow Prizet Farmhouse Kendal UK LA8 8AB +44 (0) 1539 561830
On 01/01/18 15:27, Paul A. Bristow via Boost wrote:
-----Original Message----- From: Boost [mailto:boost-bounces@lists.boost.org] On Behalf Of Andrey Semashev via Boost Sent: 31 December 2017 16:14 To: boost@lists.boost.org Cc: Andrey Semashev Subject: Re: [boost] Interest for a Fixed Point Arithmetics library
On December 31, 2017 6:50:01 PM "Paul A. Bristow via Boost" <boost@lists.boost.org> wrote:
An implementation of fixed-point has been written (and is in real-life use) by Chris Kormanyos
see https://github.com/BoostGSoC15/fixed_point/
this uses a specified (fixed at compile-time) fixed-point format. Big formats are possible and it works seamlessly with many Boost.Math functions, often using constexpr for compile-time computations).
It did not receive much attention when tentatively proposed for Boost.
Is this proposal stalled? I vaguely remember a fixed-point proposal but I don't remember the details.
I'm interested in a fast compile-time configured fixed-point type as I had to write one for one project.
It is working and in 'real-life' use 'in anger', and reasonably documented with examples of usage.
But nobody in Boost seemed to be at all interested, so it wasn't proposed formally.
Well, I'm interested. :) Is there a chance it will be proposed?
It seems there is at least some people interested so I will propose it finished for reviewing it :) 2018-01-01 13:41 GMT+01:00 Andrey Semashev via Boost <boost@lists.boost.org> :
On 01/01/18 15:27, Paul A. Bristow via Boost wrote:
-----Original Message-----
From: Boost [mailto:boost-bounces@lists.boost.org] On Behalf Of Andrey Semashev via Boost Sent: 31 December 2017 16:14 To: boost@lists.boost.org Cc: Andrey Semashev Subject: Re: [boost] Interest for a Fixed Point Arithmetics library
On December 31, 2017 6:50:01 PM "Paul A. Bristow via Boost" <boost@lists.boost.org> wrote:
An implementation of fixed-point has been written (and is in real-life use) by Chris Kormanyos
see https://github.com/BoostGSoC15/fixed_point/
this uses a specified (fixed at compile-time) fixed-point format. Big formats are possible and it works seamlessly with many Boost.Math functions, often using constexpr for compile-time computations).
It did not receive much attention when tentatively proposed for Boost.
Is this proposal stalled? I vaguely remember a fixed-point proposal but I don't remember the details.
I'm interested in a fast compile-time configured fixed-point type as I had to write one for one project.
It is working and in 'real-life' use 'in anger', and reasonably documented with examples of usage.
But nobody in Boost seemed to be at all interested, so it wasn't proposed formally.
Well, I'm interested. :) Is there a chance it will be proposed?
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman /listinfo.cgi/boost
On Mon, Jan 1, 2018 at 4:25 PM, Julien Vernay via Boost <boost@lists.boost.org> wrote:
2018-01-01 13:41 GMT+01:00 Andrey Semashev via Boost <boost@lists.boost.org> :
On 01/01/18 15:27, Paul A. Bristow via Boost wrote:
But nobody in Boost seemed to be at all interested, so it wasn't proposed formally.
Well, I'm interested. :) Is there a chance it will be proposed?
It seems there is at least some people interested so I will propose it finished for reviewing it :)
Thanks. I'll be looking forward to it.
> Is this proposal stalled? I vaguely remember a fixed-point> proposal but I don't remember the details.
In my opinion, the proposal was halted on purpose,or not forced to continue might be a better way ofputting it...
There were several reasons.
1) There was some, but not much, interest in the review.2) There is a slight conflict between the proposedboost::Fixed_point and the work of C++ standards groups,Mr. Macfarlane and SG14. In the Boost proposal,bound arithmetic is the default (unbound not supported).Whereas in the proposal for the standard, unboundarithmetic is the default, but both bound and unboundare supported.3) The work for Boost only supports 1 fixed-point type(similar to negatable). But the standards work supports4 types used for both signed and unsigned manipulationof both fractional and integral representations.
4) The work for Boost only supports 2 rounding modes.But the standards work supports 5.5) A few other details.
At the time, it was difficult for us to overcomethe slight discrepancies in these two bodies ofwork, in particular number 2 above.
The work for Boost is, nonetheless, available.It is tested and documented, but onlyfor what it does.
Best regards, Chris
On Sunday, December 31, 2017, 5:14:06 PM GMT+1, Andrey Semashev via Boost <boost@lists.boost.org> wrote:
On December 31, 2017 6:50:01 PM "Paul A. Bristow via Boost"
<boost@lists.boost.org> wrote:
>
>
>> -----Original Message-----
>> From: Boost [mailto:boost-bounces@lists.boost.org] On Behalf Of Julien
>> Vernay via Boost
>> Sent: 30 December 2017 18:30
>> To: boost@lists.boost.org
>> Cc: Julien Vernay
>> Subject: [boost] Interest for a Fixed Point Arithmetics library
>>
>> Hello everybody.
>> I am working on a project that needs and can use fixed point numbers, so I
>> started searching for a library implementing fixed point numbers, but I
>> didn't find anything. So I was wondering if people would need fixed point
>> arithmetics, and here I am.
>>
>> The aim of this FixedPointNumber is to fix the point at compile-time using
>> template, and having another template argument for the underlying type.
>> So we have :
>>
>> template<typename T, size_t N>
>> class FixedPointNumber {
>> T value;
>> public:
>> //interface
>> };
>>
>> The N last bits of value is the decimal part : so to convert a
>> FixedPointNumber to an int, we need only to use a bitshift operator (and
>> same for construction from int).
>> sizeof(FixedPointNumber<T,N>) == sizeof(T) , so there is no overhead in
>> this implementation.
>> Arithmetic and relational operators will be implemented.
>> Multiplication needs one int multiplication and one bit shift.
>> Division needs one int division and one bit shift.
>> There is only one value for zero.
>> Sign can be extracted with Most Significant Bit.
>>
>> I have already a basic implementation but if it interests more people I
>> will do a proper implementation.
>
> An implementation of fixed-point has been written (and is in real-life use)
> by Chris Kormanyos
>
> see https://github.com/BoostGSoC15/fixed_point/
>
> this uses a specified (fixed at compile-time) fixed-point format. Big
> formats are possible and it works seamlessly with many
> Boost.Math functions, often using constexpr for compile-time computations).
>
> It did not receive much attention when tentatively proposed for Boost.
Is this proposal stalled? I vaguely remember a fixed-point proposal but I
don't remember the details.
I'm interested in a fast compile-time configured fixed-point type as I had
to write one for one project.
_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
participants (4)
-
Andrey Semashev -
Christopher Kormanyos -
Julien Vernay -
Paul A. Bristow