Re: [go-nuts] Q16.16 fixed point numbers in go
ср, 24 окт. 2018 г. в 19:52, Michael Jones : > > https://en.wikipedia.org/wiki/Q_(number_format) says it all. > > you can use fixed point all the way...just shift after multiplies and before > divides and use double-width (int32) or wider math. > Thanks! -- Vasiliy Tolstov, e-mail: v.tols...@selfip.ru -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [go-nuts] Q16.16 fixed point numbers in go
Guilty. ;-) It's also more complicated because of the desire for saturating arithmetic. On Thu, Oct 25, 2018 at 9:57 AM Scott Cotton wrote: > > > On Wednesday, 24 October 2018 18:52:27 UTC+2, Michael Jones wrote: >> >> https://en.wikipedia.org/wiki/Q_(number_format) says it all. >> >> you can use fixed point all the way...just shift after multiplies and >> before divides and use double-width (int32) or wider math. >> >> > Yes, that's close enough for many purposes. But it gets more complicated > with carries/rounding and other functions (such as trig). > > Scott > > > >> On Mon, Oct 22, 2018 at 3:24 PM Vasiliy Tolstov >> wrote: >> >>> Thanks for suggestions. >>> If somebody can look at github.com/unistack-org/go-crush package >>> (ported ceph crush to go) i'l be happy. >>> Now weight represented as float32, i'm try to use >>> https://github.com/shopspring/decimal but code looks ugly with many >>> conversations. And also as i see ceph code (mapper.c for crush) they >>> does not use something >>> different for weights but uint32. may be i miss something but i don't >>> understand how they deal with weights like 0.580 in such cases. >>> So if somebody wants to help - i'l be happy. >>> пн, 22 окт. 2018 г. в 17:14, 'Ingo Oeser' via golang-nuts >>> : >>> > >>> > https://godoc.org/golang.org/x/image/math/fixed might provide some >>> hints how to work with this kind of type. >>> > >>> > The idea is to use a special type and provide all operations you need. >>> > >>> > https://godoc.org/math/big#Rat might be helpful to create first tests >>> of the basic math operations with maximum precision, before reducing >>> precision with a conversion to a more limited data type. >>> > >>> > -- >>> > You received this message because you are subscribed to the Google >>> Groups "golang-nuts" group. >>> > To unsubscribe from this group and stop receiving emails from it, send >>> an email to golang-nuts...@googlegroups.com. >>> > For more options, visit https://groups.google.com/d/optout. >>> >>> >>> >>> -- >>> Vasiliy Tolstov, >>> e-mail: v.to...@selfip.ru >>> >>> -- >>> You received this message because you are subscribed to the Google >>> Groups "golang-nuts" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to golang-nuts...@googlegroups.com. >>> For more options, visit https://groups.google.com/d/optout. >>> >> >> >> -- >> >> *Michael T. jonesmichae...@gmail.com* >> > -- > You received this message because you are subscribed to the Google Groups > "golang-nuts" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to golang-nuts+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/d/optout. > -- *Michael T. jonesmichael.jo...@gmail.com * -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [go-nuts] Q16.16 fixed point numbers in go
On Wednesday, 24 October 2018 18:52:27 UTC+2, Michael Jones wrote: > > https://en.wikipedia.org/wiki/Q_(number_format) says it all. > > you can use fixed point all the way...just shift after multiplies and > before divides and use double-width (int32) or wider math. > > Yes, that's close enough for many purposes. But it gets more complicated with carries/rounding and other functions (such as trig). Scott > On Mon, Oct 22, 2018 at 3:24 PM Vasiliy Tolstov > wrote: > >> Thanks for suggestions. >> If somebody can look at github.com/unistack-org/go-crush package >> (ported ceph crush to go) i'l be happy. >> Now weight represented as float32, i'm try to use >> https://github.com/shopspring/decimal but code looks ugly with many >> conversations. And also as i see ceph code (mapper.c for crush) they >> does not use something >> different for weights but uint32. may be i miss something but i don't >> understand how they deal with weights like 0.580 in such cases. >> So if somebody wants to help - i'l be happy. >> пн, 22 окт. 2018 г. в 17:14, 'Ingo Oeser' via golang-nuts >> >: >> > >> > https://godoc.org/golang.org/x/image/math/fixed might provide some >> hints how to work with this kind of type. >> > >> > The idea is to use a special type and provide all operations you need. >> > >> > https://godoc.org/math/big#Rat might be helpful to create first tests >> of the basic math operations with maximum precision, before reducing >> precision with a conversion to a more limited data type. >> > >> > -- >> > You received this message because you are subscribed to the Google >> Groups "golang-nuts" group. >> > To unsubscribe from this group and stop receiving emails from it, send >> an email to golang-nuts...@googlegroups.com . >> > For more options, visit https://groups.google.com/d/optout. >> >> >> >> -- >> Vasiliy Tolstov, >> e-mail: v.to...@selfip.ru >> >> -- >> You received this message because you are subscribed to the Google Groups >> "golang-nuts" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to golang-nuts...@googlegroups.com . >> For more options, visit https://groups.google.com/d/optout. >> > > > -- > > *Michael T. jonesmichae...@gmail.com * > -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [go-nuts] Q16.16 fixed point numbers in go
https://en.wikipedia.org/wiki/Q_(number_format) says it all. you can use fixed point all the way...just shift after multiplies and before divides and use double-width (int32) or wider math. On Mon, Oct 22, 2018 at 3:24 PM Vasiliy Tolstov wrote: > Thanks for suggestions. > If somebody can look at github.com/unistack-org/go-crush package > (ported ceph crush to go) i'l be happy. > Now weight represented as float32, i'm try to use > https://github.com/shopspring/decimal but code looks ugly with many > conversations. And also as i see ceph code (mapper.c for crush) they > does not use something > different for weights but uint32. may be i miss something but i don't > understand how they deal with weights like 0.580 in such cases. > So if somebody wants to help - i'l be happy. > пн, 22 окт. 2018 г. в 17:14, 'Ingo Oeser' via golang-nuts > : > > > > https://godoc.org/golang.org/x/image/math/fixed might provide some > hints how to work with this kind of type. > > > > The idea is to use a special type and provide all operations you need. > > > > https://godoc.org/math/big#Rat might be helpful to create first tests > of the basic math operations with maximum precision, before reducing > precision with a conversion to a more limited data type. > > > > -- > > You received this message because you are subscribed to the Google > Groups "golang-nuts" group. > > To unsubscribe from this group and stop receiving emails from it, send > an email to golang-nuts+unsubscr...@googlegroups.com. > > For more options, visit https://groups.google.com/d/optout. > > > > -- > Vasiliy Tolstov, > e-mail: v.tols...@selfip.ru > > -- > You received this message because you are subscribed to the Google Groups > "golang-nuts" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to golang-nuts+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/d/optout. > -- *Michael T. jonesmichael.jo...@gmail.com * -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [go-nuts] Q16.16 fixed point numbers in go
Thanks for suggestions. If somebody can look at github.com/unistack-org/go-crush package (ported ceph crush to go) i'l be happy. Now weight represented as float32, i'm try to use https://github.com/shopspring/decimal but code looks ugly with many conversations. And also as i see ceph code (mapper.c for crush) they does not use something different for weights but uint32. may be i miss something but i don't understand how they deal with weights like 0.580 in such cases. So if somebody wants to help - i'l be happy. пн, 22 окт. 2018 г. в 17:14, 'Ingo Oeser' via golang-nuts : > > https://godoc.org/golang.org/x/image/math/fixed might provide some hints how > to work with this kind of type. > > The idea is to use a special type and provide all operations you need. > > https://godoc.org/math/big#Rat might be helpful to create first tests of the > basic math operations with maximum precision, before reducing precision with > a conversion to a more limited data type. > > -- > You received this message because you are subscribed to the Google Groups > "golang-nuts" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to golang-nuts+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/d/optout. -- Vasiliy Tolstov, e-mail: v.tols...@selfip.ru -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
[go-nuts] Q16.16 fixed point numbers in go
https://godoc.org/golang.org/x/image/math/fixed might provide some hints how to work with this kind of type. The idea is to use a special type and provide all operations you need. https://godoc.org/math/big#Rat might be helpful to create first tests of the basic math operations with maximum precision, before reducing precision with a conversion to a more limited data type. -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [go-nuts] Q16.16 fixed point numbers in go
Does it possible to avoid conversation and works only with fixed point (if this can bring performance speedup)? пн, 22 окт. 2018 г. в 16:13, Vasiliy Tolstov : > > пн, 22 окт. 2018 г. в 16:02, Jan Mercl <0xj...@gmail.com>: > > > > On Mon, Oct 22, 2018 at 2:35 PM Vasiliy Tolstov wrote: > > > > > Does reading this 4 bytes to uint32 and use math.Float32frombits looks > > > right? > > > > It does not look right, see > > https://en.wikipedia.org/wiki/Q_(number_format)#Q_to_float > > > Thanks, so i need to read uint16 Num and uint16 Exp? > After that use number Num*2−exp ? > > -- > Vasiliy Tolstov, > e-mail: v.tols...@selfip.ru -- Vasiliy Tolstov, e-mail: v.tols...@selfip.ru -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [go-nuts] Q16.16 fixed point numbers in go
пн, 22 окт. 2018 г. в 16:02, Jan Mercl <0xj...@gmail.com>: > > On Mon, Oct 22, 2018 at 2:35 PM Vasiliy Tolstov wrote: > > > Does reading this 4 bytes to uint32 and use math.Float32frombits looks > > right? > > It does not look right, see > https://en.wikipedia.org/wiki/Q_(number_format)#Q_to_float Thanks, so i need to read uint16 Num and uint16 Exp? After that use number Num*2−exp ? -- Vasiliy Tolstov, e-mail: v.tols...@selfip.ru -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [go-nuts] Q16.16 fixed point numbers in go
On Mon, Oct 22, 2018 at 2:35 PM Vasiliy Tolstov wrote: > Does reading this 4 bytes to uint32 and use math.Float32frombits looks right? It does not look right, see https://en.wikipedia.org/wiki/Q_(number_format)#Q_to_float -- -j -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
[go-nuts] Q16.16 fixed point numbers in go
Hi! I have some binary format that contains in 4 bytes Q16.16 fixed point number in LittleEndian. Does reading this 4 bytes to uint32 and use math.Float32frombits looks right? And also as i understand if in my code i use float32 i have performance decrease vs if i work with fixed point int? -- Vasiliy Tolstov, e-mail: v.tols...@selfip.ru -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
