You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@plc4x.apache.org by Christofer Dutz <ch...@c-ware.de> on 2019/12/09 12:59:03 UTC
Changing the format for floating point in mspec?
Hi all,
while implementing the mspec for KNXnet/IP I stumbled into the first pit regarding floating point numbers.
In the spec they claim to be normal IEEE 754 half precision floating point numbers and therefore I thought a “float 16” would be ok.
However it turned out that they don’t use the formal half precision types. The exponent is one bit less and the mantissa is one bit more.
So I think it would be great to change the float datatype to something like
float +4.11
If there’s a “+” it’s a signed floating point (Alternatively we could add a “ufloat” type) … the main thing is that then comes the number of exponent bits and after the dot the mantissa.
Right now I “solved” the problem like this:
['9' KnxDatapointF16
[reserved uint 8 '0x0']
[simple bit 'sign']
[simple int 4 'exponent']
[simple uint 11 'mantissa']
[virtual float 16 'val' '(sign ? -1 : 1) * (0.01 * mantissa) * (2 ^ exponent)']
]
Which is not that ideal …
What do you think?
The more I think of it, I think a: “float” and a “ufloat” is the better idea as this continues the int / uint concept for integers.
Chris
Re: Changing the format for floating point in mspec?
Posted by Otto Fowler <ot...@gmail.com>.
Hi everyone,
If you wanted to do MSPEC such that you could or could have generated the
currently supported protocols with it,
you may need to do more than this.
I have been mentioning anecdotally on slack about modbus float permutations
in the wild, and took a moment to find someplace
with that kind of support that I could show,
https://libmodbus.org/docs/v3.1.6/
check out the set or get float sections.
On December 11, 2019 at 05:27:11, Christofer Dutz (christofer.dutz@c-ware.de)
wrote:
Hi Julian,
I know ... I haven't encountered a "ufloat" yet, but thought ... if we do
it, we do it right ;)
Chris
Am 11.12.19, 10:57 schrieb "Julian Feinauer" <j....@pragmaticminds.de>:
Although I have to state that this ufloat thing is overall weird-ish.
But overall, I think the way you implemented it is the best.
Julian
Am 10.12.19, 21:55 schrieb "Christofer Dutz" <ch...@c-ware.de>:
Hi Lukas,
Yeah ... the more I thought about it, the more I liked it and since today
that version is implemented ... in the feature/knxnet-ip branch __
Chris
Am 10.12.19, 18:09 schrieb "Lukas Ott" <ot...@gmail.com>:
+1 for ufloat Option as Mspec already supports uint so in my humble opinion
it would just be consistent to support ufloat as well..
Christofer Dutz <ch...@c-ware.de> schrieb am Mo., 9. Dez. 2019,
13:59:
> Hi all,
>
> while implementing the mspec for KNXnet/IP I stumbled into the first pit
> regarding floating point numbers.
> In the spec they claim to be normal IEEE 754 half precision floating
point
> numbers and therefore I thought a “float 16” would be ok.
> However it turned out that they don’t use the formal half precision
types.
> The exponent is one bit less and the mantissa is one bit more.
> So I think it would be great to change the float datatype to something
like
>
> float +4.11
>
> If there’s a “+” it’s a signed floating point (Alternatively we could add
> a “ufloat” type) … the main thing is that then comes the number of
exponent
> bits and after the dot the mantissa.
>
> Right now I “solved” the problem like this:
>
> ['9' KnxDatapointF16
> [reserved uint 8 '0x0']
> [simple bit 'sign']
> [simple int 4 'exponent']
> [simple uint 11 'mantissa']
> [virtual float 16 'val' '(sign ? -1 : 1) * (0.01 * mantissa) * (2 ^
> exponent)']
> ]
>
> Which is not that ideal …
>
> What do you think?
>
> The more I think of it, I think a: “float” and a “ufloat” is the better
> idea as this continues the int / uint concept for integers.
>
> Chris
>
>
>
>
Re: Changing the format for floating point in mspec?
Posted by Christofer Dutz <ch...@c-ware.de>.
Hi Julian,
I know ... I haven't encountered a "ufloat" yet, but thought ... if we do it, we do it right ;)
Chris
Am 11.12.19, 10:57 schrieb "Julian Feinauer" <j....@pragmaticminds.de>:
Although I have to state that this ufloat thing is overall weird-ish.
But overall, I think the way you implemented it is the best.
Julian
Am 10.12.19, 21:55 schrieb "Christofer Dutz" <ch...@c-ware.de>:
Hi Lukas,
Yeah ... the more I thought about it, the more I liked it and since today that version is implemented ... in the feature/knxnet-ip branch __
Chris
Am 10.12.19, 18:09 schrieb "Lukas Ott" <ot...@gmail.com>:
+1 for ufloat Option as Mspec already supports uint so in my humble opinion
it would just be consistent to support ufloat as well..
Christofer Dutz <ch...@c-ware.de> schrieb am Mo., 9. Dez. 2019,
13:59:
> Hi all,
>
> while implementing the mspec for KNXnet/IP I stumbled into the first pit
> regarding floating point numbers.
> In the spec they claim to be normal IEEE 754 half precision floating point
> numbers and therefore I thought a “float 16” would be ok.
> However it turned out that they don’t use the formal half precision types.
> The exponent is one bit less and the mantissa is one bit more.
> So I think it would be great to change the float datatype to something like
>
> float +4.11
>
> If there’s a “+” it’s a signed floating point (Alternatively we could add
> a “ufloat” type) … the main thing is that then comes the number of exponent
> bits and after the dot the mantissa.
>
> Right now I “solved” the problem like this:
>
> ['9' KnxDatapointF16
> [reserved uint 8 '0x0']
> [simple bit 'sign']
> [simple int 4 'exponent']
> [simple uint 11 'mantissa']
> [virtual float 16 'val' '(sign ? -1 : 1) * (0.01 * mantissa) * (2 ^
> exponent)']
> ]
>
> Which is not that ideal …
>
> What do you think?
>
> The more I think of it, I think a: “float” and a “ufloat” is the better
> idea as this continues the int / uint concept for integers.
>
> Chris
>
>
>
>
Re: Changing the format for floating point in mspec?
Posted by Julian Feinauer <j....@pragmaticminds.de>.
Although I have to state that this ufloat thing is overall weird-ish.
But overall, I think the way you implemented it is the best.
Julian
Am 10.12.19, 21:55 schrieb "Christofer Dutz" <ch...@c-ware.de>:
Hi Lukas,
Yeah ... the more I thought about it, the more I liked it and since today that version is implemented ... in the feature/knxnet-ip branch __
Chris
Am 10.12.19, 18:09 schrieb "Lukas Ott" <ot...@gmail.com>:
+1 for ufloat Option as Mspec already supports uint so in my humble opinion
it would just be consistent to support ufloat as well..
Christofer Dutz <ch...@c-ware.de> schrieb am Mo., 9. Dez. 2019,
13:59:
> Hi all,
>
> while implementing the mspec for KNXnet/IP I stumbled into the first pit
> regarding floating point numbers.
> In the spec they claim to be normal IEEE 754 half precision floating point
> numbers and therefore I thought a “float 16” would be ok.
> However it turned out that they don’t use the formal half precision types.
> The exponent is one bit less and the mantissa is one bit more.
> So I think it would be great to change the float datatype to something like
>
> float +4.11
>
> If there’s a “+” it’s a signed floating point (Alternatively we could add
> a “ufloat” type) … the main thing is that then comes the number of exponent
> bits and after the dot the mantissa.
>
> Right now I “solved” the problem like this:
>
> ['9' KnxDatapointF16
> [reserved uint 8 '0x0']
> [simple bit 'sign']
> [simple int 4 'exponent']
> [simple uint 11 'mantissa']
> [virtual float 16 'val' '(sign ? -1 : 1) * (0.01 * mantissa) * (2 ^
> exponent)']
> ]
>
> Which is not that ideal …
>
> What do you think?
>
> The more I think of it, I think a: “float” and a “ufloat” is the better
> idea as this continues the int / uint concept for integers.
>
> Chris
>
>
>
>
Re: Changing the format for floating point in mspec?
Posted by Christofer Dutz <ch...@c-ware.de>.
Hi Lukas,
Yeah ... the more I thought about it, the more I liked it and since today that version is implemented ... in the feature/knxnet-ip branch __
Chris
Am 10.12.19, 18:09 schrieb "Lukas Ott" <ot...@gmail.com>:
+1 for ufloat Option as Mspec already supports uint so in my humble opinion
it would just be consistent to support ufloat as well..
Christofer Dutz <ch...@c-ware.de> schrieb am Mo., 9. Dez. 2019,
13:59:
> Hi all,
>
> while implementing the mspec for KNXnet/IP I stumbled into the first pit
> regarding floating point numbers.
> In the spec they claim to be normal IEEE 754 half precision floating point
> numbers and therefore I thought a “float 16” would be ok.
> However it turned out that they don’t use the formal half precision types.
> The exponent is one bit less and the mantissa is one bit more.
> So I think it would be great to change the float datatype to something like
>
> float +4.11
>
> If there’s a “+” it’s a signed floating point (Alternatively we could add
> a “ufloat” type) … the main thing is that then comes the number of exponent
> bits and after the dot the mantissa.
>
> Right now I “solved” the problem like this:
>
> ['9' KnxDatapointF16
> [reserved uint 8 '0x0']
> [simple bit 'sign']
> [simple int 4 'exponent']
> [simple uint 11 'mantissa']
> [virtual float 16 'val' '(sign ? -1 : 1) * (0.01 * mantissa) * (2 ^
> exponent)']
> ]
>
> Which is not that ideal …
>
> What do you think?
>
> The more I think of it, I think a: “float” and a “ufloat” is the better
> idea as this continues the int / uint concept for integers.
>
> Chris
>
>
>
>
Re: Changing the format for floating point in mspec?
Posted by Lukas Ott <ot...@gmail.com>.
+1 for ufloat Option as Mspec already supports uint so in my humble opinion
it would just be consistent to support ufloat as well..
Christofer Dutz <ch...@c-ware.de> schrieb am Mo., 9. Dez. 2019,
13:59:
> Hi all,
>
> while implementing the mspec for KNXnet/IP I stumbled into the first pit
> regarding floating point numbers.
> In the spec they claim to be normal IEEE 754 half precision floating point
> numbers and therefore I thought a “float 16” would be ok.
> However it turned out that they don’t use the formal half precision types.
> The exponent is one bit less and the mantissa is one bit more.
> So I think it would be great to change the float datatype to something like
>
> float +4.11
>
> If there’s a “+” it’s a signed floating point (Alternatively we could add
> a “ufloat” type) … the main thing is that then comes the number of exponent
> bits and after the dot the mantissa.
>
> Right now I “solved” the problem like this:
>
> ['9' KnxDatapointF16
> [reserved uint 8 '0x0']
> [simple bit 'sign']
> [simple int 4 'exponent']
> [simple uint 11 'mantissa']
> [virtual float 16 'val' '(sign ? -1 : 1) * (0.01 * mantissa) * (2 ^
> exponent)']
> ]
>
> Which is not that ideal …
>
> What do you think?
>
> The more I think of it, I think a: “float” and a “ufloat” is the better
> idea as this continues the int / uint concept for integers.
>
> Chris
>
>
>
>