discussion
discussion copied to clipboard
Are there any Posit implementations out there?
From what I have read, Posits has many advantages over Floats.
Does anyone know about any Forth implementations of it?
I don't know about posit but I am almost sure that integers are better than floating point
I only found one case where I couldn't replace them but I think I couldn't find the proper algorithm for this, this is the correct perspective textured triangle drawing.
The easy way to translate the programs is using the fixed point
The approximation of numbers to extend the range of representation does not seem to me a good idea, especially since there will always be a loss of precision, it is preferable to manage these limits of representation
I am not suggesting to replace the integer operations with Posit, but for those cases where you need non-integer numbers, Posit numbers are better in many ways over floating point numbers.
Sorry, I don't think I explain myself well. I think it is not necessary to use the floating point numbers in any case.
There are posit libraries like this one: https://gitlab.com/cerlane/SoftPosit . It would be possible to use, say CForth and replace the floating point operations in floatops.c with calls to the corresponding Posit operations.
If you haven't seen Gustafson's talk on Posits, then it is well worth the hour or so ... on YouTube. I look forwared to hardware manufacturers moving from IEEE-754 to Posits, or at least implementing both in hardware.
On Fri, 3 Apr 2020, Mitch Bradley wrote:
Date: Fri, 03 Apr 2020 11:31:50 -0700 From: Mitch Bradley [email protected] Reply-To: ForthHub/discussion [email protected] To: ForthHub/discussion [email protected] Cc: Subscribed [email protected] Subject: Re: [ForthHub/discussion] Are there any Posit implementations out there? (#86)
There are posit libraries like this one: https://gitlab.com/cerlane/SoftPosit . It would be possible to use, say CForth and replace the floating point operations in floatops.c with calls to the corresponding Posit operations.
? You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.[AAS2KERZ674RFU6C3CQNRVDRKYTRNA5CNFSM4L4BJPM2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOERDGWRI.gif]
-- -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Robert S. Sciuk [email protected] Principal Consultant 905.706.1354 Control-Q Research 97 Village Rd. Wellesley, ON N0B 2T0
Thanks folks!