[libre-riscv-dev] [isa-dev] Re: FP transcendentals (trigonometry, root/exp/log) proposal

[lkcl]

On Thursday, August 8, 2019 at 6:30:11 AM UTC+1, Jacob Lifshay wrote:
>
> On Wed, Aug 7, 2019, 22:20 lkcl <luke.l... at gmail.com <javascript:>> wrote:
>
>> this tends to suggest that three platform specs are needed:
>>
>> * Embedded Platform (where it's entirely up to the implementor, as there 
>> will be no interaction with public APIs)
>> * UNIX Platform (which would require strict IEEE754 accuracy, for use in 
>> GNU libm, OR repeatable numericalist-acceptable accuracy)
>> * a *NEW* 3D Platform, where accuracy is defined by strict conformance to 
>> a high-profile standard e.g. OpenCL / Vulkan.
>>
> That wouldn't quite work on our GPU design, since it's supposed to be both 
> a GPU and a CPU that conforms to the UNIX Platform, it would need to meet 
> the requirements of the UNIX Platform and the 3D Platform, which would 
> still end up with correct rounding being needed.
>

yes, sorry: forgot to mention (so worth spelling out explicitly) - hybrid 
systems intended for multi-purpose obviously would need to meet the 
standard of the highest-accuracy purpose for which it is intended.

although doing FP64 as well, even that would need to meet the UNIX Platform 
spec standard.

adding these three options is to make it so that other implementors make 
the choice.  where interoperability matters (due to distribution of 
precompiled binaries that are targetted at multiple independent 
implementations), requirements have to be stricter.

l.