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

[lkcl]

On Thursday, August 8, 2019 at 2:17:37 AM UTC+1, MitchAlsup wrote:
>
> An old guy at IBM (a Fellow) made a long and impassioned plea in a paper 
> from the late 1970s or early 1980s that whenever something is put "into the 
> instruction set" that the result be as accurate as possible. Look it up, 
> it's a good read.
>
> At the time I was working for a mini-computer company where a new 
> implementation was not giving binary accurate results compared to an older 
> generation. This was traced to an "enhancement" in the F32 and F64 accuracy 
> from the new implementation. To a customer, they all wanted binary 
> equivalence, even if the math was worse.
>

someone on the libre-riscv-dev list just hilariously pointed out that 
Ahmdahl-compatible IBM370 had FP that was more accurate than the 370: 
customers *complained* and they had to provide libraries that 
*de-accurified* the FP calculations :)

My gut feeling tell me that the numericalists are perfectly willing to 
> accept an error of 0.51 ULP RMS on transcendental calculations.
> My gut feeling tell me that the numericalists are not willing to accept an 
> error of 0.75 ULP RMS on transcendental calculations.
> I have no feeling at all on where to draw the line.
>

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.

l.