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

[lkcl]

On Sunday, September 15, 2019 at 10:15:44 PM UTC+8, MitchAlsup wrote:

> > i am reminded of a message by hendrik [allo similarly old person, told ya your input was valuable :) ] on iibre-riscv-dev a couple months back about how Ahmdahl got similar customer complaints, and had to have an OS patch providing *EXACTLY* bit-for-bit the brain-dead broken FP accuracy of the mainframe that Ahmdahl were copying.
> 
> 
> 
> 
> 
> 
> > is this what you're referring to, Mitch?

> This is the basic problem; Yes.

Ok great. In the fpacc proposal I put that it is NOT to be used for compatibility with forced de-facto proprietary vendor standards, except as a last resort, and only then in a bit-for-bit compatible way, just as Allen raised concerns about.

*However*, if there is an *existing* public and prominent Industry standard, such as one publishee by the Khronos Group, where they *specifically* designed it well in advance with a "notification" to applications writers, "under no circumstances write applications thst assume better accuracy from any one hardware vendor", then and ONLY then is it ok.

The Khronos Group makes this really clear, and I found only just yesterday a set of slides by an AMD developer which emphasises that games/app writers *have* been warned on this, and asks them to actually follow the Khronos Group Developer advice properly.

That developers are not doing so is *not* "fixable" in the 3D market by making a particular bit for bit requirement in RISCV a hard one.

All that would achive was... yet another vendor with yet another answer which was not bit compatible with any of the other 100 GPUs *already out there and nit bit compatible either*.

Bottom line is: this is just how it is in 3D, there is nothing to "fix" that RISCV Compliance will help with by trying to set standards that are not *exactly* those of the Khronos Group, or, more to the point, trying to do so would step on the Khronos Group Trademark for which there is already precedent and priority.

Quite fascinating, in a horrifyingly messy sort of way.

L.