gideony2

Classic case of slowdown due to global rounding mode is
int=float;
the rules say you chop, and on the 80×87 this forces rounding mode to change twice, which costs a few
dozen cycles; see http://stereopsis.com/FPU.html.

There is a special optimization https://docs.microsoft.com/en-us/cpp/build/reference/qifist-suppress-ftol?view=vs-2017 to avoid this double change and allow rounding.

• This reply was modified 5 years, 11 months ago by  gideony2.

It may make sense at very low precision (way below IEEE). See https://arxiv.org/pdf/1502.02551v1.pdf

Russian web-page (accessed just now; pointer in Wikipedia), last line:
Ожидаемый год завершения работы: 2015

which translates as
Expected Completion Year 2015

Is C200x the proposed Unix time standard?

The A5/1 cipher should be as cheap as 3 small LFSRs

(1) decent crypto is just plain cheap; and: does an LFSR _have_ a middle? (if all bits were shifted 17 bits around, it would be just as good an LFSR)

I don’t know how to think about pseudo-random (it is supposed to look random, unless you play “unfairly”); I do know how to think about crypto: if the key is secret, distinguishing the output from random is _VERY_ expensive.

(2) without randomized rounding, 1+0.1+0.1+0.1+0.1 …. (all the way across 100 pages; rounded to integer) will still be 1. Randomized, it will be 0.1*(# of adds), +- epsilon.

If you go for stochastic rounding, make sure the random numbers are decent (some are not — see https://en.wikipedia.org/wiki/RANDU ).

Simplest fix: stream cipher with exportable key-length (or: encrypt 0,1,2… under a block-cipher — also exportable, if jail-time is an issue)

In an embedded system, with code in ROM, how can the prediction table get back into the load module?

Microsoft calls /QIfist an optimization, so I call it one.

Kahan: “Optimization is replacing something that works by something that almost works &is cheaper”

There is enough horsepower in /QIfist that MS implements &publishes this “optimization”.