[Cfp-interest] 18661-4 and 754

[Jim Thomas]

Many of the differences are explained by the following:

1. For C math functions we don’t specify IEEE underflow and overflow at the level of individual functions.

2. 754 doesn’t need to say things like cospi(0) returns 1 because it requires correct rounding.

So those differences are benign.

More comments are inlined below.


> On Feb 20, 2018, at 9:30 PM, Fred J. Tydeman <tydeman at tybor.com> wrote:
> 
> 18661-4 asinpi has no mention of underflow (but does mention range error).
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 atanpi has no mention of underflow (but does mention range error).
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 atan2pi has no mention of underflow (but does mention range error).
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 atan2pi has extra "for finite x" for atan2pi(+/-inf,+/-inf).

Yikes! Will remove two instances of “for finite x” as an editorial change.
> 
> 18661-4 atan2pi has no mention of underflow (but does mention range error).
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 cospi has cospi(+/-0) returns 1.
> 754 has no such mention.
> 
> 18661-4 sinpi has no mention of underflow or range error.
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 has tanpi 
> 754 has no mention of tanpi.

Known. Is it a problem?

> 
> 18661-4 exp10 has exp10(+/-0) returns 1.
> 754 has no such mention.
> 
> 18661-4 exp10 has no mention of underflow or overflow (but does mention range error).
> 754 in table 9.1 mentions underflow and overflow.
> 
> 18661-4 exp10m1 has no mention of underflow or range error for small x.
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 log2p1 has no mention of underflow or range error.
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 log10p1 has no mention of underflow (but does mention range error).
> 754 in table 9.1 mentions underflow.
> 
> 18661-4 compoundn has
>  compoundn(x, 0) returns 1 for x>= -1 or x a NaN. (allows signaling NaN)

In annex F NaN means quiet NaN unless stated otherwise.

> 754 has in 9.2.1
>  compoundn(x, 0) returns 1 for x>= -1, +inf or quiet NaN.
> 
> 18661-4 rootn has no mention of underflow or overflow (but does mention range error).
> 754 in table 9.1 mentions underflow and overflow.
> 
> 18661-4 rootn(+/-inf,n) is mentioned.
> 754 has no mention of that case.
> 
> 18661-4 pown(+/-inf,n) is mentioned.
> 754 has no mention of that case.

Does 754 need to add it?
> 
> 18661-4 has no mention of these cases
> 754 has in 9.2.1
>  powr(x,qNaN) is qNaN for x >= 0
>  powr(qNaN,y) is qNaN

Default quiet NaN behavior is covered globally.

- Jim Thomas

> 
> 
> ---
> Fred J. Tydeman        Tydeman Consulting
> tydeman at tybor.com      Testing, numerics, programming
> +1 (702) 608-6093      Vice-chair of PL22.11 (ANSI "C")
> Sample C99+FPCE tests: http://www.tybor.com
> Savers sleep well, investors eat well, spenders work forever.
> 
> _______________________________________________
> Cfp-interest mailing list
> Cfp-interest at oakapple.net
> http://mailman.oakapple.net/mailman/listinfo/cfp-interest