[OpenPOWER-HDL-Cores] [Libre-soc-dev] Power ISA v3.0B FP "STORE" function, potential specification discrepancy

Jacob Lifshay programmerjake at gmail.com
Sun May 16 17:24:17 UTC 2021


On Sat, May 15, 2021, 06:02 Luke Kenneth Casson Leighton <lkcl at lkcl.net>
wrote:

> No Denormalization Required (includes Zero / Infin-
> ity / NaN)
> if FRS1:11 > 896 or FRS1:63 = 0 then
>       WORD0:1 <- FRS0:1
>       WORD2:31 <- FRS5:34
>
> should that be "FRS[12:63] = 0", testing the mantissa for zero?
> or is it genuinely testing "both exponent and mantissa for zero"?
>

yes, it's both mantissa and exponent for zero, since that's +-0.0. if you
tested just exponent for zero you'd miss sending subnormal f64 numbers
through the rest of the algorithm.

Jacob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mailinglist.openpowerfoundation.org/pipermail/openpower-hdl-cores/attachments/20210516/0dbb708b/attachment.html>


More information about the OpenPOWER-HDL-Cores mailing list