Bruno Daniel <bruno.daniel@...> writes:
>> Bruno Daniel <bruno.daniel@...> writes:
>> The inability to support the lisp constant double-float-epsilon, due
>> to not being in complete control of rounding -- if a float is stored
>> to the stack, it is stored as a 53-bit-mantissa float, not a 64-bit
> Thank you very much for the explanation. I understand that C and C++ don't
> care about such issues, as usual. I wonder whether Java does.
> In the code of set-fpu-word-for-c there is a condition
> (define-vop (set-fpu-word-for-c)
> (:node-var node)
> (:generator 0
> (when (policy node (= sb!c::float-accuracy 3))
> Is this policy of float-accuracy the default?
I believe so, yes.