>>>>> "Martin" == Martin Raspaud <[EMAIL PROTECTED]> writes:
Martin> Martin Raspaud wrote:
>> Hi all
>> [...]
>>
>> (defmacro define-IEEE-float (name
>> &key (sign-bits 1)
>> (exponant-bits 8)
>> (mantissa-bits 23)
>> (bias 127)
>> (associated-type 'float))
>> (with-gensyms (significant exponant sign s-bits e-bits m-bits b
>> value)
>> `(defun ,(intern (string-upcase (format nil "read-~s" name))) ()
>> (let* ((,s-bits ,sign-bits)
>> (,e-bits ,exponant-bits)
>> (,m-bits ,mantissa-bits)
>> (,b ,bias)
>> (,value (read-unsigned-integer (/ (+ ,s-bits ,e-bits
>> ,m-bits) 8)))
>> (,significant (+ (ash 1 ,m-bits) (logand (1- (ash 1
>> ,m-bits)) ,value)))
>> (,exponant (- (logand (1- (ash 1 ,e-bits)) (ash ,value (-
>> ,m-bits))) ,b))
>> (,sign (ash ,value (- (+ ,e-bits ,m-bits)))))
>> (* ,significant (expt 2 ,exponant) (expt (coerce -1
>> ,associated-type) ,sign))))))
>> [...]
Martin> Ok, I found a solution : take que coerce out of the multiplication and
Martin> put it arround (* ,significant ....)
Martin> Apparently it's better to let cmucl handle the type itself until the end...
The compiler is trying to derive the result type of a call to,
essentially, (float x 1d0) when x is very large. The compiler should
just give up in this case, probably.
But at least you have a workaround for now.
Ray
