On 27 December 2010 23:04, Nicolas Cellier
<nicolas.cellier.aka.nice at gmail.com> wrote:
> 2010/12/27 Igor Stasenko <siguctua at gmail.com>:
>> On 27 December 2010 20:09, Nicolas Cellier
>> <nicolas.cellier.aka.nice at gmail.com> wrote:
>>> It seems that
>>>    1.0 exp = (Float classPool at: #E)
>>> is now false with the fdlibm version, at leat on Cog/MacOSX.
>>>
>>
>> tried on both John's 5.8b12
>> and own-built Cog VM , it returns true in both cases.
>> Maybe my image is obsolete and E is different there?
>>
>> (Float classPool at: #E) at: 1
>> 1074118410
>>
>> (Float classPool at: #E) at: 2
>>  2333366121
>>
>
> Hi Igor,
> your E seems correct.
> You can simply use (Float classPool at: #E) hex.
>
> Do you have the FloatMathPlugin plugged in your image ?
> In my trunk, Float>>exp is defined like this:
>
> exp
>        "Answer E raised to the receiver power.
>         Optional. See Object documentation whatIsAPrimitive."
>
>        <primitive: 'primitiveExp' module: 'FloatMathPlugin'>
>        self isNaN ifTrue:[SignalNaN ifTrue:[NaNError signal]. ^self].
>        "For now, fall back to the Squeak version of exp if FloatMathPlugin is absent"
>        ^self primitiveExp
>
> The old primitive works OK for exp(1.0) :
>    1.0 primitiveExp hex = (Float classPool at: #E) hex
>
i found that i using old primitives. not fdlibm ones.

indeed, now after using a floatmath plugin prim i got false from:

1.0 primitiveExp = (Float e)

>>> (Float classPool at: #E) is the closest Float approximation of e,
>>> which you can check with:
>>> (1.0 asArbitraryPrecisionFloatNumBits: 200) exp asFloat = (Float
>>> classPool at: #E)
>>>
>>> It's bad that fdlibm be able to compute (1.0e32 cos) with less than
>>> 1/2 ulp error, but (1.0 exp) with more than 1/2 ulp !
>>>
>>
>> Nicolas, how about writing own float math plugin?
>> You , among all of us seems care about float math most. So how about
>> getting your hands dirty with some C hacking?
>> :)
>>
>
> You mean rewriting fdlibm ? Less fun than writing in Smalltalk !
>

yeah.  but you can solve the problems :)

> Nicolas
>
>



-- 
Best regards,
Igor Stasenko AKA sig.