Daily Commit Log

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Daily Commit Log

commits-2
Changes to Trunk (http://source.squeak.org/trunk.html) in the last 24 hours:

http://lists.squeakfoundation.org/pipermail/packages/2010-October/003866.html

Name: KernelTests-nice.167
Ancestors: KernelTests-nice.166

Assert that #degreeCos and #degreeSin are exact for any multiple of 90.

This is a nice property to have that radianCos and radianSin cannot afford due to approximate floating point representation of pi.

=============================================

http://lists.squeakfoundation.org/pipermail/packages/2010-October/003867.html

Name: Kernel-nice.510
Ancestors: Kernel-ul.509

Provide an implementation of #degreeCos and #degreeSin such that results are exact for any multiple of 90.

Care is also taken to evaluate the sine between -90° and 90°, this will avoid #degreesToRadians and i386 FPU sine fonction to accumulate round off errors due to approximate representation of pi.
We can thus evaluate 240 degreeCos with at most 1 ulp error. It's not perfect, but better than previous implementation.

For cosine, we know that:
        cosd(x)=cosd(abs(x))
        cosd(x)=sind(90-x)
thus the trick is to evaluate:
        cosd(x)=sind(90-abs(x)) after appropriate modulo in [-180,180[
This way, we are sure to evaluate the sine between -90° and 90°
The #degreesToRadians and #sin are used rather than #degreeSin to avoid cycles.

For sine, it would be necessary to evaluate either
sind(x) if abs(x) <=90
or sind(180-x) if abs(x) >= 90
A possible implementation would be:
        | x |
        x := 90 + self \\ 360 - 90.
        x >= 180 ifTrue: [x := 180 - x].
        ^x degreesToRadians sin
We prefer evaluating cosd(90-x) thus providing a branch free implementation.

=============================================

http://lists.squeakfoundation.org/pipermail/packages/2010-October/003868.html

Name: Collections-ul.398
Ancestors: Collections-ar.397

- restore the behavior of the old finalization scheme

=============================================

http://lists.squeakfoundation.org/pipermail/packages/2010-October/003869.html

Name: Collections-ul.399
Ancestors: Collections-ul.398

- added a postscript to apply the changes

=============================================