remove setrounding from functions #443
Conversation
lucaferranti
changed the title
lucaferranti
force-pushed
the
lferrant/1.6-on-windows
branch
from
03690da
to
4599197
Compare
|
@dpsanders @lbenet I think this could be merged at this point and investigate the |
lucaferranti
changed the title
There was a problem hiding this comment.
Thanks for looking into this. I have added some comments to clarify the way in which you implemented fma and the helper functions; references or details are welcome! There is maybe only a comment, on an if-block which may deserve some changes. As for mig, I agree there is no need to have any rounding.
| if signbit(lo) | ||
| lo = prevfloat(hi) | ||
| elseif !iszero(lo) | ||
| lo = hi | ||
| hi = nextfloat(hi) | ||
| else | ||
| lo = hi | ||
| end |
There was a problem hiding this comment.
I'm not sure if I follow the algorithm here, so references or details are appreciated, nor if lo and hi may be any arbitrary values.
Yet, I think it is worth to check first if lo is zero (instead of getting its sign) and then deal with the sign part. If you have lo=-0.0 (zero is signed in floating point!) this block yields lo = prevfloat(hi) (first condition is satisfied), whereas if you have lo=+0.0 it yields hi (the else is used). My point is that you get two different answers for the same "value" of lo, and I do know know which one is the correct one.
| """ | ||
| two_fma(a, b, c) | ||
|
|
||
| Computes `val = fl(fma(a, b, c))` and `err = fl(err(fma(a, b, c)))`. |
There was a problem hiding this comment.
Can you provide a reference or more details, to understand why you do what you do?
|
|
||
| Interval(lo, hi) | ||
| return Interval(lo, hi) |
There was a problem hiding this comment.
I see that, in the way you define lo and hi, the returned interval includes the true answer. However, I can't see if this is the tightest interval or not, and may be it isn't. Can you comment on this?
|
This PR may have improvements from what was implemented in PR #593. @lucaferranti what do you think? Otherwise let us close the PR. |
The functions using
setrounding¨do not seem to work properly with 1.6 on windowsnow work, but it was discussed to remove setrounding from the implementation. For what I can say, the functions using setrounding areRelated issues: #442