-
Notifications
You must be signed in to change notification settings - Fork 0
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Failure to capture full precision for low (maybe subnormal) values #145
Comments
I think we might be able to fix this by scaling up small numbers (via |
Thank you for investigating this. Indeed it looks like subnormal numbers. |
Over the top of my head question (I could be entirely silly here): is -292 within the supported exponent range? I think -308 + 106 / 3.3 == -273 (+-1) is the minimum supported exponent. Below this, everything falls into subnormal (for our case: zero) territory. |
You are reading my thoughts. I'm not sure about this one. I think we can hold numbers down to the currently set understanding of the minimum value, but we need to be careful when reading them in. I'm also not entirely sure here. Let's take the minimum value of So it makes sense that the minimum value is approximately of the order we have. I'm not exactly sure if this is correct. But I think it might be. Then we need to ensure that we have the right value set for Thoughts? Cc: @sinandredemption and @cosurgi |
You're absolutely right. |
Indeed. I had also yesterday changed the value of
Yes. We have a few edge cases that we still need to catch in the operations. We have already caught quite a few of them. But i fear we still need to find the remaining hiding ones, that we missed in the obvious first passes. In addition, I will also very soon PR some slight modifications to the I will get these new changes in ASAP. Thanks Fahad! (@sinandredemption) |
CI is cycling now in #146 |
This sounds like a job #136 could handle. I will work on this and report back to you. |
Indeed Fahad (@sinandredemption). I believe that is a good place to hammer the limits and edge cases. There will be a bit of tiny work forthcoming from my side on Other than that, I think tightening up our detection of edge cases will be helpful for the failing tests. The failing
The failures in BesselK are trivially insignificant, all related to I'll look into the above two files. i think the I wil linvestigate and isolate a root (and its brackets) that does not coverge. Keep going on tightening up the edge case in general Fahad and we will keep moving forward. kind regards, Chris Cc: @cosurgi |
Fixed by #146 |
When investigating #144, the problem is actually string read of a small number.
The reduced test case finds that
cpp_double_double
loses precision when reading in the following control value.Test code follows below.
The text was updated successfully, but these errors were encountered: