Search Issue Tracker

Won't Fix

Votes

0

Found in [Package]

1.8.4

Issue ID

BUR-2399

Regression

No

Floating Point Precision degrades when Burst Compilation is enabled

Package: Burst

-

How to reproduce:
1. Open the attached Project "burstTest" and load Scene "SampleScene"
2. In the Hierarchy, select the "Bug Test" GameObject
3. In the Inspector, change the value of "F" under "Bug Testing MB" to between 0.9999 and 0.999999
4. Observe the Console

Expected result: Values in the Console are less than or equal to 1
Actual result: Values are greater than 1

Reproducible with: 1.6.6 (2021.3.26f1), 1.8.4 (2022.3.0f1, 2023.1.0b18, 2023.2.0a16)

Notes:
- With Burst compilation disabled, all F values <=1 return values that are <=1

  1. Resolution Note:

    This is actually mono being weird. The mono runtime will treat all floating point calculations as doubles and convert the result back to a float at the end.
    Burst does the "right thing" here and treats all the intermediate calculations as single precision operations, not double.

    Burst will generate the same results as the equivalent C code. And if you run that C# code with .NET Core instead of mono you'll also get the same results.

Add comment

Log in to post comment

All about bugs

View bugs we have successfully reproduced, and vote for the bugs you want to see fixed most urgently.